A Data Analytics and Machine Learning Approach to Develop a Technology Roadmap for Next-Generation Logistics Utilizing Underground Systems

Abstract

The increasing density of urban populations has spurred interest in utilizing underground space. Underground logistics systems (ULS) are gaining traction due to their effective utilization of this space to enhance urban spatial efficiency. However, research on technological advancements in related fields remains limited. To address this gap, we applied a data-driven approach using patent data related to the ULS to develop a technology roadmap for the field. We employed Latent Dirichlet Allocation (LDA), a machine learning-based topic modeling technique, to categorize and identify six specific technology areas within the ULS domain. Subsequently, we conducted portfolio analytics to pinpoint technology areas with high technological value and to identify the major patent applicants in these areas. Finally, we assessed the technology market potential by mapping the technology life cycle for the identified high-value areas. Among the six technology areas identified, Topic 1 (Underground Material Handling System) and Topic 4 (Underground Transportation System) showed significant patent activity from companies and research institutions in China, the United States, South Korea, and Germany compared to other countries. These areas have the top 10 patent applicants, accounting for 20.8% and 13.6% of all patent applications, respectively. Additionally, technology life cycle analytics revealed a growth trajectory for these identified areas, indicating their rapid expansion and high innovation potential. This study provides a data-driven methodology to develop a technology roadmap that offers valuable insights for researchers, engineers, and policymakers in the ULS industry and supports informed decision-making regarding the field’s future direction.

1. Introduction

Rapid urbanization and growing population densities in modern cities have necessitated innovative solutions to optimize space utilization and alleviate surface congestion. Underground Logistics Systems (ULS) have emerged as a promising alternative for efficient freight transport and distribution within urban environments. By leveraging the vast, undeveloped potential of underground space, ULS offers a sustainable and efficient approach to addressing the logistics challenges of megacities [1,2,3].

Despite the growing interest in ULS technology, a comprehensive roadmap for technology trends and promising technologies in detailed technology fields (subfields) is still insufficient. This hinders the development of ULS technology and makes it difficult to understand the technology landscape. This study aims to identify detailed technology trends in the ULS technology field through patent data analytics and to derive promising technologies by comparing technological advantages in each field. Ultimately, it aims to contribute to the establishment of a comprehensive technology roadmap for ULS technology development and implementation.

Patent data offer several advantages for analyzing ULS technology. It is a vast repository of technological information that provides insights into broad development trends within the ULS domain. Furthermore, the rigorous examination process that patent applications undergo makes them an excellent data source for assessing the originality and advancement of technology [4,5,6]. Analyzing patent applicants and inventors allows for the identification of major players and the competitive landscape [7,8,9]. By examining the application trends and technological advancements, we can also predict the future growth potential of the ULS field. This study aims to develop a data-driven technology roadmap for next-generation ULS. To achieve this, we will address three primary research questions.

Firstly, we classify specific technology areas within the ULS environment. At a time when the exact definition of underground logistics is ambiguous, a detailed technology classification and definition of the underground logistics field are necessary to identify technological opportunities in the field. Therefore, this study aims to identify and classify the various technological components that make up the ULS using topic modeling of patent data.

Secondly, we identify the specific entities that are leading high-value ULS technologies. It is necessary to identify promising sub-technology areas, as there are subfields in the underground logistics field, but not all fields can be promising. Therefore, we will identify companies and research institutes leading innovation in the most promising areas of ULS technology through portfolio analytics.

Lastly, we identify technological opportunities for high-value ULS sub-technologies. Without a clear definition of underground logistics and technology development strategy, it is possible to meaningfully utilize the identification of technological opportunities in the R&D strategy. Therefore, we will analyze the technology life cycle of the identified high-value areas to evaluate the growth trajectory of the area and predict the potential impact on the future of ULS. These predictions are very valuable to policymakers and investors who make strategic decisions about the future of urban infrastructure. By answering these questions, this study will provide valuable insights to researchers, engineers, and policymakers, enabling informed decision-making regarding the future direction of ULS development and implementation. Ultimately, this research aims to pave the way for the widespread adoption of ULS, contributing to the creation of a more sustainable and efficient urban transportation network.

Through this study, we reveal the key contributions to underground logistics systems (ULS). Through ULS patent analytics, critical patent applicants and high-value technology opportunities were identified, along with key technical topics. A quantitative methodology for establishing R&D strategies and uncovering technology opportunities was developed to address the gap in comprehensive research utilizing patent data in the ULS field. Furthermore, portfolio analytics pinpointed specific high-value technology areas and leading companies/research institutes. Patent technology life cycle analytics then predicted the future growth potential and the impact of these technologies on ULS development. These findings will be invaluable to researchers, policymakers, and stakeholders in the ULS domain.

2. Literature Review

The increasing population density in urban areas has increased interest in utilizing underground space more effectively. Underground logistics systems (ULSs) have emerged as a promising solution that offer efficient utilization of this space for freight transportation and distribution. Guo et al. [1] explored the planning and application of underground logistics systems (ULS) in new Chinese cities, focusing on the suitability of various goods and cargo types for ULS, demand forecasting, and network integration with existing infrastructure. Dong, et al. [2] employed a system dynamics approach to analyze the impact of ULS implementation on urban sustainability in Beijing, simulating different strategies and evaluating the impact on traffic congestion, delivery times, and emissions. Hu, et al. [3] developed a discrete event simulation model to support real-time operational decision-making in metro-based ULS, investigating different integration schemes and their performance in various scenarios. However, despite the potential benefits of the ULS, studies on identifying technology opportunities in this field through patent analytics still need to be completed. Related studies have identified technology opportunities through patent analytics in various fields.

Table 1. Summary of Studies Identifying Technology Opportunities through Patent Analytics (1963–2022).

Technology Topic	Period	Dataset	Reference
Blockchain	1974–2021	6206	[10]
Direct Seawater Electrolysis (DSE)	1963–2022	138	[11]
Photovoltaic Cells	2004–2013	22,682	[12]
5G-V2X	2014–2021	10,671	[13]
AI and Wind Power	1990–2022	2411	[14]
Airless and Smart Tire	1966–2022	47,998	[15]
Traditional Chinese Medicine Prescriptions	2001–2019	2962	[16]
Sensors in Postharvest	All Period	881	[17]
ARD Green Inventory	2001–2020	69,412	[18]
Offshore LNG Storage and Transportation	1969–2020	689	[19]
Biological Water Treatment	1967–2021	50,326	[20]
Silicon Fen (SF)	1999–2021	1140	[21]

summarizes the literature that identifies technology opportunities through patent analytics. Stating the period of the dataset helps to acknowledge its limitations. Data collected over a specific period may not be representative of the entire historical context or future developments. In Table 1, the absence of data for certain periods may limit the insights that can be drawn from the dataset.

Bhatt et al. [10] investigated technology innovation by analyzing patent citations using the global standard main path and global key-route main path analyses. The study identified five major technological clusters in the evolution of blockchain technology, which led to the identification of the evolution phases of this disruptive technology. By considering the evolution trajectory of blockchain technology, the research revealed the path dependence and knowledge flow of disruptive innovation, which helps decision-makers understand the landscape of disruptive technologies.

de Moura, et al. [11] conducted a comprehensive study to analyze global technological advancements in Direct Seawater Electrolysis (DSE) technology. The research aimed to provide a systematic overview of technological progress and identify potential opportunities for development by utilizing patent data from the Derwent Innovations Index (DII) database (1963–2022). Key findings revealed that China, South Korea, Japan, the US, and the UK are leading players in DSE technology advancement. The analyzed patents primarily focused on the evolution of various electrolytic systems and electrocatalysts, showcasing a continuous drive toward green hydrogen production through DSE. The study employed patent mapping to visualize the intricate network of research and development efforts, highlighting the critical role of DSE technology in addressing pressing concerns like freshwater scarcity and carbon reduction initiatives. Furthermore, the identified growth trajectory within the DSE patent landscape underscores the emerging opportunities for innovation and commercialization in this promising field of green hydrogen production.

Sampaio et al. [12] conducted patent analytics to identify the development of photovoltaic cell technologies. Their study found that the number of patents filed for photovoltaic cells has increased annually, with China, Japan, and the United States leading the way. The study also identified the main areas of knowledge and technology concentration in photovoltaics. This information can be used to monitor the development of photovoltaic technologies and identify potential opportunities for innovation.

Su et al. [13] identified eight critical areas of 5G-V2X technology by conducting a comprehensive search of patent data. They then evaluated the future applications of 5G-V2X in autonomous driving using technology predictions, patent trend analytics, country analytics, patent assignee analytics, IPC network analytics, and growth modeling.

