Customer-Oriented Strategic Planning for Hotel Competitiveness Improvement Based on Online Reviews

Abstract

The hotel industry has been facing fierce competition in recent years. It is important for hotels to conduct effective strategic planning for competitiveness improvement to achieve sustainable development. Prior studies on hotel strategic planning mainly use questionnaire data or hotel internal data, which have the problems of insufficient data or neglecting customer perspectives. The purpose of this study was to develop an integrated method for customer-oriented strategic planning for hotel competitiveness improvement based on text mining of online reviews. First, text mining of online reviews was conducted to extract customer-concerned service attributes and evaluate customer concern level and the performance of the service attributes through Latent Dirichlet Allocation (LDA) and sentiment analysis. Second, the competitive structures of the hotels were analyzed and the main competitors were identified from the competitive hotels through correspondence analysis. Third, SWOT analysis of the target hotel toward the main competitors was conducted, and the priorities of factors in each SWOT category were determined. An empirical study on a five-star hotel is given to illustrate the feasibility and effectiveness of the proposed method. The results indicate that the proposed method can help managers in strategic planning to obtain more specific strategies for hotel competitiveness improvement.

1. Introduction

With economic globalization and the rapid development of the tourism industry, the number of new entrants, hotel brands and hotel outlets in the hotel industry has continuously increased, leading to increasingly fierce competition among hotels [1,2]. In order to maintain competitive advantages, achieve sustainable development and gain favorable business revenue, hotel managers need to carry out effective strategic planning for competitiveness improvement [3]. Hotel competitiveness is the ability of a hotel to provide high-performance services so that the hotel can maintain its competitive advantages and gain a superior position against competitors [2]. Hotels are typical service-oriented businesses; thus, the customer-concerned service attributes and the customer’s evaluation of these attributes are vital factors that should be considered in strategic planning for hotel competitiveness improvement [4]. In pursuit of improving hotel competitiveness, a hotel needs to make efforts to provide high-performance services and gain higher customer satisfaction. Through hotel competitiveness improvement, a hotel can achieve sustainable development amid changing consumption and service patterns. The performance of the target hotel and its competitors can be evaluated through analyzing the customer-concerned service attributes and the customer’s evaluation of the attributes [5,6], thus laying a solid foundation for the development of effective competitiveness improvement strategies. Most of the existing methods for competitive strategy planning are based on questionnaires or the subjective judgment of managers [7,8], which often lack in-depth consideration of the customer’s perspective. Considering the service attributes that customers are concerned with and the performance of these service attributes, how to develop a customer-oriented hotel competitiveness improvement strategy to help hotels gain more effective competitive advantages is an important issue to be considered by hotel managers.

In recent years, with the rapid development of the internet economy and travel e-commerce, online travel agencies such as Ctrip.com (www.ctrip.com), Tripadvisor.com (www.tripadvisor.com) and Booking.com (www.booking.com) have provided consumers with more convenient and faster channels to choose and book hotels. Consumers tend to choose and book hotels through online travel platforms, and share their experiences, feelings and comments on hotel services in the form of online reviews after experiencing hotel services [9,10]. A large number of online hotel reviews have been accumulated on the online travel platforms which contain rich information including the customer-concerned service attributes and the customer’s evaluation of these attributes [11,12]. Through analyzing and mining the online review data of the target hotel and its competitors, it is possible to discover the service attributes that are of concern to customers and evaluate the performance of the target hotel and its competitors on each service attribute [13,14], thus providing effective support for analyzing the hotel’s competitive relationships and formulating a competitiveness improvement strategy for the target hotel. With the characteristics of large data volume, timely content update and easy access, online reviews can compensate for the shortcomings of traditional questionnaire data which may have little volume, lagging content and high acquisition costs [15]. Using online reviews as a data source to carry out customer-oriented strategic planning for competitiveness improvement can help hotels obtain more specific and effective strategies for competitiveness improvement. However, online reviews are unstructured data in the form of free text [16], and effective text mining methods should be adopted to obtain meaningful information from online reviews so as to support hotels in formulating competitiveness improvement strategies. Based on the above analysis, it is necessary to investigate how to carry out customer-oriented strategic planning for hotel competitiveness improvement based on online review data.

Currently, although some studies on theories and methods for hotel strategic planning can be found, studies on how to carry out strategic planning for hotel competitiveness improvement based on online reviews from the perspective of customers are still scarce. Therefore, it is necessary to develop an integrated method for customer-oriented strategic planning for hotel competitiveness improvement based on online reviews. This study aimed to answer the following three questions:

1. How can we extract meaningful information for strategic planning for hotel competitive improvement from online reviews?

2. How can we analyze the competitive structure and identify the main competitors of the target hotel?

3. How can we obtain effective strategic planning results for the target hotel considering the main competitors?

To address these research questions, we propose an integrated method for customer-oriented strategic planning for hotel competitiveness improvement based on online reviews. In the method, online reviews of the target hotel and the competitive hotels are first collected and preprocessed by web crawler software and the Jieba library in Python. Second, customer-concerned service attributes are extracted from online hotel reviews through the LDA method. Customer concern level and performance of the service attributes are evaluated through frequency statistics and multi-granularity sentiment analysis, respectively. Third, the competitive structure of the hotel is analyzed through correspondence analysis, and the main competitors of the target hotel are identified. Then, SWOT analysis toward the main competitors is conducted. In the SWOT analysis process, factors in each SWOT category are identified through the performance comparison of the target hotel and the main competitor hotels, and the priorities of the factors in each SWOT category are determined considering customer concern level, performance and performance difference of the factors. Furthermore, strategic planning results for competitive improvement of the target hotel are obtained based on the SWOT factors and the factor priorities. The proposed method can provide hotel managers with effective support and reference when conducting customer-oriented strategic planning for hotel competitiveness improvement based on online reviews. In addition, it can enrich related research on hotel strategic planning and application of online hotel reviews.

The reminder of this paper is organized as follows. In Section 2, literature on strategic planning for hotel competitiveness improvement, text mining of online hotel reviews and SWOT analysis for strategic planning are systematically reviewed. In Section 3, the method of customer-oriented strategic planning for hotel competitiveness improvement based on online reviews is presented. The results of the case study are deployed in Section 4. Some discussions about the results are given in Section 5. Finally, Section 6 summarizes the findings and highlights the theoretical and practical contributions of this study.

2. Literature Review

Three research streams are closely related to this study, i.e., strategic planning for hotel competitiveness improvement, text mining of online hotel reviews and SWOT analysis for strategic planning. A detailed literature review concerning these research streams is given in this section.

2.1. Strategic Planning for Hotel Competitiveness Improvement

Strategy can be considered as a set of decisions, plans and activities to reach a firm’s objectives [17,18]. Several studies have emphasized the competitive perspective in strategy management. Porter [19] argued that strategy should be evaluated according to the level of competition and strategic decisions should be made to provide a superior position against competitors. According to [20,21], it is meaningful to consider the competitors in strategy formulation so as to obtain competitive advantages. Koseoglu et al. [22] found in their empirical study that hotel managers prioritize market position analysis and assessment of competitors to formulate effective strategies. Their findings emphasize the consideration of competition structure analysis in hotel strategic planning. Furthermore, there are some studies which pointed out the necessity of considering customer perspectives in strategy management of various types of firms [23,24,25], especially in service sectors [26]. Thus, in order to maintain competitive advantages, a hotel should conduct effective strategy planning considering its competitive performance and resources [3,27,28,29].

Currently, some studies can be found focusing on strategic planning for hotel competitiveness improvement. Some scholars have studied the process, influence factors, activities or supporting tools for the practice of hotel strategic planning. For instance, Aldehayyat [30] examined the effects of organizational characteristics on strategic planning in Jordanian hotels. Senturk [31] studied the use of supporting tools for the strategic management of different types of hotels. Chen et al. [32] examined the differences in competitive strategy effects across two business cycle phases of the Taiwanese hotel industry. Okumus et al. [33] pointed out in their study that big data and artificial intelligence will impact and shape the strategy process. Some scholars have proposed effective methods to support competitive strategic planning for hotels. Their methods are concerned with different aspects of strategic planning for hotel competitiveness improvement. For example, Chen [34] proposed a model for evaluating service quality based on the competitive zone of tolerance by benchmarking against competitors. The proposed method can help managers in decision making to improve hotel performance considering competitors so as to achieve a competitive advantage. Matthew and Zheng [35] proposed a text analysis approach to create perceptual maps of hotel brands which could help hotels understand market structure and brand differentiation so as to obtain a positioning strategy. Xia et al. [2,36] presented methods based on data mining of hotel feature ratings for evaluating the competitiveness of a hotel or hotel brand. The proposed methods could help hotel managers evaluate hotel competitiveness. Hu and Trivedi [37] proposed a method for hotel brand positioning and competitive landscape mapping by text mining of online hotels reviews. Their method can support developing hotel positioning strategies to face competitors.

2.2. Text Mining of Online Hotel Reviews

As a typical and common form of user-generated content, online reviews are adopted as data sources in many studies focusing on different practical problems such as product ranking and selection [38,39], customer satisfaction analysis for products/services [40], product/service design and improvement [41], etc. In recent years, a large volume of online hotel reviews have accumulated on online travel platforms. Some studies can be found investigating the effect of online hotel reviews from the customer perspective and the hotel perspective [42,43]. Their findings have indicated that online reviews have a significant influence on customer behavior, and online reviews are important sources of information that can be used to develop effective strategies for hotels. Meanwhile, extracting useful information from online hotel reviews is a challenging task. Some studies have made meaningful attempts in text mining of online hotel reviews for different problems in the field of hotel management. Most existing studies involved two types of tasks in the application of online hotel reviews, i.e., extracting key service attributes (topics) and obtaining customer opinions toward the service attributes [13,14].

Online hotel reviews reflect the service attributes that the customers are most concerned about. Some studies extracted service attributes from online hotel reviews as evaluation dimensions for further analysis to support hotel management. For example, Mankad et al. [44] adopted the Latent Dirichlet Allocation method to extract topics from online reviews of 57 hotels in Moscow, Russia. They analyzed topic distribution in different types of online reviews to obtain managerial insights for hotel managers. Guo et al. [45] used the Latent Dirichlet Allocation method to identify the key service dimensions of customer satisfaction from a large dataset including online reviews of 25,670 hotels located in 16 countries. Their results uncovered 19 controllable dimensions for managing customer satisfaction with hotels. Hu et al. [46] adopted a structural topic model-based text analysis method to analyze online reviews of New York hotels. They identified the topics that appeared more frequently in negative reviews and analyzed the effect of hotel grade on the topics. Their findings enhanced the understanding of customer complaints and provided hotel managers with suggestions for customer satisfaction improvement. Moro et al. [47] used the Latent Dirichlet Allocation method to conduct topic modeling for online reviews of airport hotels. The obtained topic was used as a customer-concerned service in analyzing the service quality of airport hotels. Zhang et al. [48] used the Latent Dirichlet Allocation method to extract service attributes that are key factors affecting consumer satisfaction. They proposed a method based on penalty–reward contrast analysis to determine the prioritization of the extracted service attributes for hotel service improvement. Kim and Kim [49] derived fundamental selection attributes of customers from online hotel reviews through a method based on word frequency analysis and semantic network analysis. Their study investigated the association of hotel attributes with customer satisfaction and provided managerial implications for developing hotel strategies.

