Technical Trends and Competitive Situation in Respect of Metahuman—From Product Modules and Technical Topics to Patent Data

Abstract

As a form of technological integration, metahuman has a significant influence on sustainable production because of its consistent technological evolution. However, few studies have provided insights into the technical assessment of metahuman by employing patents. In this paper, patent analysis is conducted to identify technological trends and the competitive situation in respect of metahuman from a product modularity perspective. First, we identify 17 highly relevant metahuman keywords by combining a literature analysis and an expert interview method and identify 42,256 patents from the Derwent Innovation Index (DII), thus improving the accuracy and validity of the data collection process. Then, metahuman product modularity is implemented using the function-behavior-structure (FBS) model, and seven technical topics are extracted from patents via latent Dirichlet allocation (LDA). Lastly, the procedure for identifying technology areas in respect of metahuman is improved by applying an optimized method to establish the connecting paths of product modules, technical topics, and patent data. The analysis results show that the development of metahuman technology can be divided into three periods. Different patent priority countries have distinctive competitive advantages and characteristics at the product module level. The findings of this study are intended to aid R&D enterprises and the government in formulating sustainable decision-making and promoting the development of the metahuman industry.

1. Introduction

Metahuman is being integrated into various aspects of people’s daily lives. More than 28 metahumans participated in the Beijing 2022 Winter Olympics. These included virtual singer Tianyi Luo, who sang “Time to Shine”; a digital version of the freestyle skier Ailing Gu; an artificially intelligent AI sign language anchor who appeared for 24 h; a digital human from the Olympic public service ambassador REAI, etc. [1]. Along with the greater social impacts in respect of metahuman in various application scenarios, metahuman has demonstrated strong commercialization potential for sustainable development [2]. According to the ”Global Intelligent Virtual Assistant Market” report published by Research and Markets in 2022, the size of this market is expected to reach $45.1 billion in 2027, increasing at a compound annual growth rate of 34% [3]. Furthermore, the size of the global virtual activity market is expected to exceed approximately $474.6 billion in 2028. Thus, it is unsurprising that, in recent years, several famous funds, such as Sequoia Capital in the US, Hillhouse Capital in China, and Softbank Capital in Japan, have begun to invest in the metahuman industry [4]. These investments focus on hardware and IP incubation, as well as content production and operation, etc. [5].

Considering that creating metahuman requires integrating various technologies, the development of metahuman technology has become a major factor affecting the richness of application scenarios and industry development [6,7]. Thus, the analysis of metahuman technology is important for research and development (R&D) enterprises to formulate suitable strategies for technology iteration and innovation. The benign development of enterprises can drive economic growth, create tax revenue, and serve society. It is also essential for policymakers to put forward strategic guidelines to support the sustainable development of metahuman, thus promoting socially responsible research. In addition, in some scenarios, metahuman has been tasked with simulating human beings in medical experiments or performing difficult tasks in environments that are not conducive to human survival [8]. Such metahuman is of great medical and scientific value and can contribute to the sustainable development of human beings from the social attribute level.

Previous studies have mainly focused on qualitative research regarding the realization and development of metahuman technology from the macro-perspective [9,10]. There is still a lack of quantitative exploration and discovery from the micro-perspective, which makes it possible to further deepen the research between the theoretical analysis of technology and the implementation of technical practice. Therefore, in this research, we evaluate and investigate the technological convergence of metahuman, and specifically address the following issues:

• How do we identify metahuman product modules and relevant technology patents?
• Which countries are more actively adopting metahuman technology and developing more advanced technology?
• How is metahuman technology evolving over time?

To answer these questions, we adopted a novel systematic analysis framework that mapped product modules, technology topics, and patent data. Metahuman is an intelligent product with advanced technological integration, and intelligent product requires a series of processes, such as data collection, data parsing, data prediction, and data distribution [11]. Division of the component modules can successfully illustrate the information domain and physical domain characteristics of intelligent products so as to realize a comprehensive analysis [12]. Thus, this framework provides support for conducting quantitative research on metahuman technology at a more microscopic level, i.e., from a modular perspective.

To solve the first research question, we applied a function-behavior-structure (FBS) model, which clearly describes the product morphology matrix through a top-down solution approach [13]. Meanwhile, LDA topic modeling provides methodological support for the patent text mining in this research. We also identified patents that are closely related to metahuman by collecting and refining highly relevant keywords.

To solve questions two and three, we conducted an in-depth analysis of metahuman technology utilizing patent data. Patents are considered a trustworthy source of technological information [14]; they are easily accessible through commercial databases [15]. Furthermore, patent analysis was applied as a useful method with which to provide quantitative and objective details on current and emerging technological trends, evolution, and growth [16]. Thus, analyzing patents to determine technological development trends and the competitive situation can indicate a feasible way forward for the sustainable development of metahuman technology. Specifically, we explored the development trends of metahuman over the decades through patent measurement analysis, identified the technology layout of typical priority countries, and explored the technology turnover and evolution based on technology lifecycle theory.

The rest of the paper is organized as follows. The Literature Review section summarizes existing studies on metahuman, such as typical methods to divide product modularity, and identifying technical topics based on patents. The research methodology is outlined and explained in the Materials and Methods section, in which the data collection procedure, FBS model for product modularity, LDA method for topic mining, and links between product modules, technical topics, and patent data are introduced in detail. Patent analysis results are provided in the Results Section, in which we analyze the evolution process, development trends, and competitive situation of metahuman technology from the dimensions of time, geography, and level. The Conclusions Section discusses how our findings relate to previous studies and summarizes the implications of this work. We also describe how we might address current limitations and pursue future work.

2. Literature Review

2.1. Metahuman

The Visible Human Project (VHP), started by the American National Library of Medicine in 1989, is where the word “metahuman” first appeared. The metahuman is a “human” that exists in the digital world. It is usually made highly similar to real humans through motion capture, 3D modeling, voice synthesis, and other technologies. At the same time, a three-dimensional “human” is presented with the help of augmented reality (AR)/mixed reality (MR)/virtual reality (VR), etc. At present, scholars define the term “metahuman” mainly from a technological or functional perspective. The metahuman is a computer-generated character with a human appearance from a technical perspective [17,18]. From a functional perspective, it is a comprehensive virtual image with human-like characteristics, such as appearance, behavior, thought, or interaction ability. It is created using computer graphics, speech synthesis technology, deep learning, brain-like science, biotechnology, computational science, and other convergent technologies or computer methods [19]. The metahuman is an individual with the ability to interact with artificial intelligence from a functional perspective [20]. It is also an individual with a realistic appearance who speaks naturally and expresses emotions more fully [21].