Wang, et al. [14] proposed a patent analytics framework that integrates three data-driven approaches: Derwent Manual Codes co-occurrence network, Latent Dirichlet Allocation (LDA) topic modeling, and link prediction from multiple perspectives and layers. This framework captures the dynamics of AI and wind power technology knowledge interaction, the interaction evolution of technological topics, and the future development direction.

Ghaffari, et al. [15] used a search string and IPC codes to extract data from tire patent certificates over 20 years. They then used LDA to identify relevant technology areas and measured the compound annual growth rate (CAGR) of patent shares within these areas. As a result, they predicted that airless and innovative tires would be future technology topics in the tire industry.

Ni, et al. [16] analyzed the patent information of traditional Chinese medicine (TCM) prescriptions for diabetes treatment and identified the medication rules. These results can be used as references for other domestic enterprises in R&D and subsequent patent applications of TCM formulations. The paper also compares drug discovery and patent layout differences among three domestic pharmaceutical enterprises. Based on the patent analytics results, SWOT analytics was conducted to summarize and analyze the strengths, weaknesses, opportunities, and threats of TCM.

Mastilović, et al. [17] analyzed 881 patents to identify emerging sensor trends in the postharvest of fresh produce. They used a multistep search and refinement approach to generate a patent portfolio and then employed Latent Dirichlet Allocation (LDA) to model and interpret the unstructured patent database. As a result, they identified several potential technology opportunities, including developing new sensor technologies for monitoring the freshness, quality, and safety of fresh produce.

Liu, et al. [18] used LDA and community detection algorithms to identify the evolution and landscape of ARD green technology in China and the US This paper compares ARD green technology capabilities between the two countries, focusing on development priorities and strategic layout.

Li, et al. [19] delve into patent data to analyze offshore LNG storage and transportation’s core technologies and development trajectory. Their study sheds light on historical patenting activities, prominent patent holders and countries, and key terminology clusters. By identifying areas with high patent density and technological advancements, they indirectly pointed to potential opportunities for innovation in this field.

Jin, et al. [20] conduct patent classifications to uncover hotspots and trends in biological water treatment. Their analytics uses temporal and spatial distribution patterns and keyword frequency to reveal two prevailing technologies, their evolutionary paths, and potential application areas, which unveils promising avenues for further research and development within biological water treatment technologies.

Rezaei and Kamali [21] evaluate the technological collaboration between Silicon Fen (SF) companies and the University of Cambridge (UoC) through joint patent analytics. They calculate a “technological collaboration strength” (TCS) metric based on co-authored patents, identifying Pharma/Biotech as the sector with the highest TCS and most joint patents. They further highlight the role of spin-off companies from UoC in acting as bridges for knowledge transfer, suggesting potential opportunities for further collaboration and technology commercialization in this sector.

A review of prior studies confirms that patent analytics in various fields has identified technology opportunities. The rich technical details within patent data enable effective trend forecasting and identification of key technological components. Patent analytics is, therefore, widely used in industrial research to predict technology trends and assess patentability. Within this context, existing studies have drawn on detailed technical areas to identify technology opportunities and monitor technology advances, even identifying potential innovation areas. However, despite the growing importance of the underground logistics system (ULS) field, more comprehensive studies need to be conducted leveraging patent data in this relevant technical field. To address this gap, we extract and classify the detailed technical fields within ULS patents and identify the major patent applicants and technology opportunities with high technical value.

3. Methodology

3.1. Research Questions and Data Extraction

In this study, detailed technology fields are interpreted using detailed technology topics classified through topic modeling of the extracted patent data. Through portfolio analytics, we derive the topics with high technological value compared to other topics and the companies and research institutes that lead patent applications in those topics. It then identifies the technology growth potential of technology fields through a technology life cycle analytics of the topic with high technological value. To conduct the study, three research questions are set, and various analytical methodologies are applied. A summary of the relevant analytical methodologies based on the three research questions is shown in

Table 2. Research Questions and Analytics Methodologies.

Research Question	Description	Methodology
Q1	What are the detailed technology areas within the underground logistics system (ULS)?	TF-IDF, Topic Modeling, N-gram
Q2	Which companies and research institutes lead patent applications for detailed technology areas and high-value technology areas that have a higher technology value than other areas?	Scientometrics, Portfolio Analytics, Bibliographic
Q3	What is the potential for technological growth in specific high-value areas within the underground logistics system?	Technology Life Cycle

. Figure 1 visualizes and presents a research flow chart based on the analytics methodology.

To extract related patent data, all IPCs (International Patent Classification) related to logistics technology were extracted, and patents that included technology keywords related to underground logistics systems were extracted. The IPCs and technology keywords for extraction were searched and collected using Bard and Google’s large language model (LLM) based on the review of logistics experts. The results are shown in

Table 3. Results of Extracting IPC and Technology Keywords Related to Underground Logistics Systems.

IPC or Technology Keyword	Description
Technology Keyword	Drilling, Mining, Tunneling (Tube), Underground
B60B	VEHICLE WHEELS; CASTORS; AXLES FOR WHEELS OR CASTORS; INCREASING WHEEL ADHESION
B60P	VEHICLES ADAPTED FOR LOAD TRANSPORTATION OR TO TRANSPORT, TO CARRY, OR TO COMPRISE SPECIAL LOADS OR OBJECTS
B65B	MACHINES, APPARATUS OR DEVICES FOR, OR METHODS OF, PACKAGING ARTICLES OR MATERIALS; UNPACKING
B65C	LABELLING OR TAGGING MACHINES, APPARATUS, OR PROCESSES
B65D	CONTAINERS FOR STORAGE OR TRANSPORT OF ARTICLES OR MATERIALS, e.g., BAGS, BARRELS, BOTTLES, BOXES, CANS, CARTONS, CRATES, DRUMS, JARS, TANKS, HOPPERS, FORWARDING CONTAINERS; ACCESSORIES, CLOSURES, OR FITTINGS THEREFOR; PACKAGING ELEMENTS; PACKAGES
B65G	TRANSPORT OR STORAGE DEVICES, e.g., CONVEYORS FOR LOADING OR TIPPING, SHOP CONVEYOR SYSTEMS OR PNEUMATIC TUBE CONVEYORS
B65H	HANDLING THIN OR FILAMENTARY MATERIAL, e.g., SHEETS, WEBS, CABLES
B66	HOISTING; LIFTING; HAULING
G06Q	DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTING PURPOSES, NOT OTHERWISE PROVIDED FOR [2006.01]
G07C	TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
H04L	TRANSMISSION OF DIGITAL INFORMATION, e.g., TELEGRAPHIC COMMUNICATION
H04W	WIRELESS COMMUNICATION NETWORKS

.

We utilized Google’s patent database. First, we conducted a comprehensive International Patent Classification (IPC) search that yielded 189,034 patents. Then, we employed a Document-Term Matrix (DTM) built from technology keywords to select 2022 patents for further analytics.

Table 4. Detailed Information about the Selected Patent Data.

	Information
Identified Region	AR, AT, AU, BE, BR, CA, CH, CL, CN, CO, CZ, DE, DK, EA, ES, FR, GB, GR, HK, HU, IE, IT, JP, KR, LU, MX, NL, NO, PL, RU, SE, SU, UA, US, WO, YU, ZA
Period	1909∼2023 (Application Year)
Database	Google Patent DB
Number of Selected Data	2022

provides detailed information on the selected patent data.

Table 4 shows the extraction process and extraction ratio for the Selected Data.

Table 5. Extracted Patent Data Details from Raw Data.

IPC	Number of Raw Data (%)	ULS Technology Keyword Data (%)	Data Selection Rate
B60B	19,282 (10.2%)	140 (6.9%)	0.726%
B60P	20,227 (10.7%)	136 (6.7%)	0.672%
B65B	11,910 (6.3%)	270 (13.4%)	2.267%
B65C	11,154 (5.9%)	268 (13.3%)	2.403%
B65D	9074 (4.8%)	294 (114%)	3.240%
B65G	8129 (12%)	283 (14.0%)	3.481%
B65H	5860 (3.1%)	61 (3.0%)	1.041%
B66	29,301 (15.5%)	122 (6.0%)	0.416%
G06Q	21,544 (11.4%)	112 (5.5%)	0.520%
G07C	18,526 (9.8%)	100 (4.9%)	0.540%
H04L	17,392 (9.2%)	106 (5.2%)	0.609%
H04W	16,635 (8.8%)	130 (6.4%)	0.781%
Total	189,034 (100%)	2022 (100%)	1.06%

provides details on the raw data, including the number and ratio of extracted IPC presented in Table 3, the number and ratio of patent data extraction by patent technology keywords (‘Drilling’, ‘Mining’, ‘Tunneling (Tube)’, and ‘Underground’), and the data selection rate compared to raw data.