Online hotel reviews also contain abundant information about customer opinions toward service attributes. Some scholars have investigated the problem of extracting customer opinions from online hotel reviews, and different methods have been proposed for sentiment analysis of online hotel reviews so as to acquire customer opinions. For instance, Ali et al. [50] proposed a method based on SVM and Fuzzy Domain Ontology to classify the customer opinion toward hotel features into more detailed sentiment information. Bi et al. [51] proposed an IOVO-SVM algorithm to classify the sentiment strengths of online hotel reviews into five categories. The sentiment analysis results of online reviews were employed as customer opinions on attribute performance to conduct importance performance analysis of hotels. Sánchez-Franco et al. [52] constructed a sentiment classifier based on the TF-IDF approach and the Naive Bayes algorithm to classify customer satisfaction based on online hotel reviews. The resulting model can help hotel managers understand customer satisfaction and improve the hotel services. Cheng and Jin [53] utilized a hybrid sentiment analysis method based on lexicon and machine learning to identify a user’s attitude towards the attributes of home accommodation hotels in Airbnb. The study offered an alternative approach to understand the Airbnb user experience. Yadav and Roychoudhury [54] performed aspect-based sentiment analysis of online hotel reviews on TripAdvisor using SentiwordNet lexicon. Based on the results of sentiment analysis, opinions of reviewers associated with hotel aspects could be discovered, which could help understand the expectations of travelers in different trip modes. Tsai et al. [55] employed a powerful toolkit, OpinionFinder, to identify sentence polarity in online hotel reviews. The obtained sentiment polarity of sentences was used to generate review summaries to improve information processing of travelers. Hu et al. [56] conducted sentence-level sentiment analysis of online hotel reviews based on a lexicon-based approach to detect customers’ emotions regarding hotel attributes. Their method could help prioritize hotel attributes and optimize service offerings. Zhou and Liao [57] proposed a center term-based short sentence sentimental orientation algorithm to extract customer opinions from online hotel reviews collected from Ctrip. The results of sentiment analysis were used together with word frequency statistics to dynamically measure the customer satisfaction with hotels. Chen et al. [58] built a new sentiment lexicon for hospitality using the PolarityRank algorithm and performed sentiment analysis on a large dataset of online reviews of London hotels. The sentiment analysis result was used as attribute performance in the Kano-IPA model to explain customers’ rating behaviors and prioritize attributes for improvement. Lee et al. [59] conducted a lexicon-based sentiment analysis at the attribute level on online reviews of hotels in Seoul, Korea. Kano types of the attributes were identified based on fusing term frequency and sentiment analysis results with conjoint analysis to support prioritizing the attributes for design and marketing purposes.

2.3. SWOT Analysis for Strategic Planning

SWOT analysis is a widely used method in strategic planning [60]. In the process of strategic planning, a firm’s strengths, weaknesses, opportunities and threats can be identified through SWOT analysis [61]. A firm can make strategic decisions on how to direct resources to various factors in a competitive market environment based on the results of SWOT analysis [62,63]. To help managers analyze the firm’s situation and develop strategies more effectively, Weihrich [64] proposed a TOWS matrix for matching a firm’s opportunities and threats with its weaknesses and strengths. The TOWS matrix provides an effective framework for formulating strategies based on SWOT factors. Due to the simplicity of the execution process and the rich implication of the results, SWOT analysis and related extension methods have been utilized for strategic planning in various contexts [65,66,67].

In the traditional SWOT analysis approach, the results of SWOT factor identification mainly rely on the subjective judgment of the managers, with the problem of lacking customer consideration and quantitative factor prioritization [7]. To overcome the lack of factor prioritization, some studies proposed methods that integrate SWOT analysis with the multi-attribute decision-making method [63,68,69]. Although these studies provide methods for prioritizing SWOT factors, the prioritization is also obtained based on the subjective evaluation of experts. On the other hand, the data supporting the SWOT analysis process is mostly acquired through interviews and questionnaires or directly obtained from the firm’s internal record [7,67]. These data acquisition methods may encounter problems including insufficient data and a lack of customer perspectives. In recent years, some studies conducted customer-oriented SWOT analysis based on online review data. For example, Pai et al. [70] constructed an ontology-based SWOT analysis model to conduct customer-oriented SWOT analysis using electronic word-of-mouth information. The proposed ontology-based SWOT analysis mechanism can help a firm to carry out more effective strategic planning and increase the value of eWOM of their products. Srinivas and Rajendran [71] conducted customer-oriented SWOT analysis by analyzing online student reviews for university strategic planning. The proposed method integrated online review text mining and SWOT analysis to help university managers conduct effective strategic planning. Mitra and Jenamanis [72] conducted customer-oriented SWOT analysis for a camera brand based on an online brand image score, which was quantified from online reviews. Their method could support the decision-making process of market strategic planning for products. Cheng [73] proposed a change mining framework to conduct customer-oriented SWOT analysis based on product reviews. Their method could provide managerial implications for strategic planning of highly competitive and easily replaced products.

Based on the above literature review, it can be found that the existing studies provide some methods and references to this study, such as methods for text mining of online reviews, the SWOT analysis method for strategic planning, etc. However, it is necessary to point out that online reviews can be a promising data source for hotel strategic planning, and studies on how to carry out effective strategic planning for hotel competitiveness improvement based on online reviews are still scarce. Although the existing studies can provide support and reference to this study, they still have the following limitations. (1) Existing studies on hotel competitiveness or strategic planning usually consider the situation that the competitors are already known. However, due to the complex market environment in the hotel industry, the process of identifying main competitors should be integrated with the process of strategic planning to obtain a more targeted competitive strategy. (2) The applications of online review text mining in hotel strategic planning is still not deep enough. Online reviews are usually utilized in problems such as hotel selection, hotel customer satisfaction analysis, hotel service improvement, etc. Although several studies can be found that use online review data in hotel competitiveness analysis or strategic planning, the existing studies usually use online ratings or sentiment polarity of online text reviews, and more detailed information including multi-granularity sentiment strength obtained from online reviews is seldom utilized. (3) The existing studies on customer-oriented SWOT analysis seldom provide a specific method for determining the priorities of the SWOT factors, which may hinder obtaining detailed strategic planning results for hotel competitiveness improvement. Hence, it is necessary to conduct an in-depth study on how to carry out customer-oriented strategic planning for hotel competitiveness improvement based on text mining of online reviews.

3. Materials and Methods

According to the solution framework shown in Figure 1, a detailed description of the proposed method for customer-oriented strategic planning for hotel competitiveness improvement based on online reviews is given in this section. The method consists of four parts: (1) online review data collection and preprocessing; (2) text mining of online reviews for hotel strategic planning; (3) analyzing competitive structure and identifying main competitors; (4) conducting SWOT analysis toward main competitors. The description of the method is given below.

3.1. Online Review Data Collection and Preprocessing

In this study, we take strategic planning for competitiveness improvement of a five-star hotel as an example to conduct an empirical study. We consider the Mandarin Oriental Pudong Shanghai Hotel as the target hotel, which is located in the Lujiazui Area, Shanghai, China. Lujiazui is the core functional area of Shanghai International Financial Center, and there are quite a few five-star hotels in this area, which has led to fierce competition among hotels.

We selected 13 other hotels in the same area as the competitors which are candidate main competitors of the target hotel. The competitors are all five-star hotels in the Lujiazui Area and their overall online ratings are close to that of the target hotel. Online reviews of the target hotel and the 13 competitor hotels were collected from Ctrip.com (https://www.ctrip.com), which is one of the most famous online travel platforms in China. Bazhuayu 8.0 (https://www.bazhuayu.com/) web crawler software was employed to collect online reviews of the hotels. Data items including hotel name, hotel star level, hotel location, hotel overall online rating on Ctrip.com and online text reviews of each hotel were collected. By March 2022, we had gathered 38,901 Chinese online reviews of these hotels. The hotel name, overall online rating and the number of collected online reviews of each hotel are shown in

Table 1. Hotel names, overall online ratings and number of collected online reviews.

Hi	Hotel Name	Overall Online Rating	Number of Online Reviews
$H_0$	Mandarin Oriental Pudong Shanghai	4.7	2532
$H_1$	Oriental Riverside Hotel Shanghai	4.7	1965
$H_2$	Shanghai Grand Trustel Purple Mountain Hotel	4.7	4133
$H_3$	Grand Hyatt Shanghai	4.6	5460
$H_4$	Grand Kempinski Hotel Shanghai	4.6	5573
$H_5$	The Ritz-Carlton Pudong Shanghai,	4.7	1868
$H_6$	JW Marriott Marquis Hotel Pudong Shanghai	4.7	1097
$H_7$	Regent Pudong Shanghai	4.7	1488
$H_8$	Park Hyatt Shanghai	4.5	1421
$H_9$	JOYA Shanghai Lujiazui Hotel	4.8	2827
$H_{10}$	InterContinental Pudong Shanghai	4.6	1070
$H_{11}$	Shangri-La Pudong Shanghai	4.6	6198
$H_{12}$	Atour S Hotel financial center Shanghai	4.7	1379
$H_{13}$	Grand Soluxe Zhongyou Hotel Shanghai	4.8	1890

.

In the data preprocessing process, first, data cleaning was conducted. Default reviews provided by Ctrip and extremely short and meaningless reviews were filtered out. After data cleaning, we had 35,558 valid online text reviews of the target hotel and the competitive hotels. Let $H_0$ denote the target hotel, $H_i$ denote the ith competitive hotel and $i=1,2,\cdots ,I$, where I is the number of competitive hotels considered in this study. Second, Chinese word segmentation and part of speech (POS) tagging were conducted through the Jieba library in Python. Considering the grammatical and semantic structure of the text, each Chinese online review was segmented into words and each word was tagged with the corresponding part of speech such as nouns, adjectives, adverbs, etc. Third, stop words which appear frequently in the text but have no essential meanings were removed by comparing the results of word segmentation with the Chinese stop word list. In this study, words that frequently appeared in online hotel reviews but have no value in topic extraction and sentiment analysis, such as “酒店 (hotel)”, “滴 (“di”, a commonly used auxiliary word in Chinese online reviews)”, etc., were also added to the stop word list. Words on the stop word list were removed from the word set obtained from word segmentation.

3.2. Text Mining of Online Reviews for Hotel Strategic Planning

As typical service-oriented businesses, hotels mainly compete on customer-concerned service attributes. Customer concern level and the performance of the service attributes are important details needed in strategic planning for hotel competitiveness improvement. Hence, online review text mining methods are given in this section, including a method for extracting customer-concerned service attributes and a method for evaluating customer concern level and performance of the service attributes.

3.2.1. Extracting Customer-Concerned Service Attributes

LDA (Latent Dirichlet Allocation) is an unsupervised machine learning model for text topic modeling in the field of natural language processing. In LDA, it is assumed that each text is presented by a Dirichlet distribution of topics and each topic is presented by a Dirichlet distribution of words [74]. Through LDA, the potential structure and hidden semantic information can be discovered from a large amount of online reviews, and topic content discussed in online reviews can be mined effectively [45]. In recent studies, LDA has been widely used for extracting attributes of a product/service from online customer reviews [40,48,75]. In this study, we adopted the LDA method to extract customer-concerned service attributes from online hotel reviews.