In this paper, “metahuman” refers to a virtual image that relies on artificial intelligence technology to imitate humans in appearance, behavior, thinking, and emotional expression [22]. To perform interactive duties, it specifically possesses human perceptual qualities like vision, hearing, and touch [23]. It generally has three characteristics: Specific human characteristics, such as appearance, sex, personality, etc.; human behavior that can be conveyed through words, expressions, and movements; and recognition of the outside world and interaction with others using human intellect [24].

2.2. Division Rules of Product Modularity

Many products are made up of multiple product modules, especially complex products, which comprise numerous functional modules by combining various construction blocks or components [25], such as personal computers. Modular design is an engineering methodology that can organize and assemble a set of independent components for complex products [26]. From the perspective of system engineering, the architecture of a complex product can be seen in both the design and specification of its components [27]. Grouping a product’s components into modules is a practical way to develop a modular architecture [28]. Modular design is an effective way to define and build a metahuman’s inborn functional components. We systematically sorted out the typical methods of product modularization, as shown in

Table 1. Summary of division methods and standards for product modularization.

Modular Rules	Theory	Methods	References
Function features	Functional flow of product	Formal functional decomposition and heuristic methods	Stone et al. (2000) [29]
Function features	Functional flow of product	Quality function deployment (QFD) and mapping matrix	Li et al. (2012) [30]
Function features	Functional flow of product	Axiomatic design; relevance matrix; clustering	Gu et al. (2014) [31]
Function and structure features	Functional flow of product	Fuzzy consistent comparison matrix	Xiao et al. (2015) [32]
Function and structure features	Lifecycle perspective	Correlation matrix of product components	Gu et al. (1999) [33]
Function and structure features and developing a strategy	Developing strategy for modularity	Adding strategies into integrated product module creation	Asan et al. (2004) [34]
Function-behavior-structure relevance	System theory	Design structure matrix (DSM)	Nie (2013) [35]

.

There are three kinds of modular rules: Function features, function and structure features, and function-behavior-structure relevance. The primary aspect supporting modularity at an earlier stage is the product’s function features, and several methods have been developed to support the development process. The heuristic method was developed to identify the product architecture [29]. The quality-function-development (QFD) method is used to design processes for developing the product architecture [30]. An analysis framework of product modularity division using axiomatic design was proposed by constructing an incidence matrix and clustering [31].

The process of breaking down product modules soon incorporated both structure and function features. The characteristics of the service fuzzy consistency judgment matrix were considered to propose service functional modules [32]. To execute a modular design technique, a correlation matrix of product components was created, with the product lifecycle serving as the foundation for the theory [33]. Strategy issues were later integrated into the process of creating the product module [34].

The product modularization process has been further refined. From the system theory perspective, the function-behavior-structure (FBS) model was proposed, in which the function features, behavior features, and structure features were all taken into account simultaneously [35]. The fundamental idea behind the FBS model is breaking down a function into sub-functions, concretizing the resulting abstract function into a specific behavior, and finally seeking a structural solution in order to carry out the behavior. In particular, the FBS model decomposes functions through top-down solving, which reduces the complexity of problem-solving and achieves product modularity decomposition from a more in-depth standpoint.

2.3. Technical Topic Identification of Patents

Patents are the wellspring of technology development and innovation [36]. Patent data can be used to identify technical subjects, which is one method of analyzing technological trends. Citation network clustering and text mining are the two main approaches for identifying technological topics based on patent data, as shown in

Table 2. Main methods of technical topic identification based on patent data.

Methodology		References	Advantages	Disadvantages
Citation network clustering	Direct citation	Zhang et al. (2016) [37]	Reveal relationships between individual patents	Diversification of citation motivation, literature with citation relationship may not have thematic similarity, lagging in publication
	Co-citation	Breitzman (2015) [38]
	Coupling	Li and Chen (2015) [39]
Text mining	Word frequency analysis	Zhang et al. (2018) [40]	Easy to operate, fine-grained, and objective analysis	Difficult to reflect inter-word associations, weak semantic relationships
	Co-word analysis	Choi et al. (2011) [41]	Objectively reveal the connections between technical topics	Difficult to deal with synonyms and polysemous words, lack of semantic information mining
	Topic model analysis	Li et al. (2022) [42]	Handles synonyms and polysemy issues well, identifies hidden topics well	Insufficient indication of important low-frequency words

.

Based on the citation linkages between patents, the citation network clustering methods combine patents with comparable content or subject matter to produce a technical topic. Direct citation networks, co-citation networks, and coupling networks are the three main categories of citation networks. The direct citation network method refers to identifying emerging technical topics based on patent direct citation network clustering analysis [37]. The co-citation network method clusters and evaluates the co-citation relationships among “hot patents” and extracts the set of technologies among them [38]. The coupling network method calculates the coupling strength between patents and performs a cluster analysis to identify the technical topics [39]. By examining the underlying logic of the citation network clustering method, we found that it can reflect dynamic changes in technology, but it exhibits a time lag and is blind to the motivations behind citations.

The text mining method reveals semantic connotations and associations by deeply analyzing text content. This method includes word frequency analysis, co-word analysis, and topic model analysis. The word frequency analysis method illustrates technical topics and evolutionary trends by statistically analyzing the frequency of important words in the text [40]. While word frequency analysis is simple to use and produces precise objective analytical results, it struggles to capture cross-word relationships. The co-word analysis method, based on subject-action-object (SAO) structural-semantic analysis, detects new technologies through the functional properties of patented technologies [41]. The co-word analysis method can reveal the relationships between technical topics objectively. However, due to a lack of semantic information mining, dealing with synonyms and polysemy is challenging. Topic models, which mainly include the LDA model, CTM topic model, author-topic model, and vector space model, can extract comprehensible and useful structures by computing the probabilistic distribution of words in the text [43].