The percentage of data selected varies significantly among the different IPC codes. For example, IPC codes B65B, B65C, B65D, and B65G have a higher selection rate (more than 2%) compared to codes such as B60B, B60P, and B66 (less than 1%). However, a low selection rate does not necessarily mean a less efficient extraction process. For instance, IPC code B66 has a high raw data count (15.5%) but a low selection rate (0.416%). This may be because B66 represents a broader category covering various lifting and transport technologies, and not all technologies are relevant to underground logistics.

In addition, some of the patent data, including the thoroughly investigated International Patent Classification (IPC) codes related to underground logistics presented in Table 4, were filed in the early 20th century. However, it is questionable whether ULS was considered in the early 20th century and whether the technologies mentioned in patents over 100 years ago are likely to be significantly different from modern times. Therefore, in this study, we attempted to verify whether controversial patents are related to technologies that can affect ULS development through a discussion on whether to use patent data from that era for analytics. To this end, in

Table 6. Top 5 Oldest Selected Patents: Technological Evolution and Historical Connections to Underground Logistics Systems.

Patent Number	IPC	Application Year	Title
US1076737A	B65G	1909	Pneumatic-dispatch-tube system
DE387804C	B66B	1923	Equipment for loading the vessels and processing of the conveyed material underground
DE507903C	B65G	1927	Conveyor belt for underground operation in mines with an arrangement of the drive in the lower belt run between the end reversing rollers
DE497092C	B65G	1928	Conveyor belt for the simultaneous conveyance of goods in both directions, especially for underground operations
DE507904C	B65G	1928	Conveyor belt that can be laid in horizontal curves for underground operation

, we explored technological evolution and historical connections to the top oldest 5 patents among the selection data presented in Table 4 and included them in the analytics to prevent artificial data removal. We have summarized the relevance of the patents presented in Table 6 through

Table 7. Summary of Patent Relationships to ULS.

Patent Number	ULS Relevance
US1076737A	Pneumatic tubes are a basic technology for transporting goods through enclosed tubes. Modern ULS might utilize similar concepts for specific applications (Possibly Relevant).
DE387804C	This patent deals with underground material handling, a core function of ULS (Likely Relevant).
DE507903C	Conveyor belts are a fundamental technology for ULS, especially for bulk material transport (Possibly Relevant).
DE497092C	Bidirectional conveyor belts are useful for ULS where efficient movement in both directions is required (Possibly Relevant).
DE507904C	Flexible conveyor belts that can navigate curves could be valuable for complex ULS networks (Possibly Relevant).

.

Although the patent technology presented may differ somewhat from modern ULS, it can be confirmed that concepts such as material handling, conveyor belts, and directional movement are still connected to IPCs related to current ULS technology. Patents have been filed in Germany and the United States, and the relevant IPCs consist of ‘B65G’ and ‘B66B’. These patents show that they were developed with an emphasis on how to effectively treat mineral treatment, transportation systems, and transportation systems in mines for underground excavation. In other words, they have been extended from mineral transport to human transport, and it is thought that the evolution of hyperloop technology in pneumatic systems is taking place.

3.2. Detailed Technology Classification of the Underground Logistics System Field through Topic Modeling

To facilitate the classification of detailed technologies within the domain of underground logistics systems, a multi-pronged approach is employed. Use LDA (Latent Dirichlet Allocation)-based topic modeling to identify technology-related topics within individual documents. This discrimination contributes to a more nuanced understanding of the technological landscape [14,15,17,18].

Subsequently, an N-gram language model is applied to interpret the identified technology topics, enabling an objective interpretation of detailed technology areas. LDA topic modeling posits that each document can be represented as a blend of topics, where each topic is characterized by a distinct distribution of words [15,22,23,24,25,26,27,28]. The foundation for LDA topic modeling is the Document-Term Matrix (DTM), a matrix that encapsulates the frequency of each word within a document [29,30,31,32,33,34,35,36].

To construct the Document-Term Matrix (DTM), we employed a keyword extraction technique based on Term Frequency-Inverse Document Frequency (TF-IDF). This method assigns greater weights to terms that hold significant relevance within specific documents, effectively distinguishing them from common words that appear frequently across a broader corpus of text [37,38,39,40,41]. The mathematical foundation of the TF-IDF is outlined in Equations (1)–(3).

$$tf\left(t,d\right)=0.5+{\displaystyle \frac{0.5\times f\left(t,d\right)}{\max \left\{f\left(w,d\right):w\in d\right\} }}$$

(1)

$$idf\left(t,D\right)=log{\displaystyle \frac{\left|D\right|}{\left|\left\{dϵD:tϵd\right\}\right|}}$$

(2)

$$tfidf\left(t,d,D\right)=tf\left(t,d\right)\times idf\left(t,D\right)$$

(3)

Leveraging the constructed Document-Term Matrix (DTM), we implemented Latent Dirichlet Allocation (LDA) topic modeling to cluster documents into distinct technology topics effectively. To identify the optimal number of topics (K), we utilized perplexity, a metric that gauges a model’s ability to accurately predict words within a given document [22,42,43,44,45,46,47]. Lower scores on this measure signal stronger model performance. The mathematical formulation of perplexity is presented in Equations (4) and (5).

$$perplexity=\exp \{-{\displaystyle \frac{{{\displaystyle \sum }}_{d=1}^{M}logp\left(W_d\right)}{{{\displaystyle \sum }}_{d=1}^M{N}_d}}\}$$

(4)

$$p\left(W_d\right)={{\displaystyle \sum }}_{t=1}^{K}p(t_d)p\left(W_t\right)$$

(5)

In this study, we identify discriminatory technology keywords by deriving unique keywords corresponding to the lowest log ratios of beta values for each technology keyword between a specific topic and other topics. The two keywords with the lowest log ratios are the most discriminative. To derive discriminatory technology keywords, we utilize the tidytext package in R (R-4.4.1), which offers a method for extracting topic-specific word probabilities, referred to as “beta” values. Let us suppose that we have extracted beta values for a specific topic and the average beta values for other topics using the tidytext package, as displayed in

Table 8. Utilizing the Tidytext Package to Extract Beta Values for Discriminatory Technology Keywords.

Technology Keyword	Specific Topic	Average Value of Other Topics	Log Ratio
administration	0.000431	0.00138	1.68
ago	0.00107	0.000842	−0.339
agreement	0.000671	0.00104	0.630
aid	0.0000476	0.00105	136
air	0.00214	0.000297	−2.85
american	0.00203	0.00168	−0.270
analysts	0.00109	0.000000578	−10.9
area	0.00137	0.000231	−2.57
army	0.000262	0.00105	2.00
asked	0.000189	0.00156	3.05

.

For instance, the term “analysts” has a probability of 0.00109 being generated from a specific topic, while it has a probability of 0.000000578 being generated from the average of other topics. Calculating the log ratio of these two probabilities reveals that the probability for “analysts” is substantially higher in the specific topic compared to the average probability across other topics, resulting in a significantly lower negative value.

To objectively elucidate the specific technology areas within each topic, we leveraged N-gram analytics to extract tokens that encompass discriminative technology keywords. These keywords were identified from the LDA topic modeling results generated using the perplexity-based optimal number of topics (K). The N-gram model is used to extract consisting of meaningful word combinations from patent data into technology topics using LDA topic modeling. The N-gram model is a statistical language model (SLM) that predicts the next word in a sequence based on the previous N-1 words [48,49,50,51,52]. The probability that the word ‘$W_N$’ will follow from the previous words N-1 is called the ‘N-gram conditional probability’, and is mathematically expressed as below Equations (6)–(9) [53].

$$W_1^n=W_1\dots W_n$$

(6)

$$P\left(w_1^n\right)=P\left(w_1\right)P(w_{2 }|w_1)P(w_{3 }|w_1^2)\dots P\left(w_n|w_1^{n-1}\right)={\displaystyle \prod }_{k=1}^{n}P\left(w_k|w_1^{k-1}\right)$$

(7)

$$P\left(w_1^n\right)={\displaystyle \prod }_{k=1}^{n}P\left(w_k|w_{k-N+1}^{k-1}\right)$$

(8)

$$P\left(w_n|w_{n-N+1}^{n-1}\right)={\displaystyle \frac{C\left(w_{n-N+1}^{n-1}{w}_n\right)}{C\left(w_{n-N+1}^{n-1}\right)}}$$

(9)

The N-gram model is more accurate than traditional SLMs, which consider all words in a sentence because it focuses only on meaningful word combinations [37,54,55,56]. By extracting trigrams containing discriminative technology keywords identified through LDA topic modeling, we created meaningful phrases that more accurately represent the technological areas within each topic. This objective interpretation aids in understanding the specific technology domains covered by the patents. Trigrams allow the model to consider not only individual words but also their surrounding context, which is crucial in technical fields like underground logistics systems, where specific phrases may be more significant than individual words. For example, “tunnel boring machine” is more informative than “tunnel”, “boring”, and “machine” alone.