The process of extracting customer-concerned service attributes based on LDA can be summarized in three steps. First, the review–word matrix is constructed based on the preprocessed online reviews of all the considered hotels. The review–word matrix is a structured representation of the online reviews. Each row of the review–word matrix represents an online review, and each element in the row represents the frequency with which a word appeared in the review. Second, train the LDA model using the review–word matrix as input and assess the performance of the LDA model when various topic numbers is used. For each topic number, perplexity and coherence score are used to assess the performance of the LDA model. Third, obtain the results of customer-concerned service attributes extraction. To obtain more interpretable and more substantial topic information, the topic number of LDA is selected considering the values of perplexity and coherence score and the opinions of the decision maker or the hotel managers comprehensively. The result of topic extraction through LDA is presented in the form of a topic–word matrix. Each row of the topic–word matrix represents a topic and each element in the row is a keyword related to the topic. Furthermore, we manually merge the topics with similar meanings and filter the noisy words to obtain the final result of topic extraction. In this study, we used the Gensim library in Python to execute the LDA method. Following [40,45,48], the topics obtained by LDA can be regarded as customer-concerned hotel service attributes. Let $F=\left\{f_1,f_{2,\cdots ,}{f}_J\right\}$ denote the set of service attributes (topics) obtained through LDA, where $f_j$ is the jth service attribute, $j=1,2,\cdots ,J$, and J is the number of extracted service attributes. Let $W_j^F=\left\{w_{j1},w_{j2},\cdots ,w_{j{P}_j}\right\}$ denote the keyword set of service attribute $f_j$, where $w_{jp}$ is the pth keyword related to attribute $f_j$ and $P_j$ is the number of keywords in $W_j^F$. Furthermore, we label each obtained service attribute according to the practical meaning of the keywords related to it.

3.2.2. Evaluating Customer Concern Level and Performance of the Service Attributes

In this study, the customer concern level of a service attribute is referred to as how much importance the customers attach to the attribute. Meanwhile, the performance of a service attribute is how the customer evaluates the hotel service for the attribute. The customer concern level and performance of the service attributes should be considered in customer-oriented strategic planning for hotel competitiveness improvement. The mentioned frequency of an attribute in online reviews can reflect customer concern level, and the customers’ sentiment in the online reviews can reflect their evaluation of the performance of the attribute [47,48]. In this study, a method based on frequency statistics and multi-granularity sentiment analysis was employed to evaluate the customer concern level and performance of the service attributes.

Generally, several service attributes may be mentioned in an online hotel review. To conduct more accurate and detailed analysis, it is necessary to identify online review information concerning each service attribute. Considering characteristics of the Chinese language and writing patterns in online reviews, an online review is divided into several types of information according to adjacent punctuation, including “, (Chinese comma)”, “。 (Chinese period)”, “? (Chinese question mark)” and “! (Chinese exclamation mark)”. That is, words between adjacent punctuation are exacted from an online review as candidate online review information concerning the service attributes [76]. Let $S_i=\left\{s_{i1},s_{i2},\cdots ,s_{i{E}_i}\right\}$ denote the set of online review information extracted from online reviews of hotel $H_i$, where $s_{ie}$ is the eth online review information. Based on the preprocessed online reviews, $s_{ie}$ can be represented by word set $S{W}_{ie}=\left\{S{W}_{ie}^1,S{W}_{ie}^2,\cdots ,S{W}_{ie}^{Z_i^e}\right\}$, where $S{W}_{ie}^z$ is the zth word in $S{W}_{ie}$, $z=1,2,\cdots ,Z_i^e$, and $Z_i^e$ is the number of words in $S{W}_{ie}$. Then, the set of online review information of hotel $H_i$ concerning service attribute $f_j$ can be obtained by comparing each $S{W}_{ie}$ with $W_j^F$, which is the keyword set of service attribute $f_j$ obtained in Section 3.2.1. If $S{W}_{ie}{{\displaystyle \cap }}^{ }{W}_i^F\ne \varnothing$, $s_{ie}$ is put into the set of online review information of hotel $H_i$ concerning service attribute $f_j$. Let $F{S}_{ij}=\left\{F{S}_{ij}^1,F{S}_{ij}^2,\cdots ,F{S}_{ij}^{G_i^j}\right\}$ denote the set of online review information of hotel $H_i$ concerning attribute $f_j$, where $F{S}_{ij}^g$ is the gth online review information in $F{S}_{ij}$ and $G_i^j$ is the number of online review information in $F{S}_{ij}$, $i=0,1,2,\cdots ,I$, $j=1,2,\cdots ,J$, $g=1,2,\cdots ,G_i^j$.

Based on the obtained online review information of each hotel concerning each service attribute, the customer concern level is evaluated by frequency statistics toward online review information concerning a service attribute. Let $Fre{q}_{ij}$ denote the frequency that attribute $f_j$ is mentioned in online reviews of hotel $H_i$. $Fre{q}_{ij}$ can be obtained by counting the number of online review information in $F{S}_{ij}$, i.e., $Fre{q}_{ij}=G_i^j$. Let $A{T}_{ij}$ denote the customer concern level of attribute $f_j$ with respect to hotel $H_i$. $A{T}_{ij}$can be calculated by:

$$A{T}_{ij}=\frac{Fre{q}_{ij}}{{{\displaystyle \sum }}_{j=1}^{J}Fre{q}_{ij}}, i=0,1,2,\cdots ,I; j=1,2,\cdots ,J$$

(1)

It can be seen from Equation (1) that the customer concern level can reflect how the customer emphasizes a certain service attribute. If attribute $f_j$ has higher $A{T}_{ij}$, it means that customers of hotel $H_i$ mention attribute $f_j$ more frequently in online reviews of $H_i$; that is, the customers of hotel $H_i$ attach more importance to attribute $f_j$.

In order to evaluate the performance of the service attributes, following Li et al. [77], the dictionary-based multi-granularity sentiment analysis method is employed to obtain a more accurate and detailed evaluation of the service attributes. The detailed process of the method is described as follows.

First, the adjectives, verbs and adverbs in online review information $F{S}_{ij}^g$ are extracted to construct an opinion word set of online review information [76,78]. Let $F{W}_{ij}^g=\left\{F{W}_{ij}^{g1},F{W}_{ij}^{g2},\cdots ,F{W}_{ij}^{g{L}_{ij}^g}\right\}$ denote the opinion word set corresponding to online review information $F{S}_{ij}^g$, where $F{W}_{ij}^{gl}$ denotes the lth word in $F{W}_{ij}^g$ and $L_{ij}^g$ denotes the number of words in $F{W}_{ij}^g$, $i=0,1,2,\cdots ,I ; j=1,2,\cdots ,J; g=1,2,\cdots ,G_i^j; l=1,2,\cdots ,L_{ij}^g$.

Second, as the sentiment word set concerning each service attribute may be different, a domain sentiment dictionary for each service attribute is constructed to improve the accuracy of sentiment analysis. Let $W_{ij}$ denote the opinion word set concerning attribute $f_j$ in online reviews of hotel $H_i$. $W_{ij}$ can be obtained by:

$$W_{ij}=F{W}_{ij}^1{{\displaystyle \cup }}^{ }F{W}_{ij}^2{{\displaystyle \cup }}^{ }\cdots {{\displaystyle \cup }}^{ }F{W}_{ij}^{G_i^j}, i=0,1,2,\cdots ,I, j=1,2,\cdots ,J$$

(2)

Based on $W_{ij}$, opinion word set ${\overline{W}}_j$ concerning attribute $f_j$ in online reviews of all the considered hotels can be represented by:

$${\overline{W}}_j=W_{0j}{{\displaystyle \cup }}^{ }{W}_{1j}{{\displaystyle \cup }}^{ }\cdots {{\displaystyle \cup }}^{ }{W}_{Ij}, j=1,2,\cdots ,J$$

(3)

After obtaining ${\overline{W}}_j$, the domain sentiment dictionary of service attribute $f_j$ can be constructed using HowNet (http://www.keenage.com) as the base sentiment dictionary. Let ${\overline{W}}_{HN}^{+}$ and ${\overline{W}}_{HN}^{-}$ denote the set of positive sentiment words and negative sentiment words in HowNet, respectively. The domain positive sentiment dictionary and the domain negative sentiment dictionary of attribute $f_j$ can be obtained by:

$${\overline{W}}_j^{+}={\overline{W}}_{HN}^{+}{{\displaystyle \cap }}^{ }{\overline{W}}_j, j=1,2,\cdots ,J$$

(4)

$${\overline{W}}_j^{-}={\overline{W}}_{HN}^{-}{{\displaystyle \cap }}^{ }{\overline{W}}_j, j=1,2,\cdots ,J$$

(5)

In the case that some words in ${\overline{W}}_j$ may belong to neither ${\overline{W}}_{HN}^{+}$ nor ${\overline{W}}_{HN}^{-}$, or if the sentiment polarity of some words cannot be identified by the above method, those words are manually added to ${\overline{W}}_j^{+}$ or ${\overline{W}}_j^{-}$ based on domain knowledge [38].

Then, five-granularity sentiment analysis is carried out to obtain a more detailed evaluation of the service attributes [77,78]. Let $v_{ij}^g$ denote the sentiment strength value of online review information $F{S}_{ij}^g$. In the five-granularity sentiment analysis, $v_{ij}^g\in \left\{-2,-1,0,1,2\right\}$, where −2(2) represents the high strength value of negative (positive) sentiment, −1(1) represents the plain strength value of negative (positive) sentiment and 0 represents the neutral sentiment. The degree adverbs in HowNet are divided into two degree adverb dictionaries according to the strength level of the words. Let $De{g}^1$ and $De{g}^2$ denote the plain strength degree adverb dictionary and the high strength degree adverb dictionary, respectively. For example, words that represent plain sentiment strength such as “轻微的(slightly)” and “有点儿(a little)” are in $De{g}^1$, and words that represent high sentiment strength such as “非常(very)” and “尤其(especially)” are in $De{g}^2$. Degree adverbs in $De{g}^2$ all have higher sentiment strength than those in $De{g}^1$. Moreover, a negative adverb dictionary $Neg$ can be obtained based on the set of negative adverbs of HowNet. Based on the opinion word set corresponding to online review information and the constructed dictionaries, the sentiment value $v_{ij}^g$ of online review information $F{S}_{ij}^g$ can be determined by:

$$v_{ij}^g=\left({\eta }_{ij}^{g+}+{\eta }_{ij}^{g-}\right)·{\eta }_{ij}^{gNeg}·{\eta }_{ij}^{gDeg}$$

(6)

In Equation (6), ${\eta }_{ij}^{g+}$,${\eta }_{ij}^{g-}$,${\eta }_{ij}^{gNeg}$ and ${\eta }_{ij}^{gDeg}$ are indicator variables indicating the positive sentiment orientation, negative sentiment orientation, negative adverb and strength of degree adverb, respectively. The values of ${\eta }_{ij}^{g+}$,${\eta }_{ij}^{g-}$,${\eta }_{ij}^{gNeg}$ and ${\eta }_{ij}^{gDeg}$ can be determined by comparing the word set $F{W}_{ij}^g$ with positive sentiment dictionary ${\overline{W}}_j^{+}$, negative sentiment dictionary ${\overline{W}}_j^{-}$, negative adverb dictionary $Neg$ and degree adverb dictionary $De{g}^2$:

$${\eta }_{ij}^{g+}=\{\begin{array}{cc}1,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }{\overline{W}}_j^{+}\ne \varnothing \\ 0,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }{\overline{W}}_j^{+}=\varnothing \end{array}$$