As one of the most famous topic model methods [44], LDA is a three-layer Bayesian document topic generation model for mining and selecting target document topics, and it was developed by Blei et al., (2003) [45]. Each document in a corpus is a random mixture of latent topics, and each latent topic is identified by a distribution of characteristic words, according to the model’s central tenet [46]. The effectiveness of text semantic processing and the precision of topic recognition results have been particularly enhanced by LDA, which has arisen along with the development of natural language processing, data mining, and deep learning technologies. LDA has been used to identify crucial technologies in emerging smart industries, such as robotics [47] and intelligent manufacturing equipment [48].

Compared to other methods of identifying technical topics from the literature review, the advantages of the topic model are that it can effectively handle synonyms and polysemy issues and recognize hidden topics. LDA topic modeling is more appropriate for the study of cutting-edge technology themes contained in patents since it is used to discover latent topic information in large document collections or corpora [49], particularly for the identification of critical technologies in emerging industries. From this perspective, LDA topic modeling performs better in terms of recognition than other topic modeling techniques [50]. This provides a valid reason for us to adopt the LDA method with factuality and objectivity to identify the technical topics in respect of metahuman. However, we cannot disregard the fact that the LDA approach still has significant drawbacks. It overemphasizes automated analysis acquisition through models when discovering technical topics, without integrating with the actual situation of the analyzed object. To overcome the limitations of the LDA method, we mapped metahuman product modules hierarchically based on the available information and expert experience, matched technical topics identified by our algorithm to the prior modules, and then conducted a further in-depth patent analysis.

3. Materials and Methods

3.1. Research Framework

In this research, we followed the analysis path of the product module, technology theme, and patent data based on the patent text as the data source, by taking “data-driven” as the guiding ideology. This research trajectory lays a solid foundation for later analysis of metahuman technology trends and the competition situation. Figure 1 depicts our research framework, which primarily consists of four components.

Data collection is a meticulous and complicated procedure. The authors of this paper read metahuman articles, books, and sources independently and compiled 59 unique metahuman keywords covering a broad spectrum. Combining frequency statistics and the knowledge of subject-matter experts, we compiled a list of 17 search terms that were closely related to metahuman. Then, we chose the Derwent Innovation Index (DII) as the database by forming a search formula with keywords to identify metahuman patents between 1966 to 2020. Our initial search resulted in 61,998 patents. We filtered out some of the data from DII to obtain more precise results. After filtering, we identified 42,256 patents directly related to metahuman as research samples. These filtered data served as the foundation for further research.

Simultaneously, we carried out modular identification of metahuman product. First, we identified standards and division techniques for product modularization. On this basis, we conducted a comparative analysis of different methods, which provided the basis for our choice of metahuman product modularization methods. Following comparative analysis, the FBS model considers the complexity of module decomposition from abstract to concrete from a systems theory perspective. Therefore, we followed the decomposition logic of the FBS model to confirm the product modular composition of metahuman, including three steps. First, we analyzed the functional characteristics of metahuman. Then, we identified the corresponding carrier for each sub-function. Finally, we constructed an appropriate structural layer to link behavior to a module.

The following stage involved identifying technical topics in respect of metahuman based on the collected data. We first outlined standard approaches to technical topic identification to ensure the research’s maximum degree of scientific rigor. After confirming the advantages and applicability of the LDA topic modeling for emerging technology topic identification based on patent data, we chose LDA to mine the potential metahuman topics. The specific steps included pre-processing, clustering analysis, extracting feature words, determining the optimal number of topics, and identifying the topics’ meanings. We also verified whether the contents of the patent matched the topic matter of the “label”. Then, technical topics and modules were aligned by combining patent knowledge with expert opinion. In this way, we confirmed the patent distribution of different metahuman modules, which served as the foundation for the following analysis.

Based on the matching results, we conducted an in-depth analysis of the development trends and competitive situation of metahuman from a modular perspective. The basic data sources for analysis were the annual patent applications, patent priority country data, and technological development trend data. We used the patent measurement method to analyze the metahuman technologies from three dimensions:

We holistically explored the technological development history and trends based on the patent application situation.

We identified the geographical layout of each module in major R&D countries.

We found the level of technological development and evolutionary trends of different modules.

3.2. Source and Filtering for Data

After analyzing more than 80 academic articles related to metahuman, a total of 59 keywords were extracted. We invited 5 AI technical experts to discuss these keywords. Finally, 17 keywords closely related to metahuman were selected, which are listed in

Table 3. Metahuman keywords.

Keyword	Query Reformulation
Metahuman	TS = ‘Metahuman’ OR ‘Digital Human’ OR ‘Virtual Human’ OR ‘Avatar’
Digital Human
Virtual Human
Avatar
AI Anchor	TS = ‘AI Anchor’ OR ‘Virtual Anchor’ OR ‘AI digital human’ OR ‘Virtual Idol’ OR ‘Virtual KOL’ OR ‘Virtual Actor’ OR ‘Virtual Host’ OR ‘Virtual Spokesperson’ OR ‘Virtual Brand Officer’ OR ‘Virtual Customer Service’ OR ‘Virtual Tour Guide’ OR ‘Virtual Narrator’
Virtual Anchor
AI Digital Human
Virtual Idol
Virtual KOL
Virtual Actor
Virtual Host
Virtual Spokesperson
Virtual Influencer
Virtual Brand Officer
Virtual Customer Service
Virtual Tour Guide
Virtual Narrator

.

We chose the Derwent Innovation Index (DII) as the search source. DII is bi-directionally linked to the Web of Science, so the basic research results and technical application results could be effectively matched. This ensures the comprehensiveness and reliability of the data [51]. Since time lag is inevitable when recording patents in the Derwent database, the data for 2021 and 2022 were not included. The patent search period was from 1 January 1966 to 31 December 2020. The patent search expression is shown in Table 3. A total of 61,998 patents related to metahuman were first found in the DII database. After de-weighting and manual screening, 42,256 patents were finally identified for building a metahuman patent corpus.