In this research, trigrams serve as concise summaries of technology topics. Each trigram is a short phrase that encapsulates the essence of a specific technology area. For instance, if the topic is “tunnel construction methods”, the trigram “tunnel boring machine” accurately summarizes the key technology. Analyzing the collection of trigrams associated with each topic allows researchers to gain a deeper understanding of the specific technologies discussed in the patents, leading to a more organized and comprehensive classification of the underground logistics system field.

3.3. Portfolio Analytics Identifies Applicants Leading Patent Applications in Areas of High Technology Value

To derive the technology topics most relevant to high-value technology, we use the five indicators previously reviewed through portfolio analytics to derive technology topics related to high-value technology. Portfolio analytics is a powerful tool that can be used to leverage patent and patent data to derive technology topics most relevant to high-value technologies [57,58].

In this study, we derived five indicators for each technology topic using patent and patent data for portfolio analytics. Pentagon visualization is a tool used to illustrate the results of portfolio analytics. A pentagon visualization displays each technology topic as a pentagon represented by five metrics. The size of each pentagon represents the total score for that technology topic [59,60]. Each metric is evaluated on a one-point scale per metric through Min–Max normalization, with a total score of 5 for each technology topic. An example of pentagon visualization is shown in Figure 2.

The higher the total score, the more relevant the technology topic is to the high-value technology. Portfolio analytics makes it easy to identify the technology topics most relevant to high-value technologies and effectively assess the potential of technology topics. To derive the major indicators that measure the value of technology, an indicator for quantitatively measuring and analyzing the bibliographic information of patent data is devised based on a literature review. As a result of the review, five devised indicators will be used for analytics. The index consists of five indicators: (1) Technology Applicability (TAP), (2) Growth Rate (Cagr), (3) Market Share (SHARE), (4) Innovation Facilitation (Time), and (5) Activity (AI). Each indicator is derived by using patent data belonging to each technology topic corresponding to the detailed technology area classified through LDA topic modeling.

3.3.1. Technology Applicability (TAP)

TAP (Technology Applicability) signifies the breadth of applicability of patent data associated with a specific technology keyword (Term). This indicator draws upon the Hirschman-Herfindahl Index (HHI) theory, commonly used to assess market monopoly levels [61]. The HHI is a tool for analyzing market concentration [62]. As the number of dominant players in a market increases, the index value decreases, reflecting a more competitive landscape. Conversely, a market controlled by a few entities leads to a higher HHI, indicating reduced competition.

Instead of directly measuring market share, this study utilizes an alternative approach based on the distribution of technology keywords within patent data documents. We quantify the diversity of keywords within each technology topic through the reciprocal of the Herfindahl-Hirschman Index (HHI). A higher Diversification Index signifies a wider and more even distribution of keywords across documents, suggesting technology’s broader potential applicability.

To assess the versatility of identified topics, we analyze the frequency of keywords per document for each topic using its corresponding Document-Term Matrix (DTM) derived from the original patent. For a finer level of analytics, Equations (10)–(14), based on Figure 3, calculate the TAP indices, which explicitly quantify the potential application breadth of each topic. Figure 3 shows a visualization of the Document-Term Matrix (DTM) derived from patent documents and technology keywords belonging to a specific technology topic. These equations leverage the concept of the HHI to capture the degree of keyword dispersion across diverse application areas, with broader dispersal implying more significant potential usage.

$$TA{E}_m={{\displaystyle \sum }}_{i=1}^i{A}_{m, i}$$

(10)

In Equation (10), $TA{E}_m$ represents the total frequency sum of a specific technology keyword to which a patent document belongs to. $TA{E}_1$ represents the total frequency sum of a technology keyword corresponding to $P{A}_1$ that a patent document ($T_{1~i}$) belongs to.

$$TAS={{\displaystyle \sum }}_{m=1}^{m}TA{E}_m$$

(11)

Within the framework of Equation (11), TAS serves as a quantitative measure of the collective frequency of technology keywords associated with a given patent document. Specifically, it denotes the cumulative sum of individual technology keyword frequencies, spanning from $TA{E}_1$ to $TA{E}_m$, as aligned with their respective technology keywords from $P{A}_1$ to $P{A}_m$.

$$s_m={\displaystyle \frac{TA{E}_m}{TAS}}$$

(12)

Within the framework of Equation (12), the value of $s_m$ is determined by dividing $TA{E}_m$, representing the aggregate of keyword frequencies for a designated technology keyword, by TAS. To illustrate, if TAS holds a value of 100, and $TA{E}_1$, specifically associated with the technology keyword corresponding to $P{A}_1$, is 10, then the calculated value of $s_1$ would be 0.1.

$$HHI={{\displaystyle \sum }}_{m=1}^m{s}_m^2$$

(13)

As a result, HHI is derived by squaring the values of $s_m$ through Equation (13). If there are 10 types of technology keywords and the calculated values of $s_1$ to $s_{10}$ are all 0.1, then the HHI value becomes 0.1, which is 10 times 0.01.

$$TAP={\displaystyle \frac{1}{HHI}}$$

(14)

In Equation (14), the TAP is determined by employing the reciprocal of the HHI value. This inverse relationship is employed because lower HHI values signify a more even distribution of keywords across diverse fields. To illustrate, consider a scenario analogous to the one presented in Equation (13). If the calculated value of $s_1$ is 0.55, and the calculated values of $s_2$ to $s_{10}$ are each 0.05, the resulting sum of 0.3025 and 0.0225 yields 0.325, surpassing the HHI index calculated in Equation (13).

3.3.2. Growth Rate (CAGR)

The growth Rate (Cagr) tracks the annualized rate of change in patent applications for a specific technology area, indicating its expansion or contraction. Through LDA topic modeling, we identify technology areas by topic by analyzing them in terms of the compound annual growth rate (CAGR) for each clustered technology topic. We used the compound annual growth rate (CAGR) to measure the change in frequency over time by subject [15,63].

Equation (15) is an equation for calculating the compound annual growth rate (CAGR). To calculate the CAGR, divide the number of documents in the last year (V final) of each area by the number of documents in the first year (V begin) of each area for the period, then raise the result to the power of the inverse of the number of years in the period (t) and subtract 1 [64,65].

$$CAGR=({\displaystyle \frac{V_{final}}{V_{begin}}}{)}^{1/t}-1$$

(15)

In performing our methodology, it is difficult to determine the exact frequency of patent data because there is an 18-month public protection period [66,67,68]. Therefore, 2022 and 2023 are excluded from the analytics period.

3.3.3. Market Share (SHARE)

Market Share (SHARE) reflects the relative dominance of a technology area within the intellectual property landscape by capturing its share of patents compared to all areas [63]. We leverage LDA topic modeling to identify these technology areas thematically and analyze their market share (SHARE) for each clustered topic. SHARE for a specific topic is calculated by dividing the sum of its patent application occurrences by the total number of patent applications. Equation (16) is an equation for calculating the market share (SHARE).

$$\mathrm{Share}={\displaystyle \frac{Number of applications in specific topic}{Total number of applications}}$$

(16)

3.3.4. Innovation Facilitation (TIME)

The Time Index can be used to evaluate the relative speed of innovation across different technology fields. Rapid technology development and commercialization, often facilitated by swift patent screening processes, generally indicate faster innovation compared to other areas [69,70,71,72].

Unlike traditional screening systems that prioritize applications based on application orders, some countries utilize a patent-priority screening system. This system prioritizes applications related to promising new technologies, accelerated development, and commercialization [73,74,75]. The Time Index analyzes the average time between patent application and application dates for each specific technology area, providing valuable insights into the relative pace of innovation.

3.3.5. Activity (AI)

To assess the level of activity or concentration of patenting in specific technological domains, the Activity Index (AI) is a valuable index In addition to patent analytics, AI indicators are widely used in other fields under Revolved Technological Advantage (RTA) indicators [76,77,78]. This index accomplishes this by measuring the proportion of patent applicants held within a given technology area regarding their overall patent holdings [79]. This information can be used to pinpoint areas where applicants are particularly active or focused, which can then guide strategic portfolio decisions.

The Activity Index (AI) can illuminate areas of strength for applicants by revealing those technological domains where they exhibit heightened activity or concentration. Conversely, areas with a low AI index may indicate potential areas of weakness for an applicant. This indicator is calculated by dividing the percentage of patents held by an applicant within a specific technology area by the overall percentage of patents across all areas. For portfolio analytics between topics belonging to each detailed technology topic, we derive the average of the index values based on the index values for each detailed technology area (topic) derived for each applicant. Equation (17) for calculating the activity (AI) is as follows:

$$AI={\displaystyle \frac{ \frac{ Applicant’s application count for a particular field.}{Application count for a particular field.}}{\frac{ Sum of applications from certain applicants.}{Application count overall.}}}$$

(17)

Lastly, we conduct a bibliographic analytics to explore the patent application trends for high-value detailed technology areas within the underground logistics system, as identified through previous portfolio analytics [80,81,82,83]. This analytics examines the selected topics’ annual patent application trends by country and identifies the leading patent applicants driving innovation [81,84,85,86,87]. Our analytics aims to identify both patent application trends by country and the major applicants in these high-value technical areas [88,89,90]. These findings offer valuable insights for future research efforts, aligning with the evolving technological landscape of the current innovation environment [86,91,92,93,94].