(7)

$${\eta }_{ij}^{g-}=\{\begin{array}{cc}-1,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }{\overline{W}}_j^{-}\ne \varnothing \\ 0,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }{\overline{W}}_j^{-}=\varnothing \end{array}$$

(8)

$${\eta }_{ij}^{gNeg}=\{\begin{array}{cc}-1,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }Neg\ne \varnothing \\ 1,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }Neg=\varnothing \end{array}$$

(9)

$${\eta }_{ij}^{gDeg}=\{\begin{array}{cc}2,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }De{g}^2\ne \varnothing \\ 1,& F{W}_{ij}^g{{\displaystyle \cap }}^{ }De{g}^2=\varnothing \end{array}$$

(10)

Based on the sentiment analysis of online review information, the performance of hotel $H_i$ in service attribute $f_j$ can be evaluated by the average sentiment strength value of the online review information of hotel $H_i$ concerning attribute $f_j$. Let $P{M}_{ij}$ denote the performance value of hotel $H_i$ on service attribute $f_j$. $P{M}_{ij}$ can be determined by:

$$P{M}_{ij}=\frac{{{\displaystyle \sum }}_{g=1}^{G_i^j}{v}_{ij}^g}{G_i^j}, i=0,1,2,\cdots ,I, j=1,2,\cdots ,J$$

(11)

3.3. Analyzing Competitive Structure and Identifying Main Competitors

In Section 3.1, the target hotel’s competitors are preliminarily selected according to the location, star level, online rating, etc. These competitive hotels are similar to the target hotel in some aspects, but they may have different competitive situations with respect to the target hotel. To conduct more targeted and detailed strategic planning, the competitive structure of the hotels should be analyzed, and the main competitors that have closer competitive positions to the target hotel should be identified. Correspondence analysis (CA) is a statistical technique which is commonly used in market research to represent categorical marketing research data with a low-dimensional map [79]. Based on CA, the relationships between firms (brands) and attributes can be explored and thus provide support for analyzing the competitive structure and identify competitors with close positions in the market structure [35,80]. In this study, we used CA to analyze hotel competitive structure. Based on visualization of the results of CA by perceptual mapping, a method for identifying the main competitors is proposed.

Some studies can be found that use the mentioned frequency of attributes in online reviews to reflect customer perception of different attributes of the firm and conduct CA to analyze market position [35,80]. Some studies argued that sentiment strength of online reviews is important information that reflects customer perception toward the product/service [50,51]. Hence, based on the above studies, we propose a modified correspondence analysis method for competitive structure analysis and main competitor identification considering the sentiment strength of the online review information. In our method, the perception frequencies of the service attributes are calculated according to the sentiment strength of the online review information. For online review information $F{S}_{ij}^g$, the perception frequency is determined according to the sentiment strength $v_{ij}^g$. When $v_{ij}^g=$−1, 1 or 0, it means the perception of the customer is plain, i.e., the customer expressed plain sentiment through online review information $F{S}_{ij}^g$. On the other hand, when $v_{ij}^g=$−2 or 2, it means the perception of the customer is more intense, i.e., the customer expressed intense sentiment through online review information $F{S}_{ij}^g$. Let ${\gamma }_{ij}^g$ denote the perception frequency of online review information $F{S}_{ij}^g$. To consider the differences in customer perception that are reflected in online review information with different sentiment strength values, ${\gamma }_{ij}^g$ is determined by:

$${\gamma }_{ij}^g=\{\begin{array}{cc}1,& v_{ij}^g=-1,0,1\\ 2,& v_{ij}^g=-2,2\end{array}$$

(12)

Furthermore, let $Perfre{q}_{ij}$ denote the total perception frequency of service attribute $f_j$ in the online reviews of hotel $H_i$. $Perfre{q}_{ij}$ can be obtained by summarizing perception frequency with respect to all the online review information concerning $f_j$ of hotel $H_i$:

$$Perfre{q}_{ij}={{\displaystyle \sum }}_{g=1}^{G_i^j}{\gamma }_{ij}^g, i=0,1,2,\cdots ,I, j=1,2,\cdots ,J$$

(13)

According to $Perfre{q}_{ij}$, the contingency table of perception frequencies of the service attributes and the considered hotels can be constructed. As shown in

Table 2. Contingency table of perception frequencies of the service attributes concerning each hotel.

Hi	f1	f2	$\cdots$	fJ
$H_0$	$Perfre{q}_{01}$	$Perfre{q}_{02}$	…	$Perfre{q}_{0J}$
$H_1$	$Perfre{q}_{11}$	$Perfre{q}_{12}$	…	$Perfre{q}_{1J}$
$⋮$	$⋮$	$⋮$	$⋮$	$⋮$
$H_I$	$Perfre{q}_{I1}$	$Perfre{q}_{I2}$	…	$Perfre{q}_{IJ}$

, the contingency table summarizes the perception frequencies of the service attributes concerning each hotel. The data form in the contingency table is the input of correspondence analysis. In this study, we use SPSS software tool to conduct correspondence analysis and obtain the visualization result in the form of a perceptual map. In the perceptual map, the competitive structures of the hotels are represented in a two-dimensional graph. Hotels that are positioned close together in the perceptual map may have similar customer perceptions and thus are competing intensely [35,80]. Following [81], we used cosine similarity to measure the closeness of competitive positions of two hotels in the perceptual map. Let $u_0^1$ and $u_0^2$ denote the values of dimensions F1 and F2 of the target hotel, and $u_i^1$ and $u_i^2$ denote the values of dimensions F1 and F2 of the competitive hotel $H_i$, respectively. The cosine similarity between the competitive position points of the target hotel $H_0$ and competitive hotel $H_i$ can be obtained by:

$$cos{H}_0,H_i=\frac{u_0^1{u}_i^1+u_0^2{u}_i^2}{\sqrt{{\left(u_0^1\right)}^2+{\left(u_0^2\right)}^2}·\sqrt{{\left(u_i^1\right)}^2+{\left(u_i^2\right)}^2}}, i=1,2,\cdots ,I$$

(14)

The closer the cosine similarity is to 1, the closer the competitive position between the two hotels is. Through Equation (14), the cosine similarity between the target hotel $H_0$ and every competitive hotel $H_i$ can be obtained. Sort the competitive hotels according to the cosine similarity in descending order, and the main competitors can be identified as the first M competitive hotels. The value of M can be determined by the hotel managers. Let $H^C=\left\{H_1^C,H_2^C,\cdots ,H_M^C\right\}$ denote the set of identified main competitor hotels, where $H_m^C$ is the mth main competitor hotel, $m=1,2,\cdots ,M$.

3.4. Conducting SWOT Analysis toward Main Competitors

Based on the information obtained from text mining of online reviews, SWOT analysis toward main competitors was conducted to obtain targeted strategic planning results. First, factors of each SWOT category were identified considering the performance of the target hotel and the main competitor hotels. Second, priorities of SWOT factors are determined according to customer concern level, performance and performance difference between the target hotel and the main competitor hotels. Finally, the results of strategic planning for hotel competitiveness improvement are obtained by combining SWOT factors into the TOWS matrix.

3.4.1. Identifying Factors of Each SWOT Category

In the method proposed by Phadermrod et al. [7], attributes that have high performance can be identified as strengths of the firm, while attributes that have low performance can be identified as weaknesses of the firm, and opportunity and threat factors can be identified through comparing the strength/weakness factors of the target firm with those of the main competitors’s. Albayrak [5] argues that the performance difference between the target hotel and the competitors should be considered in hotel strategy. Based on the above studies, a method for identifying SWOT factors of the target hotel is proposed in this study. In our method, SWOT factors of the target hotel are identified based on a performance comparison of the target hotel and the main competitors. Strengths and weaknesses of the target hotel and the main competitor hotels are preliminarily identified through comparing a single attribute’s performance with average attribute performance. Opportunity and threat factors are identified preliminarily based on comparing the strength/weakness factors of the target hotel and the main competitor hotels. Furthermore, the preliminarily identified SWOT factors are modified considering the performance difference between the target hotel and the main competitor hotels to obtain more candidate opportunity and threat factors for strategic planning.

For the target hotel $H_0$, the attribute performance $P{M}_{0j}$ can be obtained through the method described in Section 3.2. Let ${\overline{PM}}_0$ denote the average performance value of all the service attributes of the target hotel $H_0$. ${\overline{PM}}_0$ can be determined by:

$${\overline{PM}}_0=\frac{{{\displaystyle \sum }}_{j=1}^{J}P{M}_{0j}}{J},$$

(15)

For the main competitor hotel $H_m^C$, the attribute performance $P{M}_{mj}^C$ can be obtained through the method described in Section 3.2. Let ${\overline{PM}}_j^C$ denote the average performance value of the main competitor hotels on attribute $f_j$, and let ${\overline{PM}}^C$ denote the average performance value of all the attributes of the main competitor hotels. ${\overline{PM}}_j^C$ and ${\overline{PM}}^C$ can be obtained by:

$${\overline{PM}}_j^C=\frac{{{\displaystyle \sum }}_{m=1}^{M}P{M}_{mj}^C}{M}, j=1,2,\cdots ,J$$

(16)

$${\overline{PM}}^C=\frac{{{\displaystyle \sum }}_{m=1}^M{{\displaystyle \sum }}_{J=1}^{J}P{M}_{mj}^C}{M·J},$$

(17)

The detailed rules for preliminarily identifying factors of strength, weakness, opportunity and threat are described as PS, PW, PO and PT, respectively, and the managerial implications are given according to [7]:

Rule PS: Service attribute $f_j$ is identified as a strength factor of the target hotel preliminarily if the performance of $f_j$ is above average considering both the target hotel and the main competitor hotels. That is, if $P{M}_{0j}>{\overline{PM}}_0$ and ${\overline{PM}}_j^C>{\overline{PM}}^C$, service attribute $f_j$ is identified as a strength factor of the target hotel preliminarily. This means the target hotel and the main competitors all perform well, and the target hotel faces head-to-head competition on this factor.

Rule PW: Service attribute $f_j$ is identified as a weakness factor of the target hotel preliminarily if the performance of $f_j$ is below average considering both the target hotel and the main competitor hotels. That is, if $P{M}_{0j}<{\overline{PM}}_0$ and ${\overline{PM}}_j^C<{\overline{PM}}^C$, service attribute $f_j$ is identified as a weakness factor of the target hotel preliminarily. This means the target hotel and the main competitors perform poorly on this factor, and the target hotel can improve its performance on this factor to obtain competitive advantages.

Rule PO: Service attribute $f_j$ is identified as an opportunity factor of the target hotel preliminarily if the performance of $f_j$ is above average considering the target hotel and below average considering the main competitor hotels. That is, if $P{M}_{0j}>{\overline{PM}}_0$ and ${\overline{PM}}_j^C<{\overline{PM}}^C$, service attribute $f_j$ is identified as an opportunity factor of the target hotel preliminarily. This means the target hotel performs better than the main competitors and has a competitive advantage in this factor.