3.3. Metahuman Product Modularity

The FBS model provides us with a clear and convenient process for the step-by-step decomposition of a complex product like a metahuman. As the primary task of product modularization is to identify the similarity and correlation of thousands of different components by describing characteristics [52], function and structure features are the main dimensions in dividing product modularity. The structure describes what the product is and how it is constituted, while the function explains what it does and what it is for [53]. In this paper, a method for identifying metahuman modules from functional models is presented. The methodology realizes the modularization of metahuman, combining the process of functional interaction analysis, specific behavior description, and module formation, as shown in Figure 2.

Scholars have proposed that the overfull functional goal of the metahuman is to build a nearly realistic virtual image according to the input information of real humans [23]. We first analyzed the functional features of the metahuman by reviewing and summarizing relevant information from articles, reports, and books. “Human” is the key element of creating the metahuman. The sense of intimacy, care, and immersion brought to users by a high degree of anthropomorphism is the core motivation for most consumers to use the metahuman. The ability to provide users with a natural and realistic experience through expression and interaction has become an important criterion by which to determine whether metahuman can replace real people in various application scenarios. Overall, metahuman comprise a technology-integrated product that combines three major functional features: Personification, expressiveness, and interactivity. We provide detailed descriptions of these three characteristics in Section 2.1. The functional features of metahuman correspond to the product requirements.

The logic of constructing modules following the FBS model involves a continuous process of decomposing complexity from abstract to concrete. The key step involves decomposing the total function into sub-functions according to the principle of functional independence. The real human images and character styles of metahuman further characterize the “personification” function. Facial expression, phonetic representation, and body expression decompose the “expressiveness” function according to different parts of metahuman. “Interactivity” represents the level of interaction between the metahuman and the real world and includes content dialogue and limb reaction.

Functional elements such as performance, style, expression, speech, body, interaction, and reaction comprise both the overall functional decomposition of the metahuman and the abstraction of specific behaviors. We determined the carrier of the abstract sub-function, i.e., the behavior that materializes into a certain function. The sub-function of “personification” is mainly reflected by appearance. The final rendering effect of a metahuman is achieved by a series of processes such as image design; key point binding of the face and body; key point changes in the form, eyes, and gestures; and detail production. The sub-functions of “expressiveness” are implemented by corresponding to the different expressive parts of the behavior occurring in the metahuman. Facial muscle refinement is the behavioral carrier of facial expression. The intonation rhythm handle is the behavioral carrier of phonetic representation. Limb precut adjustment is the behavioral carrier of body expression. In practical applications, the metahuman can recognize the user’s demands and respond through behavior from a preset knowledge base, perceiving and identifying user intention, managing the conversation, and analyzing decisions. In this way, hierarchical mapping from the function element to the behavior element and structure element is completed, laying a foundation for the solution of the overall structural scheme.

After finding the carrier of the obtained divided sub-functions, we need to build a suitable structural layer implementation scheme to match the execution action to the executive module, in order to realize the functional principle of the behavior layer. The “Information Collection” module provides a basis for forming a metahuman. The “Modeling & Rendering” module helps to achieve highly reductive modeling from real humans to create a realistic image. The “Intelligence-Driven” module enables intelligent synthesis and motion capture migration, the latter of which has become the current mainstream motion production method for metahuman. The “Character Generation” module can generate corresponding character voices and matching character animations based on text. The “Synthesis & Display” module synthesizes voice and animation into video and then displays it to the user. The “Recognition & Perception” module provides solutions for the recognition of voice semantics, face, and motion. The “Analysis & Decision” module enables the metahuman to have interaction capabilities that recognize users’ intention and respond to users’ needs.

In summary, we created a generic set of eight modules for the metahuman model by clarifying the relationships between the function features, behavior characteristics, and structure schemes based on hierarchical mapping.

3.4. Technical Topic Analysis of Patent Text

We chose to use the Google Colab platform to complete technical topic analysis. The major steps were as follows:

① Text pre-processing.

To ensure the scientific validity of the data analysis, we started with pre-processing of the corpus, which was composed of identified metahuman patent abstracts. The text pre-processing included the removal of stop words, case unification, root unification, and removal of symbols.

② Text clustering analysis.

Unsupervised learning was used to cluster the latent semantic structure of the text, and the Gensim (Generate Similarity) library in Python was used to vectorize the documents and build an LDA topic model with which to extract semantic topics.

③ TF-IDF processing.

To avoid including extracted feature words that were not representative of the topic, we used the term frequency-inverse document frequency (TF-IDF) algorithm to calculate the importance degree of each word in a document. Through the TF-IDF formula, we calculated the contribution of the word to the expression of a topic to obtain a vector that more reasonably represented the characteristics of the document. The formula is shown below:

$$TF-IDF\left(word-weight\right)=TF\ast IDF=\frac{N_A}{N}\ast ln\frac{\left|D\right|}{\left|D_A\right|+1}$$

(1)

In Formula (1), $N_A$ is the total number of lexical items $A$ in the document; $N$ is the total number of words in the document;$\left|D_A\right|$ refers to the total number of all documents in which lexical item $A$ appears;$\left|D\right|$ is the total number of all documents contained in the corpus.

④ Determining the optimal number of topics.

Two reference indexes, perplexity and coherence score, were calculated to evaluate the model effect and select the best number of topics. The perplexity represents the uncertainty that a particular document in the trained model belongs to a certain topic. The formula is shown below:

$$Perplexity=e^{\frac{-{\displaystyle \sum }\log \left(p\left(w\right)\right)}{N}}$$

(2)

In Formula (2), $p\left(w\right)$ is the probability of each word appearing in the test set, specifically in the LDA model, $p\left(w\right)={\displaystyle \sum zp\left(z|d\right)}*p\left(w|z\right)$ ($z$ and $d$ refer to the trained topics and each document in the test set, respectively). $N$ in the denominator denotes all the words appearing in the test set. The coherence score was measured for each topic by calculating the semantic similarity of high-probability words in the topic [54]. The formula is shown as follows:

$$Coherence\left(V\right)={\displaystyle \sum }_{\left(V_i,V_j\right)}score\left(V_i,V_j,\epsilon \right)$$

(3)

where $V$ represents the set of words describing the topic; $\epsilon$ denotes the smoothing factor [55].