3.4. Evaluation of Technological Growth Potential for High-Value Areas through Technology Life Cycle Analytics

In this study, we leverage Latent Dirichlet Allocation (LDA) topic modeling to analyze the technology life cycles (TLCs) of pre-identified high-value areas. TLC theory, first proposed by Harvey [95], offers a framework for understanding technology evolution in four stages: emergence, growth, maturity, and obsolescence. In the emergence stage, novel technologies emerge, driven by originality and creativity. Patent applications focus on technology feasibility. In the growth stage, demand for the technology surges, necessitating commercial viability and protection. Patent applications encompass both technology and business aspects. The technology becomes established and reliable in the marketplace during the maturity stage. Stability, competitiveness, and commercial aspects are paramount, with balanced technology and business considerations in patent applications. In the obsolescence stage, market saturation leads to stagnant supply and demand. Technology stability remains crucial, but competitiveness needs to improve. Patent focus shifts toward legal protection, with fewer applications filed.

A well-established approach for analyzing technology life cycles is the framework proposed by Campbell [96], further refined by Mogee, et al. [97]. This framework leverages the market’s competitive landscape, specifically the number of active players, to identify the stages of technology development [98]. Figure 4 visually depicts this framework, highlighting the correlation between market competition and technology maturity.

This framework categorizes technologies into four quadrants based on patent activity and market saturation, aiming to identify a high-value topic’s growth potential [99,100,101,102]. Quadrant I (Emergence or Obsolescence Stage) represent emerging or obsolete technologies, while Quadrant IV (Growth or Maturity Stage) signifies those approaching growth or maturity. Quadrant III (Technology not basis of competition), characterized by many players and limited patent applications, indicates mature but non-competitive technologies. Finally, Quadrant II (High technological competition) represents mature, highly competitive technologies with fewer players.

We can assess both maturity and market growth potential by analyzing individual technology life cycles. Technology maturity is determined by the stage within its life cycle: emerging technologies are nascent, while mature technologies are established in the market. Market growth potential is also evaluated based on the life cycle stage. Emerging technologies exhibit high growth potential, while mature technologies experience slower growth. Through the proposed framework, growth potential can be objectively identified by considering the maturity and market potential of technology topics corresponding to the high-value technology areas identified [98,100,103].

4. Results

4.1. Derive Technology Topics of ULS through Topic Modeling

We extracted high-scoring TF-IDF technical keywords from 2022 patents. This allows us to focus on the most prominent technical keywords in the field of underground logistics systems. Among the 835 identified keywords, we selected those with TF-IDF scores exceeding 10.6, corresponding to the top 25% (Q3). The results of the descriptive statistics of the TF-IDF score are summarized in

Table 9. TF-IDF Analytics of Underground Logistics System Patents.

	TF-IDF Score
Max	151.4535
Q3	10.60657
Median	5.349762
Q1	3.546848
Min	1.066552

.

These 208 keywords were then used to construct document-term matrixes (DTMs) with 835 and 208 rows and columns, respectively. Such DTMs can be used to identify patterns and relationships between patents and technical keywords.

Table 10. High TF-IDF technical keywords derived from TF-IDF analytics.

Technical Keyword

TF-IDF Index

‘system’, ‘storag’, ‘method’, ‘mine’, ‘conveyor’, ‘devic’, ‘oper’, ‘coal’, ‘base’, ‘water’, ‘tank’, ‘monitor’, ‘belt’, ‘mud’, ‘pipe’, ‘manag’, ‘posit’, ‘irrigation’, ‘facil’, ‘chain’, ‘cabl’, ‘transport’, ‘gas’, ‘crop’, ‘control’, ‘network’, ‘convey’, ‘communic’, ‘wireless’, ‘equip’, ‘scraper’, ‘construct’, ‘especi’, ‘detect’, ‘line’, ‘intellig’, ‘pipelin’, ‘materi’, ‘autonomous’, ‘power’, ‘station’, ‘machin’, ‘vehicl’, ‘galleri’, ‘structur’, ‘space’, ‘particular’, ‘reservoir’, ‘safeti’, ‘process’, ‘liquid’, ‘distribut’, ‘work’, ‘urban’, ‘inform’, ‘data’, ‘cavern’, ‘tunnel’, ‘locat’, ‘park’, ‘store’, ‘drive’, ‘type’, ‘instal’, ‘evalu’, ‘risk’, ‘sensor’, ‘contain’, ‘logist’, ‘driving’, ‘area’, ‘caviti’, ‘fluid’, ‘transmiss’, ‘oil’, ‘predict’, ‘operations’, ‘warn’, ‘util’, ‘pollut’, ‘void’, ‘rock’, ‘product’, ‘medium’, ‘garag’, ‘environ’, ‘bottom’, ‘wall’, ‘engin’, ‘automat’, ‘mobil’, ‘load’, ‘build’, ‘roller’, ‘remot’, ‘assembl’, ‘inspect’, ‘seal’, ‘protect’, ‘repeater’, ‘lay’, ‘support’, ‘bunker’, ‘measur’, ‘earli’, ‘and’, ‘comprehens’, ‘applic’, ‘prevent’, ‘electr’, ‘personnel’, ‘assess’, ‘energi’, ‘pressur’, ‘spiral’, ‘optim’, ‘technolog’, ‘pneumat’, ‘model’, ‘integr’, ‘face’, ‘ore’, ‘map’, ‘provid’, ‘analysi’, ‘fault’, ‘combin’, ‘bit’, ‘vault’, ‘leakag’, ‘chute’, ‘excav’, ‘hydraul’, ‘survey’, ‘vessel’, ‘termin’, ‘robot’, ‘suitabl’, ‘uwb’, ‘hydrocarbon’, ‘realtim’, ‘time’, ‘determin’, ‘platform’, ‘ground’, ‘mechan’, ‘level’, ‘storage’, ‘transfer’, ‘air’, ‘realiti’, ‘lot’, ‘track’, ‘deliveri’, ‘link’, ‘identif’, ‘format’, ‘drill’, ‘bulk’, ‘section’, ‘radio’, ‘mainten’, ‘slide’, ‘charg’, ‘soil’, ‘wire’, ‘qualiti’, ‘chamber’, ‘mean’, ‘connect’, ‘signal’, ‘navig’, ‘servic’, ‘wind’, ‘flow’, ‘roadway’, ‘sump’, ‘discharg’, ‘acquisit’, ‘arrang’, ‘unload’, ‘emerg’, ‘command’, ‘make’, ‘improvement’, ‘calcul’, ‘run’, ‘compress’, ‘form’, ‘good’, ‘comput’, ‘thing’, ‘cover’, ‘move’, ‘doubl’, ‘bim’, ‘railway’, ‘car’, ‘threedimension’, ‘visual’, ‘fill’, ‘continu’, ‘guid’, ‘field’, ‘sens’, ‘suppli’, ‘lock’, ‘site’

X > 10.6

presents these 208 keywords.

To achieve high-quality clusters through optimal cluster (k) selection, determining the optimal number of clusters (k) for LDA topic modeling is crucial. Perplexity serves as a guide in this process. As the number of topics increases, perplexity decreases. This inverse relationship is illustrated in Figure 5. We identified the optimal k value as 6, corresponding to the point where the perplexity curve begins to plateau.

Following topic modeling analytics with an optimized k value of 6, we identified six distinct technology topics, each represented by descriptive keywords with the lowest log ratios. These findings are summarized in

Table 11. Derivation of the Negative Beta Log Ratio of Technology Keywords by Topics.

Topic	T1	T2	T3	T4	T5	T6
Max	−0.30	−0.30	−0.20	−0.08	−0.18	−0.35
Q3	−1.10	−0.89	−0.73	−0.49	−0.70	−1.59
Median	−2.33	−2.25	−1.83	−1.30	−1.56	−2.81
Q1	−5.14	−143	−100	−3.18	−144	−8.26
Min	−253.6	−831.9	−58.66	−4712	−61.70	−1048

.

We first clustered patent data to delineate technology topics using Latent Dirichlet Allocation (LDA) topic modeling. Subsequently, we applied N-gram model analytics to extract meaningful word combinations for each clustered topic. To interpret each technology topic, we select the three most frequently identified tokens using a trigram (3-gram) model. These tokens, containing discriminatory technology keywords, were extracted for each topic and are presented in

Table 12. Results of identifying technology topics based on the extracted tokens.