Rule PT: Service attribute $f_j$ is identified as a threat factor of the target hotel preliminarily if the performance of $f_j$ is below average considering the target hotel and above average considering the main competitor hotels. That is, if $P{M}_{0j}<{\overline{PM}}_0$ and ${\overline{PM}}_j^C>{\overline{PM}}^C$, service attribute $f_j$ is identified as a threat factor of the target hotel preliminarily. This means the target hotel performs poorer than the main competitors and has a competitive disadvantage in this factor.

Furthermore, more opportunity and threat factors can be identified considering performance differences between the target hotel and the main competitor hotels. If the performance of the target hotel on a factor which is preliminarily identified as a strength of the target hotel is lower than that of the main competitor hotels, the factor should be modified as a threat to the target hotel since it may be a competitive disadvantage of the target hotel. Meanwhile, if the performance of the target hotel on a factor which is preliminarily identified as a weakness of the target hotel is higher than that of the main competitor hotels, the factor should be modified as an opportunity of the target hotel since it may be a competitive advantage of the target hotel.

Let $P{D}_j$ denote the performance difference of attribute $f_j$ between the target hotel and the main competitor hotels. $P{D}_j$ can be obtained by:

$$P{D}_j=P{M}_{0j}-{\overline{PM}}_j^C, j=1,2,\cdots ,J$$

(18)

Rules for modifying the preliminarily identified SWOT factors are described as UO and UT.

Rule UO: Service attribute $f_j$, which is preliminarily identified as a weakness factor by Rule PW, is modified as an opportunity for the target hotel if the target hotel has a positive performance difference toward main competitors on $f_j$. That is, if $P{D}_j>0$, attribute $f_j$, which is identified as a weakness factor preliminarily by Rule PS, is ultimately identified as an opportunity factor.

Rule UT: Service attribute $f_j$, which is preliminarily identified as a strength factor by Rule PS, is modified as a threat to the target hotel if the target hotel has a negative performance difference toward main competitors on $f_j$. That is, if $P{D}_j<0$, attribute $f_j$, which is identified as strength preliminarily by Rule PS, is ultimately identified as a threat factor.

Based on the preliminary identifying rules PS, PW, PO and PT and the modified rules UO and UT, the rules for identifying factors in each SWOT category are summarized in

Table 3. Rules for SWOT factor identification.

Target Hotel	Main Competitor Hotels	Preliminarily Identified SWOT Factors	Difference in Performance	Ultimately Identified SWOT Factors
$P{M}_{0j}>{\overline{PM}}_0$	${\overline{PM}}_j^C>{\overline{PM}}^C$	Strength	$P{D}_j>0$	Strength
			$P{D}_j<0$	Threat
	${\overline{PM}}_j^C<{\overline{PM}}^C$	Opportunity	$P{D}_j>0$	Opportunity
$P{M}_{0j}<{\overline{PM}}_0$	${\overline{PM}}_j^C>{\overline{PM}}^C$	Threat	$P{D}_j<0$	Threat
	${\overline{PM}}_j^C<{\overline{PM}}^C$	Weakness	$P{D}_j>0$	Opportunity
			$P{D}_j<0$	Weakness

. Let $F_S=\left\{f_1^S,f_2^S,\cdots ,f_{U_S}^S\right\}$, $F_W=\left\{f_1^W,f_2^W,\cdots ,f_{U_W}^W\right\}$, $F_O=\left\{f_1^O,f_2^O,\cdots ,f_{U_O}^O\right\}$ and $F_T=\left\{f_1^T,f_2^T,\cdots ,f_{U_T}^T\right\}$ denote the factor set of strengths, weaknesses, opportunities and threats of the target hotel, respectively, where $f_u^S$, $f_u^W$, $f_u^O$ and $f_u^T$ are the factors in each SWOT category and $U_S$, $U_W$, $U_O$ and $U_T$ are the number of factors in each SWOT category. $F_S{{\displaystyle \cup }}^{ }{F}_W{{\displaystyle \cup }}^{ }{F}_O{{\displaystyle \cup }}^{ }{F}_T=F$, and $U_S+U_W+U_O+U_T=J$.

3.4.2. Determining Priorities of Factors in Each SWOT Category

In strategic planning for hotel competitiveness improvement, managers need to obtain the prioritization of the factors in each SWOT category so as to develop a strategy effectively when facing resource constraints. To determine the priority of the factors in each SWOT category, the customer concern level, the performance of the factors (service attributes), the performance difference between the target hotel and the main competitor hotels, and the characteristics and managerial implication of the SWOT category should be considered comprehensively. The method for determining the priorities of factors in each SWOT category is described as follows.

First, the customer concern level, the performance and the performance difference of the factors between the target hotel and the main competitor hotels are normalized into values in the interval [0,1]:

$${\overline{AT}}_{0j}=\frac{A{T}_{0j}-A{T}_0^{min}}{A{T}_0^{max}-A{T}_0^{min}} , j=1,2,\cdots ,J$$

(19)

$${\overline{PM}}_{0j}=\frac{P{M}_{0j}-P{M}_0^{min}}{P{M}_0^{max}-P{M}_0^{min}} , j=1,2,\cdots ,J$$

(20)

$${\overline{PD}}_j=\frac{P{D}_j-P{D}^{min}}{P{D}^{max}-P{D}^{min}} , j=1,2,\cdots ,J$$

(21)

where $A{T}_0^{min}=min\left\{A{T}_{0j}|j=1,2,\cdots ,J\right\}$, $A{T}_0^{max}=max\left\{A{T}_{0j}|j=1,2,\cdots ,J\right\}$, $P{M}_0^{min}=min\left\{P{M}_{0j}|j=1,2,\cdots ,J\right\}$, $P{M}_0^{max}=max\left\{P{M}_{0j}|j=1,2,\cdots ,J\right\}$, $P{D}^{min}=min\left\{P{D}_j|j=1,2,\cdots ,J\right\}$ and $P{D}^{max}=max\left\{P{D}_j|j=1,2,\cdots ,J\right\}$.

As the customer concern level reflects how much the customers attach importance to the factor, the factor with a higher customer concern level should be given higher priority. On the other hand, the performance and performance difference of the factor should be considered in determining factor priority according to the characteristics and the managerial implication of the SWOT category.

• For factors in the strength category, if a factor has a lower performance value, it means the target hotel is more likely to be surpassed by the competitors on this factor, so resources should be devoted to this factor with higher priority so as to maintain its strength. Let ${\overline{AT}}_u^S$ and ${\overline{PM}}_u^S$ denote the normalized customer concern level and performance of factor ${\mathrm{f}}_{\mathrm{u}}^{\mathrm{S}}$ in the strength factor set $F_S=\left\{f_1^S,f_2^S,\cdots ,f_{U_S}^S\right\}$ of the target hotel, and let ${\lambda }_u^S$ denote the priority of $f_u^S$. ${\lambda }_u^S$ can be given by:

  $${\lambda }_u^S={\alpha }_S{\overline{AT}}_u^S+{\beta }_S\left(1-{\overline{PM}}_u^S\right) , u=1,2,\cdots ,U_S$$

  (22)

In Equation (22), ${\alpha }_S$ and ${\beta }_S$ are the weight of customer concern level and factor performance, respectively. The values of ${\alpha }_S$ and ${\beta }_S$ can be determined by hotel managers according to managerial consideration.

• For factors in the weakness category, if a factor has a lower performance value, it means the target hotel performs worse on this factor and this factor is a more notable weakness; thus, resources should be devoted to this factor with higher priority so as to improve the performance and overcome the weakness. Let ${\overline{AT}}_u^W$ and ${\overline{PM}}_u^W$ denote the normalized customer concern level and the performance value of factor $f_u^W$ in the weakness factor set $F_W=\left\{f_1^W,f_2^W,\cdots ,f_{U_W}^W\right\}$ of the target hotel. The priority of $f_u^W$ can be given by:

  $${\lambda }_u^W={\alpha }_W{\overline{AT}}_u^W+{\beta }_W\left(1-{\overline{PM}}_u^W\right) , u=1,2,\cdots ,U_W$$

  (23)

In Equation (23), ${\alpha }_W$ and ${\beta }_W$ are the weight of customer concern level and factor performance, respectively, which can be determined by the hotel managers.

• For factors in the opportunity category, the performance difference should be considered because the opportunity is identified based on the comparison of the target hotel and the main competitor hotels. The performance of the target hotel on factors in the opportunity category is higher than that of the main competitors. If a factor has a higher performance difference, it means the factor has a greater impact on the competitiveness of the target hotel; thus, resources should be devoted to this factor with higher priority so as to seize the opportunity and obtain more dramatic competitive advantages. Let ${\overline{AT}}_u^O$ and ${\overline{PD}}_u^O$ denote the normalized customer concern level and the performance difference value of factor $f_u^O$ in the opportunity factor set $F_O=\left\{f_1^O,f_2^O,\cdots ,f_{U_O}^O\right\}$ of the target hotel. The priority of $f_u^O$ can be given by:

  $${\lambda }_u^O={\alpha }_O{\overline{AT}}_u^O+{\beta }_O {\overline{PD}}_u^O , u=1,2,\cdots ,U_O$$

  (24)

In Equation (24), ${\alpha }_O$ and ${\beta }_O$ are the weight of customer concern level and factor performance difference, respectively, which can be determined by the hotel managers.

• For factors in the threat category, the performance of the target hotel on these factors is lower than that of the main competitors. If the target hotel has a more significant difference on the factor—that is, the target hotel has a larger gap on this factor compared with the main competitors—the factor has a greater impact on the competitiveness of the target hotel; thus, resources should be devoted to this factor with higher priority so as to avoid the threat and make up for the competitive disadvantage. According to Equations (18) and (21), a lower value of the normalized performance difference of factors in the threat category means a greater gap between the target hotel and the main competitors. Let ${\overline{AT}}_u^T$ and ${\overline{PD}}_u^T$ denote the normalized customer concern level and the performance difference value of factor $f_u^T$ in the threat factor set $F_T=\left\{f_1^T,f_2^T,\cdots ,f_{U_T}^T\right\}$. The priority of ${\mathrm{f}}_{\mathrm{u}}^{\mathrm{T}}$ can be given by:

  $${\lambda }_u^T={\alpha }_T{\overline{AT}}_u^T+{\beta }_T\left(1-{\overline{PD}}_u^T\right) , u=1,2,\cdots ,U_T$$

  (25)

In Equation (25), ${\alpha }_T$ and ${\beta }_T$ are the weight of customer concern level and factor performance, respectively, which can be determined by the hotel managers.

3.4.3. Obtaining Strategic Planning Result for Hotel Competitiveness Improvement

Based on the obtained SWOT factors and priorities of factors in each SWOT category, strategic planning results for hotel competitiveness improvement can be presented according to the characteristics and managerial implications of each SWOT category.

• For factors in the strength category, the target hotel should maintain its performance to ensure that it will not be surpassed by the competitors on these factors. Meanwhile, more resources should be allocated to factors of higher priority to maintain competitiveness effectively.
• For factors in the weakness category, the target hotel should improve its performance to surpass the competitors and obtain new competitive advantages on these factors. Meanwhile, more resources should be allocated to factors of higher priority to seek new competitive advantages effectively.
• For factors in the opportunity category, the target hotel should maintain the performance to seize the opportunities and highlight the competitive advantages on these factors. Meanwhile, more resources should be allocated to factors of higher priority to seize the opportunity and improve competitiveness effectively.
• For factors in the threat category, the target hotel should take urgent action to improve the performance and make up for the competitive disadvantages. Meanwhile, more resources should be allocated to factors of higher priority to avoid threat and reverse competitive disadvantages effectively.