The coherence score provides a basis for selecting more appropriate topics from the approximate range delineated by the perplexity according to quantifying semantic coherence indicators.

We set the initial numbers of output topics at 3–20 after referring to the characteristics of metahuman and product module numbers. We iteratively calculated the perplexity metrics for each model under different topics. The results show that the perplexity value decreased as the number of topics increased. In addition, the number of topics should not be too high as overfitting would occur in this case. We also comprehensively considered the numerical trend change of the coherence score for verification. Finally, we selected seven topics as the display results of the topic model after the above analysis process.

⑤ Topic-meaning mining and identification.

Each topic contains sets of feature words to describe the meaning of the topic. The basis for our technical topic mining came from the feature words corresponding to each topic and a full understanding of the technical knowledge in respect of metahuman. The results of topic naming are shown in

Table 4. Metahuman topics and their representative feature words.

No.	Topics	Feature Words
1	Audio-visual information processing and formation process	image, display, face, three-dimensional, screen, scene, VR
2	Signal connection and data analysis	license, trace, VRRP, bucket, foreground
3	Software support system	interrupt, PCIe, raid, waveguide, checksum
4	Hardware carrier equipment	memory, computing, application, information, environment
5	Network data information transmission and response	VPLS, transponder, infiniband, payload, IPv4, IPv6
6	Human–computer interaction application	avatar, game, shopping, financial, chatting, advertisement
7	Synthetic main material and design method	composition, compound, persona, profiling, campus

.

① The feature words image, display, face, three-dimensional, screen, scene, and VR in Topic 1 reflect the process of execution of a metahuman avatar image in a simulation environment through the interface system. The target sound and picture are rendered and presented in two or three dimensions using virtual reality technology. Therefore, we defined Topic 1 as “audiovisual information processing and formation process”.

② The feature words license, trace, VRRP, bucket, foreground, etc. in Topic 2 reflect the interconnection of different metahuman modular systems utilizing signals. This is a method for triggering different systems to execute data commands through vertical data processing. Therefore, we defined Topic 2 as “signal connection and data analysis”.

③ The feature words interrupt, PCIe, raid, waveguide, and checksum in Topic 3 reflect the software systems used to display the virtualized body. The underlying logic of the system includes virtual network identification of terminal attributes and sending of passwords, encryption/decryption operations for performing transactions, and authentication checking methods. Therefore, we defined Topic 3 as a “software support system”.

④ The feature words memory, computing, application, information, and environment in Topic 4 reflect the method of creating training data, driving, verifying, deploying, and testing virtual humanoid avatars in a virtual development environment. The feature words in this topic also described hardware devices including display devices, sensors, and optics. Therefore, Topic 4 was defined as “hardware carrier equipment”.

⑤ The feature words VPLS, transponder, infiniband, payload, IPv4, and IPv6 in Topic 5 reflect the steps of establishing a network link between receiving information from an input node and outputting control data for communication with a remote processing system. Therefore, we defined Topic 5 as “network data information transmission and response”.

⑥ The feature words avatar, game, shopping, finance, advertisement, and chatting in Topic 6 reflect practical applications of metahuman in the fields of games, media, culture, tourism, and finance, relying on interactive devices and interactive systems. Therefore, we defined Topic 6 as “human–computer interaction application”.

⑦ The characteristic words composition, compound, persona, profiling, and campus in Topic 7 reflect the compound, subject material, display components, and props of metahuman. The content of this topic also includes the design method of detecting the distribution of real human meridians in virtual reality. Therefore, we defined Topic 7 as “synthetic main material and design method”.

To verify whether the patent content matched the topic “label”, the top five patents with probability distributions under each topic were selected. We verified the accuracy of the topic labeling by checking whether the topic naming matched the original information of the patent. Taking Topic 1 (audiovisual information processing and formation process) as an example.

Table 5. Brief description of Topic 1′s top five metahuman patents.

Patent Publication Number	Patent-Topic Probability Distribution	Derwent Patent Title
US874675P	0.998176932	System for resecting target tissue mass from host tissue mass, where host tissue mass is deformable; involves surgical instrument; first fiducial sensor dimensioned to fit inside and next to target tissue mass; first fiducial sensor including hook to anchor first fiducial sensor inside
LT000066	0.997958779	Observation of spatial image stream by projecting stereo images at pre-determined angle using light beams and focusing onto light reflecting material e.g., reflection hologram, which creates spatial separate hologram surface viewing zones
CN10972624	0.997626662	Method of virtual reality human-computer interaction, involves determining moving speed of virtual cursor and set according to relative distance and first preset relationship, and orrespondingly setting first preset relationship and content of virtual object
CN11343006	0.997327328	Projection display device has light modulating panel which is pixel-level high-speed spatial light modulator set on side of light reflecting plate away from micro galvanometer array for modulating beam energy output by galvanometer array
CN11580108	0.997310519	Vehicle-mounted augmented reality enhancing head-up display system has camera, image controller, and augmented reality enhancing head-up display device comprising main reflector, secondary reflector, and image generator

shows that the top five patent titles all reflected the process of multimodal perception of metahuman through collecting or presetting image and speech information and focusing on the formation of spatial information flow. The content of these patents was consistent with the “label” of the topic. For the remaining six topics, the same method was also used to verify the results, which showed that the topic labeling was more accurate.

3.5. Matching Product Modularity, Technical Topics, and Patent Data

Metahuman technical topics were matched with the product modules by deeply analyzing the metahuman technical background and combining the results of the LDA analysis. Based on the patent-topic distribution information, the patents were divided into different product modules, as shown in Figure 3.

Specifically, the “Information Collection” module corresponds to Topic 7 (synthetic main material and design method), which provides the original reference samples for metahuman modeling through the collection of information on the human body, clothing, etc. The “Modeling & Rendering” and “Intelligence-Driven” modules correspond to Topic 4 (hardware carrier equipment) and Topic 3 (software support system), respectively. These are the two basic layers of support for the development of metahuman. The “Character Generation” module builds holographic, highly realistic characters through technical Topic 1 (audiovisual information processing and formation process) and Topic 7 (synthetic main material and design method). “Recognition & Perception” is realized from source input to end effect presentation through Topic 1 (audiovisual information processing and formation process) and Topic 6 (human–computer interaction application). The “Perception and recognition” module realizes the digital transmission and feedback process through analog signals, corresponding to Topic 5 (network data information transmission and response). The “Analysis & Decision” module realizes data capture and decoding, which mainly depends on key nodes, and corresponds to Topic 2 (signal connection and data analysis).