Topic	Discriminatory Technology Keyword	Selected 3 Identified Tokens	Patent Data (%)
Topic 1	drive, run	‘conveyor chain drive’, ‘chain drive mechanisms’, ‘lower belt run’	389 (17.7%)
Topic 2	operations, mud	‘underground mining operations’, ‘lot shutdown operations’, ‘mud risk evaluating’	315 (112%)
Topic 3	crop, irrigation	‘crop irrigation system’, ‘crop store consists’, ‘root crop store’	349 (15.8%)
Topic 4	autonomous, driving	‘autonomous federated learning’, ‘driving risk assessment’, ‘road driving device’	396 (18.0%)
Topic 5	bottom, void	‘cave depot bottom’, ‘void space occupied’, ‘bottom funnel bunker’	334 (15.2%)
Topic 6	repeater, wireless	‘network integrated repeater’, ‘relay wireless repeater’, ‘integrated repeater system’	419 (19.0%)

, defining the essence of each technology area.

Table 13. Patent Application Frequency from the Top 10 Countries.

Rank	Country	Abbreviation	Patent Data (%)
1	China	CN	1033 (46.9%)
2	Germany	DE	357 (16.2%)
3	South Korea	KR	266 (12.1%)
4	United States	US	259 (11.8%)
5	France	FR	47 (2.1%)
6	Russia	RU	30 (1.4%)
6	United Kingdom	GB	30 (1.4%)
7	Japan	JP	18 (0.8%)
7	Australia	AU	18 (0.8%)
8	Brazil	BR	10 (0.5%)

further explores the above patent data by analyzing the frequency of patent applications originating from the top 10 countries. Figure 6 visually depicts this distribution, highlighting a higher concentration of patent applications from China, Germany, Korea, and the United States. This confirms that applicants from these four countries largely drive patent activities in this field.

4.2. Portfolio Analytics to Derive Technology Topics of High Value in the ULS

To pinpoint the technology topic most closely aligned with high-value technologies, we conducted portfolio analytics utilizing five previously derived indicators for each topic. Visualized through pentagonal representations, this analytics enables the clear identification of the most relevant topics and a comprehensive evaluation of their potential. Figure 7 illustrates this visualization, where each technology topic is depicted as a pentagon, with its dimensions reflecting the five metrics assessed in portfolio analytics.

Table 14. Technology Value Ranking.

Topic	TAP	Cagr	Share	Time	AI	VR
T1	0.305	1	1	0.838	1	1
T2	0.756	0.312	0	1	0.494	3
T3	0	0.05	0.319	0.081	0	6
T4	1	0.921	0.787	0.662	0.742	2
T5	0.272	0.306	0.191	0	0.513	5
T6	0.088	0	0.723	0.706	0.696	4
Min	13.8	0.015	0.143	0.001664	0.69
Max	17.33	0.17	0.19	0.00191	1.21

complements this visual representation, presenting numerical scores for each technology topic based on the corresponding pentagon size. These scores signify the degree to which a topic aligns with high-value technologies, with higher scores indicating a stronger correlation. The table also includes a value ranking (VR) for each topic, providing a structured assessment of its relative potential.

Through the above analytics, we identify the technology topics with the first and second highest VR as detailed technology areas with high technology value. We evaluate the validity of Topic Description for the five patents judged to be the most important for each topic. The five patents presented in

Table 15. Results of Portfolio Analytics.

Topic Area	Description	Selected Patent (Patent Number)	Key Innovation
Underground Material Handling (Topic 1)	Systems for transporting materials within underground mines	Coal face underground mine conveyor chain drive assembly (DE19616458A1)	Improved design for conveyor chain drives used in underground mines
		A kind of underground coal mine belt-conveying dewatering system and belt conveyor (CN106219190A)	Integrated system for dewatering (removing water) from belt conveyors
		A kind of downhole intelligent belt conveyor system suitable for underground coal mine (CN109533865A)	Intelligent belt conveyor system with enhanced capabilities for underground mines
		Automatic belt frame extending device of coal mine tunnel conveyor and underground tunnel driving working face transportation system (CN115477125A)	Scalable material handling system for underground tunnels
		Method, system, equipment and medium for controlling underground coal mine belt conveyor system (CN114671213B)	Advanced control system for underground belt conveyors
Underground Transportation (Topic 4)	Systems for transporting personnel and equipment within underground mines	Positioning device and method for coal mine underground transportation robot (CN111417069A)	Uses Ultra-Wideband (UWB) technology for precise positioning of underground robots
		TOA underground person positioning system and method (CN104333905A)	Accurate personnel positioning system even in Non-Line-of-Sight (NLOS) situations
		Uwb technology-based attitude self-correcting underground transportation device (CN113766418B)	Self-correcting navigation system for underground transportation devices using UWB
		Detachable sharing express delivery device and sharing method for underground road or pipeline transportation (CN113989994B)	Express delivery system specifically designed for underground transportation networks
		Communication node arrangement method for underground rescue robot (CN106781370B)	Optimized placement of communication nodes for underground rescue robots to improve communication in challenging environments

were extracted by experts, and we did not provide information that hindered the validity derived in advance. The results of the feasibility evaluation for the Topic’s Description presented in Table 15 are presented in

Table 16. Topic Definition Validation through Patent Relevance Presentation based on Technical Tokens.

Patent Number (Topic)	Technical Effects (Token Relevance)
DE19616458A1(1)	It improves the underground coal mine conveyor chain drive assembly (‘conveyor chain drive’, ‘chain drive mechanisms’).
CN106219190A (1)	This technology offers a novel dewatering solution for underground coal mine conveyor systems. (‘conveyor chain drive’).
CN109533865A (1)	An intelligent belt conveyor system specifically designed for underground coal mines (‘conveyor chain drive’, ‘chain drive mechanisms’).
CN115477125A (1)	Automatic belt frame expansion technology streamlines transportation in coal mine tunnels and underground work surfaces. (‘conveyor chain drive’, ‘chain drive mechanisms’).
CN114671213B (1)	This comprehensive technology controls the entire underground coal mine conveyor system. (‘chain drive mechanisms’, ‘lower belt run’).
CN111417069A (4)	Improves the accuracy of positioning for underground transportation robots using UWB technology, leading to safer and more efficient navigation (‘autonomous federated learning’).
CN104333905A (4)	Provides a method for accurate underground personnel positioning, especially in situations where there’s no direct line of sight (NLOS), which is essential for worker safety and avoiding collisions (‘driving risk assessment’).
CN113766418B (4)	Utilizes UWB technology to allow underground transportation devices to self-correct their navigation, ensuring smoother and safer operation (‘autonomous federated learning’).
CN113989994B (4)	Introduces a new concept for an express delivery system designed specifically for underground transportation networks, featuring detachable and shareable components (‘road driving device’).
CN106781370B (4)	Addresses communication challenges in underground environments by proposing a method for optimal placement of communication nodes for rescue robots, enabling better emergency response (‘road driving device’).

.

The analytics of patent applications for underground logistics technologies reveals a fascinating shift. While early innovations in this field might be attributed to countries like Germany and the United States, as evidenced by patents like those in Table 6, the landscape is changing. Notably, a significant portion of recent patents originated in China (e.g., patents with CN prefixes in Table 16 and

Table 17. Top 10 Patent Applicants in Topic 1.

Rank	Applicant	Country	Patent Data (%)
1	Gutehoffnungshuette Sterkrade	DE	15 (3.9%)
2	China Coal Technology & Engineering Group	CN	12 (3.1%)
3	China University of Mining and Technology	CN	10 (2.6%)
4	Siemag Masch Stahlbau	DE	9 (2.3%)
5	Hauhinco Maschf	DE	8 (2.1%)
6	Certusview Technologies, LLC	US	6 (1.5%)
6	Dbt Gmbh	DE	6 (1.5%)
6	Demag Ag	DE	6 (1.5%)
7	Phillips Petroleum Co	US	5 (1.3%)
8	Chahoo	KR	4 (1.0%)

). This trend is further corroborated by the prominence of Chinese companies among the top applicants (Table 17). This suggests that China is emerging as a major player in driving advancements in underground logistics technologies.

We conduct bibliographic analytics to explore trends in patent applications for high-value-added detailed technical areas in underground logistics systems identified by earlier portfolio analytics. In the case of 2022 and 2023, only some of the patents filed in the last 18 months are disclosed; therefore, there is a limitation on the derived data because it needs to reflect the complete data [66,67,68].

Table 17 shows the top 10 patent applicants corresponding to Topic 1. The top 10 applicants, which consist of German, Chinese, US, and Korean companies, account for about 20.8% of all patent applications,

Table 18. Topic 1: Application Trends for Countries of the Top 10 Applicants.