Furthermore, a TOWS matrix [64] can be constructed by combining threat/opportunity factors with weakness/strength factors to obtain more choices and more detailed suggestions to support strategic planning for hotel competitiveness improvement. The strategic planning results, in the form of a TOWS matrix, are shown in Figure 2. In the TOWS matrix, SO Strategy means that the target hotel should rely on its strengths and take advantage of opportunities. WO Strategy implies that the target hotel should take advantage of opportunities and overcome its weaknesses. ST Strategy means that the target hotel should rely on its strengths and avoid threats from the competitors. WT Strategy implies that the target hotel should overcome its weaknesses and avoid threats from the competitors. For each candidate strategy in the TOWS matrix, factor priorities can provide decision support on resource allocation priorities when the target hotel faces resource constraints.

4. Results

In this section, the results of an empirical study on the Mandarin Oriental Pudong Shanghai Hotel is given to illustrate the use of the method proposed in this study. The Mandarin Oriental Pudong Shanghai Hotel is a five-star hotel located in the Lujiazui Area, Shanghai, China. Lujiazui is the core functional area of the Shanghai International Financial Center, and high-end hotels are widely distributed in this area. Since the hotels in the Lujiazui are facing fierce competition, we considered the Mandarin Oriental Pudong Shanghai Hotel as the target hotel to conduct customer-oriented strategic planning for competitiveness improvement. We selected 13 other five-star hotels in the same area as the competitive hotels of the target hotel. Online reviews of the hotels were collected and preprocessed (Section 3.1). Based on the data described in Table 1, the results of the empirical study are given in this section.

4.1. Text Mining of Online Hotel Reviews

According to the process of extracting customer-concerned service attributes in Section 3.2.1, LDA is used to extract customer-concerned service attributes from online reviews of the target hotel and the competitive hotels. After constructing the review–word matrix based on preprocessed online reviews, the number of topics is set from 2 to 20 to train the LDA model. The performance of the LDA model is evaluated under different numbers of topics using perplexity and coherence scores; the number of topics in the LDA model is determined as 14 based on performance evaluation under different topic numbers. After manually merging the topics with similar meanings and filtering the noisy words, we obtained 12 customer-concerned service attributes. The extracted customer-concerned service attributes and examples of the related keywords are shown in

Table 4. Customer-concerned service attributes and related keywords.

${\mathit{f}}_{\mathit{j}}$	Customer-Concerned Service Attributes	${\mathit{P}}_{\mathit{j}}$	Examples of Related Keywords
$f_1$	入住/退房服务 (Check-in/out service)	35	入住(Check-in), 退房(check-out), 办理(handle), 预订(booking), 排队(queuing), …
$f_2$	客房服务 (Room service)	28	服务(service), 打扫(sweeping), 客房(hotel room), 垃圾(rubbish), 收拾(clear up), …
$f_3$	客房设施 (Room facilities)	32	浴缸(bathtub), 空调(air conditioner), 大床(big bed), 浴室(bathroom), 加床(extra bed), …
$f_4$	房间格局 (Room layout)	25	房型(room type), 套房(suite), 布置(arrange), 宽敞(spacious), 空间(space), …
$f_5$	服务人员 (Service staff)	37	热情(warmth), 经理(manager), 周到(considerate), 礼宾(concierge), 服务员(waiter), …
$f_6$	个性化服务 (Customized service)	27	酒廊(lounge), 免费(free), 停车(parking), 惊喜(surprise), 赠送(present offering), …
$f_7$	娱乐休闲设施 (Entertainment facilities)	21	泳池(swimming pool), 健身房(fitness room), 娱乐(entertainment), 休闲(relaxation), 硬件(hardware facility), …
$f_8$	性价比 (performance/price ratio)	27	性价比(Performance for price), 品牌(brand), 五星级(five-star level), 价格(price), 标准(standard), …
$f_9$	内部环境与卫生 (Internal environment and cleanliness)	24	环境(environment), 整洁(tidy), 卫生(cleanliness), 电梯(elevator), 大堂(lobby), …
$f_{10}$	周边环境 (Surrounding environment)	22	周边(surrounding), 便利(convenience), 风景(scenery), 视野(view), 滨江(riverside), …
$f_{11}$	地理位置/交通(Location/traffic)	32	位置(location), 广场(square), 地铁站(subway station), 东方明珠(Oriental Pearl Tower), 购物(shopping), …
$f_{12}$	餐饮服务 (Food service)	36	早餐(breakfast), 品种(variety), 好吃(delicious), 餐厅(canteen), 晚餐(dinner), …

. As shown in Table 4, the hotel customers mainly focus on 12 service attributes, i.e., Check-in/out service ($f_1$), Room service ($f_2$), Room facilities ($f_3$), Room layout ($f_4$), Service staff ($f_5$), Customized service ($f_6$), Entertainment facilities ($f_7$), Performance/price ratio ($f_8$), Internal environment and cleanliness ($f_9$), Surrounding environment ($f_{10}$), Location/traffic ($f_{11}$) and Food service ($f_{12}$). In Table 4, $P_j$ denotes the number of related keywords of service attribute $f_j$,$j=1,2,\cdots ,12$.

Then, after identifying online review information concerning each service attribute through the method described in Section 3.2.2, the customer concern level of the service attributes with respect to the target hotel is obtained based on Equation (1). Let $Fre{q}_{0j}$ denote the frequency that attribute $f_j$ is mentioned in online reviews of the target hotel. $H_0$ and $A{T}_{0j}$ denote the customer concern level of attribute $f_j$ with respect to the target hotel. The mentioned frequency and customer concern level of the attributes with respect to the target hotel are shown in

Table 5. The mentioned frequency and customer concern level of the attributes concerning the target hotel.

${\mathit{f}}_{\mathit{j}}$	$\mathit{F}\mathit{r}\mathit{e}{\mathit{q}}_{0\mathit{j}}$	$\mathit{A}{\mathit{T}}_{0\mathit{j}}$
$f_1$	2098	0.12
$f_2$	2263	0.13
$f_3$	880	0.05
$f_4$	2256	0.13
$f_5$	1611	0.09
$f_6$	1069	0.06
$f_7$	689	0.04
$f_8$	1708	0.10
$f_9$	758	0.04
$f_{10}$	1178	0.07
$f_{11}$	1346	0.08
$f_{12}$	1655	0.09

. As shown in Table 5, service attributes including Room service ($f_2$), Room layout ($f_4$) and Check-in/out service ($f_1$) are mentioned more frequently in online reviews of the target hotel. This means the customers of the target hotel are more concerned with these service attributes.

Furthermore, multi-granularity sentiment analysis was conducted to evaluate the performance of the service attributes of each hotel through the method described in Section 3.2.2. After obtaining the opinion word set of online review information and constructing a domain sentiment dictionary for each attribute, the sentiment strength value of online review information can be obtained based on Equations (6)–(10). Here, due to limitations of space, we take the target hotel $H_0$ and the competitive hotels $H_1$ and $H_{13}$ as examples to illustrate the results of sentiment analysis. Based on the results of multi-granularity sentiment analysis, the amounts of online review information of sentiment strength values −2, −1, 0, 1, 2 concerning attribute $f_j$ of hotel $H_0$, $H_1$ and $H_{13}$ are shown in

Table 6. Results of sentiment analysis for hotels $H_0$, $H_1$ and $H_{13}$.

${\mathit{f}}_{\mathit{j}}$	${\mathit{H}}_0$	${\mathit{H}}_1$	${\mathit{H}}_{13}$
$f_1$	$-2_{\left(36\right)},-1_{\left(269\right),}{0}_{\left(996\right)},1_{\left(530\right)},2_{\left(267\right)}$	$-2_{\left(20\right)},-1_{\left(124\right),}{0}_{\left(411\right)},1_{\left(194\right)},2_{\left(108\right)}$	$-2_{\left(8\right)},-1_{\left(102\right),}{0}_{\left(466\right)},1_{\left(359\right)},2_{\left(290\right)}$
$f_2$	$-2_{\left(57\right)},-1_{\left(213\right),}{0}_{\left(631\right)},1_{\left(745\right)},2_{\left(617\right)}$	$-2_{\left(35\right)},-1_{\left(91\right),}{0}_{\left(248\right)},1_{\left(261\right)},2_{\left(224\right)}$	$-2_{\left(19\right)},-1_{\left(36\right),}{0}_{\left(248\right)},1_{\left(527\right)},2_{\left(484\right)}$
$f_3$	$-2_{\left(24\right)},-1_{\left(134\right),}{0}_{\left(331\right)},1_{\left(256\right)},2_{\left(135\right)}$	$-2_{\left(27\right)},-1_{\left(66\right),}{0}_{\left(201\right)},1_{\left(143\right)},2_{\left(61\right)}$	$-2_{\left(32\right)},-1_{\left(75\right),}{0}_{\left(189\right)},1_{\left(197\right)},2_{\left(69\right)}$
$f_4$	$-2_{\left(16\right)},-1_{\left(193\right),}{0}_{\left(624\right)},1_{\left(798\right)},2_{\left(625\right)}$	$-2_{\left(11\right)},-1_{\left(71\right),}{0}_{\left(330\right)},1_{\left(420\right)},2_{\left(409\right)}$	$-2_{\left(13\right)},-1_{\left(75\right),}{0}_{\left(231\right)},1_{\left(420\right)},2_{\left(410\right)}$
$f_5$	$-2_{\left(11\right)},-1_{\left(89\right),}{0}_{\left(471\right)},1_{\left(545\right)},2_{\left(495\right)}$	$-2_{\left(5\right)},-1_{\left(50\right),}{0}_{\left(228\right)},1_{\left(183\right)},2_{\left(178\right)}$	$-2_{\left(0\right)},-1_{\left(26\right),}{0}_{\left(354\right)},1_{\left(526\right)},2_{\left(432\right)}$
$f_6$	$-2_{\left(8\right)},-1_{\left(117\right),}{0}_{\left(460\right)},1_{\left(361\right)},2_{\left(123\right)}$	$-2_{\left(4\right)},-1_{\left(56\right),}{0}_{\left(166\right)},1_{\left(150\right)},2_{\left(56\right)}$	$-2_{\left(5\right)},-1_{\left(35\right),}{0}_{\left(175\right)},1_{\left(150\right)},2_{\left(39\right)}$
$f_7$	$-2_{\left(6\right)},-1_{\left(82\right),}{0}_{\left(214\right)},1_{\left(279\right)},2_{\left(108\right)}$	$-2_{\left(7\right)},-1_{\left(34\right),}{0}_{\left(84\right)},1_{\left(122\right)},2_{\left(51\right)}$	$-2_{\left(14\right)},-1_{\left(31\right),}{0}_{\left(76\right)},1_{\left(130\right)},2_{\left(34\right)}$
$f_8$	$-2_{\left(19\right)},-1_{\left(115\right),}{0}_{\left(314\right)},1_{\left(924\right)},2_{\left(336\right)}$	$-2_{\left(7\right)},-1_{\left(51\right),}{0}_{\left(147\right)},1_{\left(593\right)},2_{\left(218\right)}$	$-2_{\left(13\right)},-1_{\left(39\right),}{0}_{\left(189\right)},1_{\left(729\right)},2_{\left(275\right)}$
$f_9$	$-2_{\left(5\right)},-1_{\left(47\right),}{0}_{\left(174\right)},1_{\left(351\right)},2_{\left(181\right)}$	$-2_{\left(3\right)},-1_{\left(23\right),}{0}_{\left(112\right)},1_{\left(263\right)},2_{\left(162\right)}$	$-2_{\left(17\right)},-1_{\left(34\right),}{0}_{\left(86\right)},1_{\left(369\right)},2_{\left(178\right)}$
$f_{10}$	$-2_{\left(4\right)},-1_{\left(50\right),}{0}_{\left(340\right)},1_{\left(529\right)},2_{\left(255\right)}$	$-2_{\left(2\right)},-1_{\left(36\right),}{0}_{\left(365\right)},1_{\left(462\right)},2_{\left(232\right)}$	$-2_{\left(2\right)},-1_{\left(16\right),}{0}_{\left(182\right)},1_{\left(312\right)},2_{\left(153\right)}$
$f_{11}$	$-2_{\left(6\right)},-1_{\left(74\right),}{0}_{\left(478\right)},1_{\left(534\right)},2_{\left(254\right)}$	$-2_{\left(6\right)},-1_{\left(33\right),}{0}_{\left(511\right)},1_{\left(837\right)},2_{\left(504\right)}$	$-2_{\left(4\right)},-1_{\left(26\right),}{0}_{\left(438\right)},1_{\left(545\right)},2_{\left(297\right)}$
$f_{12}$	$-2_{\left(13\right)},-1_{\left(102\right),}{0}_{\left(659\right)},1_{\left(473\right)},2_{\left(408\right)}$	$-2_{\left(6\right)},-1_{\left(28\right),}{0}_{\left(269\right)},1_{\left(208\right)},2_{\left(231\right)}$	$-2_{\left(4\right)},-1_{\left(23\right),}{0}_{\left(304\right)},1_{\left(256\right)},2_{\left(309\right)}$