4. Results

4.1. Technology Trends

For the period 1973 to 2020, the number of metahuman patent applications is shown in Figure 4.

Considering both the number of the patent applications and their annual growth rate, the development of metahuman technology can be divided into three stages: The exploring development stage (1973–1999), the rapid development stage (2000–2015), and the sharp development stage (2016 to present).

The exploring development stage of metahuman technology was from 1973 to 1999. The first patent related to metahuman appeared in 1973; its patent number was US365795 and its priority country was the US. This patent belongs to computer virtualization technology, which is one of the bases for creating metahuman. From 1973 to 1993, the annual patent application number was below 100, with a large variety of annual growth rates and various types of modules appearing one after another. By 1994, the number of related patent applications worldwide exceeded 100 for the first time. By 1988, patents related to all seven types of metahuman product modules had appeared. From 1995 to 1999, the number of patent applications grew steadily, with an average annual growth rate of about 25%, and the number of applications reached 363 in 1999. The rapid development stage of metahuman was from 2000 to 2015. In 2000 the annual growth rate was 78.7%. After the sharp growth, the development entered a stable period, with an average annual growth rate of 16.25% from 2000 to 2015. This period saw a rapid rise in metahuman technology. In 2015, the number of patent applications was 3305. The Modeling & Rendering, Synthesis & Display, and Character Generation modules also evolved at a faster pace. The fast development stage was from 2016 to the present day. In 2016, the annual growth rate was 40.7%. After that, the average annual patent growth rate was 11.46% from 2016 to 2020. While the growth rate was lower than in the rapid development stage, the number of patents maintained a high level. In 2020, the number of patents reached 5422.

4.2. Geographical Layout

Usually, a patent applicant can obtain patent priority in their first patent application country, and then use the priority to apply for patents in other countries within one year [56]. Therefore, the technological strength of a country can be indicated by the number of priority patent applications it has [57]. The technology dynamics of each country in the field of metahuman can be analyzed by comparing the patent priority numbers in different countries.

Among the 42,256 patents related to metahuman, a total of 105 priority countries are involved. According to the countries’ patent priority numbers, the top 10 countries include the US, China, South Korea, and Japan. The total number of patents in these four countries accounts for 91.17% of all patents, so we analyzed the development of metahuman technology based on the data for the top four countries.

As can be seen from Figure 5, the US and Japan are the main original developers of metahuman-related technologies. The US occupied the worldwide top position for a long time until it was overtaken by China in 2017. The number of patents in the US continued to decline from 2017 to 2020.

China first appeared as a metahuman patent priority country in 1994. While China started late, its annual number of patent applications surpassed those of South Korea in 2008, Japan in 2009, and the US in 2017. From 2017 to 2020, China occupied the number one position in the world. From 2010 to 2020, the average growth rate of metahuman technology patents in China was 28.82%.

Japan accumulated certain advantages in the initial development stage of metahuman technology. The first Japanese patent appeared in 1984, 7 years earlier than South Korea and 10 years earlier than China. However, Japan was overtaken by both South Korea and China in 2009. South Korea and Japan entered a situation of alternating leadership from 2009 to the present. In 2020, Japan occupied third place globally in terms of the number of metahuman patents.

We selected the US, China, South Korea, and Japan to gain insight into the geographic distribution of metahuman product modules by using a spider diagram. The results are shown in Figure 6.

The US and China are leaders in the field but have different technological advantages. The US ranks first worldwide in patent applications for the “Modeling & Rendering” module and has an obvious advantage in this field. China, which ranks second, has 3055 patents in this field, which is about a third of that of the US. China dominates globally in the fields of “Character Generation” and “Synthesis & Display”. In 2022, the number of patent applications for the former is 6891, about 1.87 times that of the US, and the number of patents for the latter is 6520, about 1.3 times that of the US.

Both the US and South Korea have one significant technological advantage in respect of metahuman. The US has an absolute advantage in the field of “Modeling & Rendering”, with 9253 patents. This is 1.83 times that in the “Synthesis & Display” field, which ranks second in the US. The “Character Generation” module follows with 3673 patents. The numbers of patents for the other four modules range from 1440 to 2900. In South Korea, compared with other modules, the “Synthesis & Display” module has the leading number of patents at 1779, which accounts for 1/2 of the total module patents in the country. The remaining six modules all have smaller numbers of patent applications, i.e., under 1000.

Both China and Japan have two significant advantages in the field of metahuman technologies. “Character Generation” and “Synthesis & Display” are both in the top tier of these two countries, with the absolute leading position. In China, the patent applications for these two modules were 6891 and 6520, respectively, while there were only 6700 total patent applications for the other five modules. In Japan, the patent applications for these two modules were 1052 and 999, respectively, and there were only 1580 applications for the other five modules.

The “Synthesis & Display” module is dominant in all four countries. China is ahead of other countries in patent applications for the “Synthesis & Display” module, showing a clear advantage in image analysis and virtual reality systems technologies. South Korea ranks first in patent applications for this module, indicating that South Korea’s research and development in respect of related technologies is the core of the country. In both the US and Japan, this module is in the second position in terms of patents, representing the module as a key research area for both countries.

In general, the overall level of technology in the field of metahuman varies greatly from country to country. The “Modeling & Rendering” module ranks first in the world in the US, and is three times larger than in China. While China ranks first in both “Character Generation” and “Synthesis & Display”, its advantage over the US is not particularly clear. The technology that has the best advantage in Japan and South Korea is not so obvious in the international arena.

4.3. Technological Revolution

Based on the matching relationships between product modules, technical topics, and patent data, we analyzed the development of metahuman technology at the product module level using patent data. The technical trends of metahuman product modules in different periods are plotted separately, as shown in Figure 7 and Figure 8.