Period	CN	DE	KR	US
–1988	0	42	0	25
1989–1993	0	0	0	9
1994–1998	0	4	0	6
1999–2003	1	2	1	2
2004–2008	0	1	1	4
2009–2013	2	1	9	3
2014–2018	43	0	11	8
2019–2023	146	0	11	7

shows the changes in the frequency of patent applications by period in the countries where the top 10 applicants identified in Table 17 belong. Figure 8 shows a visualization of the results in Table 18. Traditionally, patent applicants from Germany and the United States were the leading patent applicants. However, China’s patent applications have proliferated in recent years, particularly since 2014. In the case of patent applications in Korea, the frequency of patent applications is increasing less than that in China, but the frequency of patent applications is increasing.

Table 19. Top 10 Patent Applicants in Topic 4.

Rank	Applicant	Country	Patent Data (%)
1	Certusview Technologies, LLC	US	9 (2.3%)
1	Anhui University of Science and Technology	CN	9 (2.3%)
2	China University of Mining and Technology	CN	8 (2.0%)
3	Hu’Nan Guoao Electric Power Equipment Co., Ltd.	CN	6 (1.5%)
3	Dimine	CN	5 (1.3%)
4	Demag Ag	DE	4 (1.0%)
4	Dbt Gmbh	DE	4 (1.0%)
4	ChinaMTA	CN	3 (0.8%)
4	Smart Geotech	KR	3 (0.8%)
4	Siemag Masch Stahlbau	DE	3 (0.8%)

shows the top 10 patent applicants corresponding to Topic 4. The top 10 applicants, which consist of German, Chinese, US, and Korean companies, account for about 13.6% of all patent applications.

Table 20. Topic 4’s Application Trends for Countries of the Top 10 Applicants.

Period	CN	DE	KR	US
–1988	0	45	2	12
1989–1993	0	3	0	1
1994–1998	0	2	1	0
1999–2003	0	4	1	0
2004–2008	0	3	1	2
2009–2013	1	1	6	7
2014–2018	53	1	5	3
2019–2023	146	0	15	3

presents the trends in patent applications for topic 4 filed in four key countries, China, Germany, South Korea, and the United States, covering the period from 1988 to 2023. Figure 9 shows a visualization of the results in Table 20. China has experienced a remarkable surge in patent activity over the past 36 years. Although no applications originated from China in 1988, their number soared to 146 by 2023. Similarly, South Korea demonstrates a substantial increase, recording only two applications in 1988 compared to 15 in 2023. However, the previously dominant players, Germany and the United States, have witnessed a decline in patent applications over the same period. The results for topic 4 were similar to those for topic 1.

4.3. Identify Technology Opportunities in High-Value ULS through Technology Life Cycle Analytics

Technology opportunities are identified through data-driven analytics of patent activity for each technology topic. This analytics utilizes annual patent applications and application frequency to characterize the life cycle of each technology. We efficiently pinpoint promising opportunities by evaluating technology maturity and market growth potential based on these life cycles, delve into each identified technology topic’s patent data, and examine annual patent applicants and application frequency to construct a technology life cycle. Normalize the frequency when visualizing the technology lifecycle. This allows us to assess the technology’s maturity and market growth potential, thus pinpointing promising opportunities. The results of this life cycle analytics (TLC) for each technology topic (1, 4) are visualized in Figure 10 below.

Figure 10 reveals a shared technology life cycle trajectory for both topics, which is characteristic of the growth phase. This consistent curve pattern suggests a transition from the introductory phase to the growth phase since 2014. Table 19 and

Table 21. Proposed Technology Roadmap for Next-Generation Underground Logistics Systems.

Term	Proposed Technology Roadmap
Short-term (1–3 years)	Focus on the refinement and optimization of existing technologies within Topic 1 (Underground Material Handling System) and Topic 4 (Underground Transportation System). This includes advancing conveyor systems, optimizing chain drive mechanisms, developing more sophisticated autonomous driving capabilities, and enhancing driving risk assessment algorithms.
Medium-term (3–5 years)	Invest in research and development to accelerate the transition of Topic 1 and Topic 4 technologies from the growth phase to the maturity phase. This involves fostering collaboration between industry and academia to address technical challenges, standardize solutions, and reduce costs.
Long-term (5–10 years)	Explore the potential of emerging technologies, such as artificial intelligence, blockchain, and robotics, to further enhance the efficiency, safety, and sustainability of ULS. Investigate the feasibility of integrating ULS with other urban infrastructure systems, such as energy grids and waste management systems, to create a more holistic and interconnected urban environment.

notably highlight 2014 as the year coinciding with a surge in Chinese patent applications. This strong correlation between Chinese patent activity and technology life cycle shifts indicates a strong link.

Looking ahead to potential market expansion, these technology areas be interpreted to reach the growth stage, spawning new markets, and amplifying competition through both technological advancements and price-driven pressures from technological pervasiveness. The accelerating market growth rate is expected to incentivize new entrants and intensify competition among existing players vying for market share expansion. This trend also implies a potential future shift in the market’s competitive landscape.

5. Conclusions and Discussions

5.1. Research Questions and Key Findings

In this study, we use patent data analytics to bridge the knowledge gap in ULS technology development and identify technological opportunities for future growth. To this end, it aims to achieve the research objectives based on the three main research questions presented earlier. Firstly, we extracted patents that include keywords related to ULS technology from patents that correspond to IPCs related to logistics. Based on the extracted data, we used topic modeling of patent data to identify and classify various technical components within the ULS. As a result of the modeling, six technology topics were identified. We then combined the identification of discriminatory technology keywords and N-gram models to derive various detailed technical areas related to the ULS.

Secondly, we identify companies and research leading the development of high-value ULS sub-technologies by using portfolio analytics that utilizes various indicators of the six identified detailed technology topics. As a result of the analytics, Topic 1 (Underground Material Handling System) and Topic 4 (Underground Transportation System) were identified as areas of high technical value. Representative detailed technology areas for Topic 1 include “conveyor chain drive”, “chain drive mechanisms”, and “lower belt run”. Representative detailed technology areas for Topic 4 include “autonomous federated learning”, “driving risk assessment”, and “road driving device.” These technology areas were found to have many patent applications from companies and research institutes in China, the United States, Korea, and Germany, compared to other countries. In addition, the two technology topics were initially patent applications centered in the United States and Germany. However, recently, the frequency of patent applications in China and Korea has increased rapidly. Also, major patent applicants were identified that were common to both topics.

Finally, we analyze the technology life cycle of the identified high-value ULS technology topics to assess their growth trajectory and potential impact on the future of the ULS. Our analytics reveal that these technologies are currently in the growth phase, which may explain the relatively low volume of patent applications observed earlier by significant players. Notably, this life cycle shift coincides with a surge in patent applications from numerous Chinese applicants. Based on this trend, we expect a further increase in patent applications from China shortly. Subsequently, as competition within the market intensifies, we anticipate a decline in patent applications from Chinese entities, marking a transition toward the market maturity phase. Based on our research findings, in Table 21, we propose the following technology roadmap for the next generation of ULS.

5.2. Key Contribution of Our Research

The main contribution of this study is that it identifies vital patent applicants and technological opportunities with high technical value topics by extracting and classifying technology topics within ULS patents. It also suggests that presenting a quantitative methodology for establishing R&D strategies and identifying technological opportunities based on derived technological opportunities is meaningful.

Through a review of previous studies, it was confirmed that technological opportunities can be identified through patent analytics in various fields. Technology topics were derived to identify technological opportunities and monitor technological progress, and potential innovation areas were identified. However, despite the increasing importance of the ULS (Underground Logistics System) field, comprehensive research using patent data in related technical fields needs to be improved.

To address this, this study used patent data analytics to systematically identify and classify technology topic components within the ULS, providing valuable insights into the field’s current state. As a result of portfolio analytics, high-value ULS technology topics were identified by revealing that Topic 1 (Underground Material Handling System) and Topic 4 (Underground Transportation System) are areas with high technological value. In addition, we identified leading companies/research institutions within high-value ULS topics 1 and 4.

Finally, we identified technological opportunities based on the technology life cycle by analyzing these high-value topics’ patent technology lifecycles, providing valuable predictions about future growth and the potential impact on ULS development. This study is expected to be a valuable tool for researchers, policymakers, and patent offices, and it is expected to provide valuable insights to policymakers and stakeholders involved in development and implementation by additionally exploring the potential economic and social impacts of ULS development. Through our study, LDA topic modeling analytics, among machine learning methodologies, was used to analyze vast amounts of patent text data and derive detailed topics for each technology.

Table 22. Contributions of Derived Topics (via LDA Topic Modeling Analytics) to ULS Development.