. For example, the content in line 1, column 1 of Table 6 is “$-2_{\left(36\right)},-1_{\left(269\right),}{0}_{\left(996\right)},1_{\left(530\right)},2_{\left(267\right)}$”, which means in online reviews of the target hotel $H_0$, there are 36, 269, 996, 530, 267 pieces of online review information whose sentiment strength value is −2, −1, 0, 1, 2, respectively.

Based on the results of multi-granularity sentiment analysis of the target hotel and the competitive hotels, the performance of the hotels on each service attribute can be evaluated based on Equation (11). The results of the performance evaluation are shown in

Table 7. Performance of service attributes of the target hotel and the competitive hotels.

${\mathit{f}}_{\mathit{j}}$	${\mathit{H}}_0$	${\mathit{H}}_1$	${\mathit{H}}_2$	${\mathit{H}}_3$	${\mathit{H}}_4$	${\mathit{H}}_5$	${\mathit{H}}_6$	${\mathit{H}}_7$	${\mathit{H}}_8$	${\mathit{H}}_9$	${\mathit{H}}_{10}$	${\mathit{H}}_{11}$	${\mathit{H}}_{12}$	${\mathit{H}}_{13}$
$f_1$	0.3446	0.2870	0.6422	0.3666	0.3146	0.3638	0.2939	0.3845	0.2536	0.5571	0.3024	0.4174	0.6978	0.6702
$f_2$	0.7300	0.6380	0.9576	0.6468	0.6362	0.7824	0.6327	0.7468	0.4684	1.0510	0.6636	0.6908	1.1787	1.0814
$f_3$	0.3909	0.2912	0.0406	0.2079	0.2586	0.1837	0.4433	0.4870	0.1888	0.5017	0.0718	0.0994	0.5728	0.3488
$f_4$	0.8081	0.9226	0.6667	0.7724	0.7872	0.6970	0.7824	0.8350	0.6742	0.9669	0.6787	0.7559	1.0186	0.9913
$f_5$	0.8839	0.7438	1.0276	0.8863	0.8662	0.9552	0.8684	1.0009	0.8582	1.0979	0.9098	0.8640	1.1586	1.0194
$f_6$	0.4434	0.4583	0.4415	0.5004	0.4439	0.4144	0.3939	0.4278	0.3384	0.4974	0.3792	0.3640	0.5605	0.4530
$f_7$	0.5820	0.5906	0.1825	0.4955	0.5382	0.4747	0.5663	0.6529	0.4732	0.6446	0.2857	0.3687	0.7137	0.4877
$f_8$	0.8448	0.9488	0.9206	0.8383	0.8830	0.8729	0.8785	0.8776	0.7019	0.9995	0.8075	0.8561	1.0386	0.9751
$f_9$	0.8654	0.9911	1.0210	0.7832	0.8801	0.7963	0.7835	0.7348	0.5865	1.0260	0.8007	0.8603	1.0999	0.9605
$f_{10}$	0.8328	0.8077	0.9537	0.8342	0.8069	0.8980	0.7373	0.8773	0.8261	0.8896	0.7870	0.8326	1.0446	0.8992
$f_{11}$	0.7103	0.9519	0.9239	0.7807	0.8902	0.8022	0.5606	0.8089	0.7108	0.6484	0.9297	0.8368	0.8788	0.8435
$f_{12}$	0.7015	0.8491	0.8360	0.6363	0.7085	0.7911	0.7136	0.7345	0.6493	0.9262	0.7542	0.7947	1.0274	0.9408

. It can be seen from Table 7 that the target hotel and the competitive hotels have differences in performance of the service attributes. The target hotel performs best on attribute “Internal environment and cleanliness ($f_9$)” and performs worst on attribute “Check-in/out service ($f_1$)”.

4.2. Competitive Structure Analysis and Main Competitors Identification

After text mining of online hotel reviews, the competitive structure of the hotels was analyzed and the main competitors of the target hotel were identified through the method in Section 3.1. Based on the results of the online review text mining shown in Section 4.1, the perception frequency of the online review information was obtained through Equation (12), and the total perception frequency of service attribute $f_j$ in the online reviews of hotel $H_i$ was obtained through Equation (13). Then, the contingency table of perception frequencies of the service attributes and the hotels was constructed, shown in

Table 8. Contingency table of perception frequencies of the service attributes and the hotels.

${\mathit{H}}_{\mathit{i}}$	${\mathit{f}}_1$	${\mathit{f}}_2$	${\mathit{f}}_3$	${\mathit{f}}_4$	${\mathit{f}}_5$	${\mathit{f}}_6$	${\mathit{f}}_7$	${\mathit{f}}_8$	${\mathit{f}}_9$	${\mathit{f}}_9$	${\mathit{f}}_{11}$	${\mathit{f}}_{12}$
$H_0$	2401	2937	1039	2897	2117	1200	803	2063	944	1437	1606	2076
$H_1$	985	1118	586	1662	827	492	356	1241	728	1331	2401	979
$H_2$	2330	3019	1331	2056	2436	685	595	2731	1213	1443	3762	1576
$H_3$	9523	7780	2808	7720	6275	2840	2083	4941	2754	4327	4454	5007
$H_4$	3880	4119	2065	4620	3070	1726	1409	3912	1729	2582	4402	3246
$H_5$	1699	1744	637	1728	1520	682	505	1080	538	908	1283	1162
$H_6$	1544	1444	673	1621	1176	601	530	1202	648	671	555	1329
$H_7$	1630	1786	933	1795	1492	874	965	1125	496	606	960	1356
$H_8$	1335	1430	610	1419	918	510	418	1010	644	873	652	1047
$H_9$	3045	4202	1432	4396	3694	1487	983	2797	1916	2046	1629	3777
$H_{10}$	804	996	430	818	669	260	241	816	379	345	785	763
$H_{11}$	7383	6259	2088	5998	5756	2079	1528	4335	2150	3319	5503	4094
$H_{12}$	1053	1635	515	1419	1217	512	294	1022	1082	549	1134	992
$H_{13}$	1523	1817	663	1572	1770	448	333	1533	879	820	1611	1209

. Based on the contingency table, SPSS software was used to conduct correspondence analysis, and the results are visualized in a perceptual map shown in Figure 3. In the perceptual map shown in Figure 3, the two dimensions F1 and F2 are the factors with the two highest inertia values in the correspondence analysis. F1 and F2 can be considered as the factors which contain the most information about the contingency table in the reduced space obtained by correspondence analysis. As shown in Figure 3, the cumulative inertia ratio of the two dimensions is 71.0% and the inertia ratios of dimensions F1 and F2 are 52.2% and 18.8%, respectively, indicating that F1 and F2 are both statistically significant dimensions.

Based on the results of correspondence analysis, the cosine similarity between the competitive position of the target hotel and each competitive hotel can be obtained through Equation (14). The coordinate values on dimensions F1 and F2 and the cosine similarity of the competitive position points of the target hotel and each competitive hotel are shown in

Table 9. Coordinate values and cosine similarity of the competitive position points.

${\mathit{H}}_{\mathit{i}}$	${\mathit{u}}_{\mathit{i}}^1$	${\mathit{u}}_{\mathit{i}}^2$	$\mathit{c}\mathit{o}\mathit{s}{\mathit{H}}_0,{\mathit{H}}_{\mathit{i}}$
$H_0$	0.208	0.134	-
$H_1$	−0.957	0.161	−0.739
$H_2$	−0.722	0.016	−0.828
$H_3$	0.246	−0.332	0.065
$H_4$	−0.254	0.122	−0.523
$H_5$	0.074	−0.145	−0.100
$H_6$	0.467	0.211	0.989
$H_7$	0.403	0.088	0.937
$H_8$	0.272	0.161	0.999
$H_9$	0.396	0.458	0.959
$H_{10}$	−0.134	0.239	0.061
$H_{11}$	−0.07	−0.345	−0.698
$H_{12}$	−0.08	0.521	0.408
$H_{13}$	−0.241	0.144	−0.443

. It can be seen from Table 9 that competitive hotel $H_8$ has the highest cosine similarity with the target hotel concerning their competitive positions. Here, we identify the four hotels with the highest cosine similarity as the main competitor hotels of the target hotel. The main competitor hotels can be sorted in descending order of cosine similarity as $H_8$, $H_6$, $H_9$ and $H_7$.

4.3. Obtaining Strategic Planning Result of the Target Hotel

Using Equations (15)–(18), the performance analysis of the target hotel and the main competitors is conducted. The average performance value of all the service attributes of the target hotel $H_0$, the average performance value of the main competitor hotels on each attribute $f_j$, the average value of all the attributes of the main competitor hotels and the performance difference of attribute $f_j$ between the target hotel and the main competitor hotels, i.e., ${\overline{PM}}_0$, ${\overline{PM}}_j^C$, ${\overline{PM}}^C$ and $P{D}_j$, are calculated. The performance value of the target hotel on each attribute and the above-mentioned performance analysis results are shown in

Table 10. Performance of the target hotel and results of performance analysis.