The “Information Collection” module is the first to appear, and it accounts for a relatively low percentage among all the modules. As time goes by, although the related patent applications are increasing, their proportions in the overall metahuman technology are gradually decreasing. While the “Information Collection” module appears earlier, its importance is relatively low due to its low technical difficulty and high technical maturity.

The “Modeling & Rendering” module also appears earlier, and the current technology development is relatively stable. Before 2000, the technologies covered by this module were in a steady development period, and after 2000, these technologies developed rapidly, with an annual growth rate of more than 50%. The technology share peaked in 2012 and has been declining steadily since then. Thus, the technology supporting this kind of module has gradually matured.

The proportion of “Intelligence-Driven” module technology has been at a relatively stable level. This module first appeared in 1985, and the level of technology continued to improve from that time; patent applications accounted for about 8–10% of all applications, maintaining a relatively stable state. The continuous development of driven technology is of great significance to metahuman in terms of emotion, movement, and user interactivity, especially when “Intelligence-Driven” is the focus of future development.

The “Character Generation” module shows a fluctuating trend in the early stage of development and continues to climb in the middle and late stages. Before 1993, there were fewer than 10 patent applications for this module. In 2020, patent applications accounted for 30.52%, and this number is still increasing. The module covers the generation of language, speech, expression, body, etc., and eventually achieved the integration of natural speech, facial animation implementation, and the action of a series of metahuman. The development of technologies in this module is the key to creating metahuman with more anthropomorphic, expressive, and interactive features at present. The patent applications supporting the “Character Generation” module not only occupied a high proportion but were also in continuous development. This module will soon be the technical focus.

The “Synthesis & Display” module developed slowly in the early stages and gradually took the lead in the development trend. Since the beginning of the 1990s, the patent applications for this module have increased to about 25%, which is a relatively stable amount and has been rising in recent years. The implementation of analysis and display functions requires the integration of more data and is based on other technologies. While the initial development of this module was slow, this area will be the focus of technology in the future.

The “Recognition & Perception” module developed rapidly in the early stages, and then developed slowly in the later years. In the 1980s, patents supporting this module began to appear, mainly in respect of data processing technologies. The number of technologies in this module has been increasing, and the percentage of patents in this category peaked in 1985 and then fluctuated for years with an average of 12%. It has been gradually declining since 2014 and currently stands at about 5.29%. This module includes technologies for memory storage, processor execution instructions, and data decoding, and is the current technology focus for optimizing the functionality of metahuman.

The “Analysis & Decision” module accounts for a relatively small proportion of patent applications, and shows a fluctuating development trend at the beginning. After 2000, the proportion remained relatively stable at about 9.96%, which was relatively low, but this module is also in the process of continuous development. The analysis and decision module can only be implemented by combining various types of data and decision models, and the technologies in this module are relatively mature.

5. Conclusions

5.1. Discussion

In this work we studied development trends and the competitive situation in respect of metahuman and proposed a novel analysis framework including the components of product modules, technical topics, and patent data. Since metahuman involves the integration of advanced technologies, the analysis results may influence technology iteration and innovation in R&D enterprises, industrial development strategies for policymakers, and necessity applications in human lives. Therefore, we discussed issues related to the economic and social aspects of sustainable development. Metahuman is a hot topic nowadays, and current studies mainly focus on the following three dimensions: Concept definition, development history, and application scenarios [4,20,58]. However, research on metahuman technological analysis is still in its infancy. Thus, to provide valuable information for promoting the sustainable development of metahuman, we conducted a patent analysis of metahuman technology.

Scholars have suggested the advantages of using patents to reveal emerging technologies, since patents provide greatly detailed descriptions of the technological aspects [59]. The search results of this research were obtained from the DII (Derwent Innovations Index) patent database, which is a web-based patent information database that integrates the Derwent World Patents Index (DWPI) and the Patents Citation Index (PCI) to provide global patent information [60]. The DII database is a powerful tool for searching worldwide invention patent information and has unique characteristics, in that it takes a series of measures when including patent data from various countries, for the purpose of improving the comprehensiveness and accuracy of patent searches [61]. In this study, cross-country analysis using the DII database [62,63,64] provided references for retrieving patents from the DII database for the comparative analysis of countries and guaranteed that the mainstream technical direction and research results provided would be of use to the greatest extent.

The discussion of the selected database is useful to ensure the balance and robustness of the data used for analysis in this research. The findings through patent analysis in the time dimension illustrate the total changes, evolutionary stages, and development trends of metahuman technology. The results indicate that the global patent applications related to metahuman are aligned with the R&D history of metahuman [65], passing through the periods of early-stage exploration, rapid development, and the recent peak stage. Due to technical limitations, metahuman in the early exploration stage were not mature enough and the product quality was low. According to the results of annual patent applications, the years of 2000 and 2016 were two key time-points marking the milestones of different development stages. The year 2000 became an important turning point in the history of metahuman development due to the presentation of the social practice function of metahuman [66]. Metahuman presented by computer animation is gradually being replaced by metahuman presented by CG, motion capture, and other computer technologies. In addition, the year 2016 has been dubbed the “Year of Artificial Intelligence”, in which a global AI trend erupted. Breakthroughs in AI technology and deep learning algorithms have led to simplification of the metahuman production process, leading the development of metahuman into a new period. Thus, this research provides objective empirical test results for the evolutionary history of metahuman.