Contribution to ULS Development	Description
Identifying Key Technology Areas	The six topics we found represent different technological components of ULS. This gives developers a clear roadmap of the most important areas to focus on.
Prioritizing Research and Development	The analytics showed that Topic 1 (Underground Material Handling) and Topic 4 (Underground Transportation) are high-value areas. This suggests that investing in these areas could have a bigger impact on advancing ULS technology.
Understanding the Competitive Landscape	By looking at who is patenting in these areas (e.g., companies from China, the US, Korea), developers can understand who their main competitors are and where innovation is happening.
Predicting Future Trends	The technology life cycle analytics suggests that Topic 1 and Topic 4 are in the growth phase, meaning we can expect to see more development and innovation in these areas in the coming years.

presents a plan in which topics derived through the proposed analytics methodology can contribute to the development of the ULS.

5.3. Validation through Literature ReviewValidation through Literature Review

We verify the validity of the findings based on a review of the relevant prior literature that conducted research on underground logistics systems. Edelenbos, et al. [104] investigated underground space use in the Netherlands and explored its future potential from diverse perspectives. They analyzed the characteristics and potential challenges of subterranean space utilization in the country, alongside the driving forces behind underground construction. Their findings anticipated a growing demand for underground space driven by increasing availability, rising environmental pollution from traffic, and rapid economic expansion. Additionally, the study suggests that urban areas and those with high traffic density hold promising potential for underground cargo transportation systems, warehouses, and shopping centers due to their relatively high return on investment.

Bobylev [105] focused on harnessing urban underground space (UUS) for sustainable city development. He emphasized the importance of rational land use encompassing surface and underground areas for efficient UUS utilization. The paper outlines the characteristics of UUS and considerations for its development. It examines the current state of underground storage facilities, transportation systems, communication networks, and industrial facilities and draws comparisons with urban subways, roads, and utilities. Notably, the research identifies obstacles to UUS development under a master plan and proposes strategies for sustainable growth.

Chen, et al. [106] employed SWOT-PEST analytics to assess the implementation and future of ULS in China. The SWOT analytics provided a balanced evaluation of ULS implementation, while the PEST analytics explored critical environmental factors affecting its development, including political, economic, social, natural, and technological aspects. These analytics shed light on the environmental challenges and potential improvement strategies for ULS in China.

Qihu [107] reviewed China’s underground space, highlighting five major challenges and four promising development trends. The identified obstacles hindering efficient utilization include fragmented management systems, incomplete planning and regulations, disjointed solutions, and unclear policies.

The development trends discussed include integrated planning, construction, and management of urban underground spaces, exemplified by projects like the Shanghai Expo and southern Ningbo’s underground spaces. Additionally, the paper examines the concept of integrated transport hubs, drawing examples from Beijing’s subway system, logistics network, and underground water storage.

In their study, Makana, et al. [108] analyze the potential of Multi-Utility-Tunnels (MUTs) as a driver for utilizing urban underground space (UUS) in Birmingham’s Eastside area. They leverage the SUURE framework and combine the Analytic Hierarchy Process (AHP) with fuzzy logic methodologies to analyze the viability of MUT configurations from diverse perspectives. Notably, their findings reveal flush fitting MUT as the optimal solution for promoting sustainable utility placement.

Hunt, et al. [109] advocate for the multifaceted utilization of UUS, showcasing its potential in building or safeguarding sustainable and resilient cities. They define sustainability and present a framework for addressing key planning challenges, ultimately demonstrating the inherent benefits of UUS integration.

Nelson [110] emphasizes the significance of proactively integrating efficient underground space utilization into urban planning to prepare for crises like extreme events and cultivate a thriving urban environment. They champion the concept of urban infrastructure stewardship and promote a multifaceted approach to system resilience. Notably, their research leverages Performance Response Function (PRF) analytics for evaluating system resilience and highlights the critical role of a data-rich cyber environment in facilitating informed decision-making.

Our analytics show that Topic 1 (Underground Material Handling System) and Topic 4 (Underground Transportation System), as well as their sub-topics, are essential technologies for the potential of underground logistics transportation systems, warehouses, shopping centers, and multi-utility tunnels (MUTs), as emphasized in the studies. In addition, we believe that technological innovation in Topic 1 and Topic 4 is an important solution to address the obstacles to the development of ULS. For example, problems such as the high cost, time constraints, and fragmented management system of ULS can be solved by reducing the operating time and costs by improving an integrated operating and management system through technological innovation in the Underground Material Handling System. In addition, human resource problems such as lack of experience and expertise can be solved by promoting automation in the ULS field through technological innovation in the Underground Transportation System, such as autonomous driving technology, driving risk assessment, and the advancement of road driving devices, to prepare for potential risks that may occur in advance.

5.4. Validation through Policy Review

The development of underground logistics systems (ULS) is undeniably influenced by government policies around the world. A review of policy documents from various countries, including China, Switzerland, and Singapore, reveals a diverse range of implemented policies. These case studies highlight the importance of government support in fostering private sector participation. By providing legal and institutional frameworks, the public sector plays a critical role in opening and growing the ULS market.

China, for instance, has prioritized ULS within its national scientific and technological initiatives. This commitment is evident in the promotion of the ULS integration management and pilot application projects. Additionally, government organizations like the Natural Science Foundation actively support collaboration across various fields, including policy, economics, transportation, environment, and technological integration [111]. Such comprehensive efforts have significantly propelled China’s ULS development.

Similarly, the Swiss government actively promotes the Cargo Sous Terrain (CST) project, a plan to build an underground tunnel network for transporting goods using unmanned vehicles. This project reflects their strategy to optimize logistics in a mountainous and densely populated country [112,113,114]. The Swiss government’s commitment is further demonstrated by incorporating CST into the national transportation network plan and actively engaging in legislative activities to create a supportive legal framework [112,113,114].

Singapore also exemplifies proactive government involvement in ULS development. Their Inter-estate Goods Mover System (IGMS) utilizes an underground tunnel network for automated logistics operations. After completing the project’s economic feasibility analytics, the Singaporean government is currently reviewing relevant regulations to facilitate its implementation [114,115].

These case studies suggest that government policies play a significant role in fostering growth in the ULS technology market. It is believed that policies promoting private sector participation and technological innovation contribute to the growth trajectory observed in the high-value-added technology areas identified in our study. We compared the major contributions of the work with those already reported in the literature by presenting comparative analytics of the major contributions to the study of underground logistics systems (ULS) through

Table 23. Comparative Analytics of Key Contributions to Underground Logistics Systems (ULS) Research.

Research Area	Existing Literature	This Study’s Contributions	References
ULS Technologies	Focus on specific technologies (material handling, transportation) in isolation, limited integration of different ULS components.	Comprehensive analytics of interdependencies between ULS technologies, identification of key technologies driving ULS potential (material handling, transportation).	[1,116,117,118]
ULS Implementation	SWOT/PEST analytics for specific countries (China), case studies on individual projects (CST in Switzerland), limited focus on policy impacts beyond individual projects.	Cross-country policy review highlighting common themes (government support, private sector participation), analytics of policy impact on high-value-added technology areas.	[106,116,119,120]
ULS Challenges	Identification of general challenges (cost, management, regulation), limited discussion of technology’s role in overcoming challenges.	Specific focus on technological solutions for ULS challenges (automation for human resources, integrated systems for management), highlighting the transformative potential of technology.	[2,116,121,122]
ULS Potential	Emphasis on specific use cases (warehouses, transportation), limited exploration of broader urban development impacts.	Broader vision for ULS integration into sustainable city development, connection to concepts like Multi-Utility Tunnels (MUTs) and urban resilience.	[2,106,116,123,124]

, based on the comparisons presented in Section 5.3 and Section 5.4.

5.5. Limitations and Future Research Directions

In this study, we investigate Underground Logistics Systems (ULS) with a focus on identifying high-priority research areas for improving energy efficiency within ULS operations (Topics 1 and 4). However, a comprehensive understanding of the ULS’s environmental impact requires further exploration beyond operational efficiency.

While the identified high-priority topics (Topic 1 and 4) suggest a focus on potentially more energy-efficient ULS operations, further investigation is necessary to understand the impact of ULS on raw materials and their life cycle carbon footprint. Future research should explore how ULS technology development can be integrated with the use of low-carbon materials in construction or components to achieve a more comprehensive low-carbon approach.

In addition, the focus on operational efficiency within the ULS (Topic 4: Underground Transportation System) suggests the potential for optimizing movement patterns. This optimization could lead to reduced overall carbon emissions within the ULS itself. To gain a holistic understanding of low-carbon behavior in the supply chain, broader analytics encompassing the carbon footprint of all transportation stages is necessary. This analytics could involve studying how the ULS integrates with other transportation modes and their combined impact on emissions.

Future research could explore the potential economic and social impacts of ULS adoption, further contributing to well-informed decision-making. By overcoming the limitations presented earlier through follow-up studies, we intend to provide a more comprehensive picture of the environmental impact of the ULS and contribute to the development of true low-carbon supply chains.