${\mathit{f}}_{\mathit{j}}$	$\mathit{P}{\mathit{M}}_{0\mathit{j}}$	${\overline{\mathit{P}\mathit{M}}}_{\mathit{j}}^{\mathit{C}}$	$\mathit{P}{\mathit{D}}_{\mathit{j}}$
$f_1$	0.3446	0.3723	−0.0277
$f_2$	0.7300	0.7247	0.0053
$f_3$	0.3909	0.4052	−0.0143
$f_4$	0.8081	0.8146	−0.0066
$f_5$	0.8839	0.9564	−0.0724
$f_6$	0.4434	0.4144	0.0290
$f_7$	0.5820	0.5842	−0.0022
$f_8$	0.8448	0.8644	−0.0195
$f_9$	0.8654	0.7827	0.0827
$f_{10}$	0.8328	0.8326	0.0002
$f_{11}$	0.7103	0.6822	0.0281
$f_{12}$	0.7015	0.7559	−0.0544
	${\overline{PM}}_0$=0.6781	${\overline{PM}}^C$=0.6825

.

Based on the performance analysis results shown in Table 10, SWOT factors of the target hotel are identified according to the rules proposed in Section 3.4.1, and the priorities of factors in each SWOT category are determined by the method described in Section 3.4.2. Here, the weights of customer concern level and factor performance are both set to 0.5. Factors and factor priorities in each SWOT category of the target hotel are shown in

Table 11. Factors and factor priorities in each SWOT category.

Strength		Weakness
Strength factor	Priority	Weakness factor	Priority
Room service	0.643	Check-in/out service	0.944
Surrounding environment	0.214	Room facilities	0.513
Internal environment and cleanliness	0.017	Entertainment facilities	0.280
Opportunity		Threat
Opportunity factor	Priority	Threat factor	Priority
Location/traffic	0.487	Room layout	0.621
Customized service	0.147	Food service	0.449
		performance/price ratio	0.412
		Service staff	0.278

.

Based on the results of SWOT analysis and the priorities of SWOT factors shown in Table 11, strategic planning results of the target hotel can be obtained.

• The strength factors of the target hotel are “Room service”, “Surrounding environment” and “Internal environment and cleanliness”. The target hotel should make efforts to maintain its performance on these factors to ensure its competitiveness. Specifically, the hotel should improve the quality of room service staff and focus on service with a smile and timeliness in solving customers’ problems to maintain the quality of room service. At the same time, the hotel should pay attention to the development of the surrounding environment and propagate the advantage of the surrounding environment to attract more customers, and maintain the strength of the hotel in terms of internal environment and cleanliness to further improve customer satisfaction. Furthermore, according to the priorities shown in Table 11, when the hotel faces resource constraints, the priority of maintaining the strength factor is “Room service” ≻ “Surrounding environment” ≻ “Internal environment and cleanliness”.
• The weakness factors of the target hotel are “Check-in/out service”, “Room facilities” and “Entertainment facilities”. The target hotel should improve its performance on these factors to obtain new competitive advantages. Specifically, the hotel should improve check-in/out processing efficiency and introduce new check-in/out processing methods to reduce customer waiting time, conduct regular inspections and maintenance of room facilities and entertainment facilities, and update old facilities to ensure satisfactory customer experience. When the hotel faces resource constraints, the priority of improving the weakness factor is “Check-in/out service” ≻ “Room facilities” ≻ “Entertainment facilities”.
• The opportunity factors of the target hotel are “Location/traffic” and “Customized service”. The target hotel should maintain the performance and highlight the competitive advantages. Specially, the hotel should take the superior location and convenient transportation as the focus of marketing and advertising to attract more customers, maintain the competitive advantage in customized services and set up a variety of customized service content to meet different customer requirements to further improve customer satisfaction. The priority of maintaining opportunity factors is “Location/traffic” ≻ “Customized service” when facing resource constraints.
• The threat factors of the target hotel are “Room layout”, “Food service”, “Performance/price ratio” and “Service staff”. The target hotel should take urgent action to make up for these competitive disadvantages. Specially, the hotel should improve the room layout according to customer opinions, enhance the variety of food and beverage options and carry out training and supervision of service staff. Through comprehensive improvement of service quality to enhance the customer perception of performance/price ratio, the hotel can further enhance its competitiveness. When the hotel faces resource constraints, the priority of avoiding the threat factor is “Room layout” ≻ “Food service” ≻ “Performance/price ratio” ≻ “Service staff”.

Based on the above results, the strategic planning strategy for the target hotel in the form of a TOWS matrix is shown in Figure 4. In the TOWS matrix shown in Figure 4, the priorities of factors in each SWOT category are also marked in the brackets after the factors to provide priority suggestions in strategic planning. The obtained TOWS matrix can provide more suggestions for strategic planning for hotel competitiveness improvement.

5. Discussion

Customer opinions are vital information that should be considered in strategic planning for hotel competitiveness improvement. The purpose of this study was to develop a method for customer-oriented strategic planning for hotel competitiveness improvement based on text mining of online reviews. In this section, some discussions on the results of the empirical study are given.

From the results of online review text mining presented in Section 4.1, it can be seen that richer and more specific information can be obtained through text mining of online reviews, and such information is important for analyzing competitive hotel relationships and developing strategies for hotel competitiveness improvement. First, more abundant and targeted service attributes that the customers are concerned with are extracted from online hotel reviews through the LDA method. As shown in Table 4, 12 customer-concerned service attributes were extracted from online reviews. In contrast, the Ctrip platform only provides online ratings on four service attributes in its review system.

The target hotel and the competitive hotels considered in the empirical study are high-end five-star hotels. Some extracted service attributes shown in Table 4, including “Customized service”, “Entertainment facilities” and “Food service”, reflect the service content commonly provided by high-end hotels when the hotels seek to achieve higher customer satisfaction. Hence, the result of service attribute extraction is consistent with the characteristics of the target hotel and the competitive hotels. From the customer concern level of the target hotel shown in Table 5, it can be seen that the customer concern level of the service attribute “Internal environment and cleanliness” is not high. This is because high-end hotels generally perform well on this attribute, and the customers of such hotels do not pay special attention to that attribute. Moreover, from the result of attribute performance evaluation shown in Table 7, it can be seen that more specific information on attribute performance can be obtained through multi-granularity sentiment analysis of online text reviews. In addition, the online reviews of the target hotel and the competitive hotels can be collected from online travel platforms with a relatively low cost of data collection. Through text mining of a large amount of online reviews, meaningful information can be obtained to provide effective information support for analyzing competitive hotel relationships and developing a strategy for competitiveness improvement.

Using the method proposed in Section 3.3, the competitive structure of hotels was analyzed and the main competitors were identified, which can help obtain more targeted strategic planning results. From the results of competitive structure analysis and main competitor identification shown in Section 4.2, the hotel competitive structure is visualized through perceptual mapping, which increases the clarity and intuitiveness of the results and can provide hotel managers with a more convenient way to understand the competitive structure and determine the main competitors. It is necessary to point out that our method also provides the hotel managers with a flexible way to determine the main competitors. Based on the results of cosine similarity calculation, suggestions on main competitors can be provided and the main competitor hotels can be added or deleted according to the hotel manager’s judgment of the competitive situation.

In the method proposed in Section 3.4, SWOT factors of the target hotel are identified based on a performance comparison of the target hotel and the main competitors, and more opportunity and threat factors can be identified to provide more choices and references for hotel competitive strategic planning. In our method, the priorities of SWOT factors are determined based on the customer concern level and the performance of the service attributes, which can be obtained from text mining of online reviews. Compared with existing studies, our method can obtain quantitative SWOT analysis results with less subjective information from the managers. To some extent, our method can make up for the shortcomings of the traditional SWOT method, which is highly subjective and lacks quantitative prioritization of SWOT factors. From the results of strategic planning for hotel competitiveness improvement shown in Section 4.3, it can be seen that the strategic planning results are based on the results of SWOT factor identification and priority determination shown in Table 11, and resource constraints are also considered in strategic planning. In addition, further strategic planning results are presented in the form of a TOWS matrix as shown in Figure 4, which can provide more choices and references on effective strategic planning. The purpose of our method is to provide hotel managers with more specific and effective decision support for strategic planning for hotel competitiveness improvement. Based on a comprehensive assessment of the resource conditions, development stage and competitive situation of the hotel, the hotel managers can choose or synthesize a variety of competitive strategies obtained through the methods in this study to formulate effective competitiveness improvement strategies for the target hotel.

6. Conclusions

In this study, we present an integrated method for customer-oriented strategic planning for hotel competitiveness improvement based on online reviews. In the method, first, online reviews of the target hotel and the competitive hotels were collected and preprocessed by web crawler software and the Jieba library in Python. Second, customer-concerned service attributes were extracted from online reviews through the LDA method. Customer concern level and the performance of the service attributes were evaluated through mentioned frequency statistics and multi-granularity sentiment analysis. Third, the competitive structures of the hotels were analyzed and visualized through correspondence analysis and perceptual mapping, and the main competitors of the target hotel were identified. Finally, SWOT analysis toward main competitors was conducted. In the SWOT analysis process, factors in each SWOT category were identified and the priorities of the factors were determined. Strategic planning results for competitive improvement of the target hotel were obtained based on SWOT analysis and the TOWS matrix. The major contributions of this study can be summarized as follows.

First, a novel and integrated solution framework for customer-oriented strategic planning for hotel competitiveness improvement based on online reviews is proposed. In the solution framework, methods for online review data collection and preprocessing, online review text mining, competitive structure analysis and main competitors identification and strategic planning based on SWOT analysis are integrated to obtain more effective and specific strategic planning results for hotel competitiveness improvement. Compared with existing studies, online reviews are used as data sources for strategic planning to obtain more detailed customer opinions, and customer-oriented strategic planning is conducted for hotels, which are typical service-oriented businesses. The proposed framework enriches the relevant research on hotel strategic planning and provides new ideas and specific methods to support effective strategic planning for hotel competitiveness improvement.

Second, meaningful information for strategic planning for hotel competitiveness improvement is obtained through an effective online review text mining method. Based on the LDA method and multi-granularity sentiment analysis, more abundant and specific information about customer-concerned service attributes and hotel performance on these attributes is obtained, which can provide effective support for strategic planning. This is an in-depth application of online review data in the research field of strategic planning.

Third, a method for identifying SWOT factors and determining quantitative priorities of SWOT factors is proposed. The proposed method utilizes information obtained from text mining of online reviews and needs less subjective information. Furthermore, more opportunity and threat factors can be identified, and quantitative priorities can be obtained through our method. The proposed method can overcome shortcomings of traditional SWOT analysis in terms of high subjectivity and a lack of quantitative results.

Moreover, the proposed method has a clear structure and is easily implemented. It has strong operability and effectiveness. This study provides a new choice in methods for customer-oriented strategic planning for hotel competitiveness improvement. The strategic planning results obtained through the proposed method are more practical and specific. This study has significant contributions not only to the literature but also in practice in strategic planning for hotel competitiveness improvement.

This study also has some limitations which provide suggestions for future research. First, online reviews were collected from the Ctrip platform in this study. Online reviews from multiple platforms can be utilized to obtain more sufficient information. Second, this study used dictionary-based sentiment analysis to evaluate attribute performance of hotels. Machine learning methods can be adopted to improve the efficiency of sentiment analysis. Third, this study utilized online review data to conduct customer-oriented strategic planning. Multiple data sources, including online review data and the firm’s internal data, can be used to obtain more comprehensive strategic planning results.