The results of analyzing four typical patent priority countries provide two main insights into national technical competitive advantages. From the time series viewpoint, the US patent applications from 1973 to 2000 reached 21,356 and China came in second with a total of 12,220 patents. The overall number of patents held by the US and China from 1973 to 2000, which accounted for 80.16% of all patents, greatly outstripped those of other nations. The US dominated the metahuman technology sector for decades in terms of overall technical prowess until it was overtaken by China in recent years. From 2017 to 2020, China filed 6560 patents, 1.2 times as many as the US (5449 patents). Thus, China has been leading the development of metahuman technology with tremendous technological advancement during the last four years. Since metahuman is a virtual image constructed by artificial intelligence technology [67], the development of metahuman is closely related to the innovation of artificial intelligence. The existing literature indicates that the US has a dominant position in the development of artificial intelligence industries and plays a crucial role in the global development of AI technical networks [68,69]. The research results show that artificial intelligence industries are also constantly and rapidly developing in China [70]. South Korea and Japan were comparable in terms of metahuman technology, and we observed a small difference in the number of patent applications between these two countries. There would have been a total of 2378 and 2151 patent applications in South Korea and Japan from 1973 to 2020. The total number of applications from 2016 to 2020 was 1047 in South Korea and 769 in Japan. Furthermore, from the technical layout viewpoint, the findings show that different countries present special characteristics at the product module level. Except for “Character Generation” and “Synthesis & Display”, which were in second place behind China in terms of patent applications, the US held a relatively dominant position in the majority of technology modules. China outperformed other nations outside of the US in the “Character Generation” and “Synthesis & Display” modules, as well as in the “Modeling & Rendering”, “Intelligence-Driven”, “Information Collection”, “Recognition & Perception”, and “Analysis & Decision” modules. Our findings fill the gaps in knowledge regarding the technological layout and competitive advantages of dominant countries in the field of metahuman.

The analysis of technological change from the modular perspective effectively bridges the product development lifecycle and technology evolution. We summarized the emergence and growth of patent technologies related to the following modules through the analysis of patent data: “Information Collection”, “Recognition & Perception”, “Synthesis & Display”, “Analysis & Decision”, “Character Generation”, “Modeling & Rendering”, and “Intelligence-Driven”. Additionally, we observed that the technologies underlying metahuman functional modules started with the relatively simple modules of “Information Collection”, “Recognition & Perception”, “Synthesis & Display”, and “Analysis & Decision”, and then moved on to the relatively intricate technologies of “Character Generation”, “Modeling & Rendering”, and “Intelligence-Driven”. The research results indicate how technological convergence evolution has taken place from 1973 to 2020. The authors found some prominent technologies that gained significant importance over this period. We also compared our findings with other studies, showing that facial recognition technology, capture-based and synthesis motion, rendering technology, and “Digital Human Modeling” based on virtual reality are all topics of interest to scholars [71,72,73,74]. However, the “Intelligence-Driven” module is hardly mentioned in previous research, possibly because there has been limited exploration of modules in the practical dimension. Our findings provide evidence for reflecting that technological progress is an important internal driving force for the development activities of metahuman.

5.2. Theoretical and Practical Implications

Major implications of this study are to be found in the methodological and practical aspects of R&D enterprises and policymakers. Unlike many previous studies that identified technical topics, we took a further step and applied an optimized and systematic analysis process, for the purpose of minimizing the possibility of misidentifying emerging technology areas and improving the usability of the research results. FBS was used to build seven modules for the metahuman product from the dimensions of function, behavior, and structure. Taking patent abstracts as the corpus based on 42,256 patents relating to metahuman, the LDA method was utilized to extract seven technological subjects and their corresponding patent data. Finally, through expert consultation, it was determined that the product modules, technical topics, and related patent codes corresponded, which provided a solid database for the patent analysis of metahuman technological development. Therefore, the process for recognizing patent data in complex items, which cover numerous patents, making it challenging to locate particular search parameters, was improved. The analytical framework we developed can provide a reference for evaluating technologies in the metahuman field using patent data.

Several notable aspects of metahuman technology were discovered through the patent analysis conducted in this research. These findings provide theoretical references for R&D enterprises and governments to comprehensively grasp the status quo of the field and make strategic decisions. The patent application trend in patent-priority countries represents the overall scale of technology R&D, current patent holdings, and international market layout capabilities of different countries. The US was the first to explore related metahuman technologies, followed by Japan, but Japan’s overall scale was far from that of the US, and it was quickly overtaken by South Korea and China. The patent trends revealed that, since 2017, patenting advantages shifted to China, which is currently leading the patent share by a large margin compared to other countries. In addition, each country has its own advantages in metahuman technologies as a priority country. By comparing and analyzing the key R&D directions and layout of each country, the differences in technology layouts between countries can be identified, which helps to determine the advantages of the national layout and clarify the future layout optimization direction. These analysis results will assist government officials in making decisions for future industrial development of metahuman based on the technology layout and competitiveness assessment.

The patent analysis results also revealed several interesting patterns of patenting activity in metahuman technology relating to different modules. The findings of characteristics help enterprises or technology developers to assess and grasp the evolutionary trend of key R&D directions. In addition, the results are useful for investors to identify the possibility of new technology investments in a predictable manner to avoid risks and maintain sustainable prosperities. It is observed that some R&D directions have also evolved from key R&D directions to non-key R&D directions. For example, “information collection” was a key R&D direction in the early development stage, but has ceased to be a key R&D direction since the rapid rise of development. In addition, “Information Collection”, “Analysis & Decision”, “Recognition & Perception” and “Modeling & Rendering” technologies have undergone the process of initial development and rapid development, and were key R&D directions before 2000, but the number of patent applications since then has shown a deceleration and they have ceased to be key R&D directions. In addition, R&D enterprises continue to pay more attention to the two key directions of “Character Generation” and “Synthesis & Display”. The number of patents belonging to these two modules has increased significantly, and the R&D direction has gradually expanded. At present, their development is currently accelerating. The “Synthesis & Display” module was observed as the most active in respect of filing metahuman patents, indicating that the technologies belonging to this module have always shown good momentum in respect of development. Thus, it is a key research and development direction in terms of enterprise sustainability. While “Intelligence-Driven” has not demonstrated a rapid development trend, it is most likely to be the next major technological breakthrough point.

5.3. Limitations and Future Research Directions

This research has the following limitations, which can be investigated further in the future. Firstly, future work could examine different types of patents, for the purpose of providing a more comprehensive picture of technical advancement. Secondly, we investigated the technological structure and development level of metahuman from a modular perspective. However, the quality of integration between modules could influence a metahuman’s final quality. The integration between technologies and its impact also needs to be taken into consideration in the future. Thirdly, the development trends of metahuman were primarily analyzed from the dimensions of time, product modules, and country distribution. In the future, the interrelationships among technologies, such as co-citations and co-occurrences, can be further analyzed, to enable a deeper understanding of metahuman development and to make suggestions that are more appropriate in reality.