Graphene industry application discovery method based on large model and knowledge graph analysis

Abstract

The invention discloses a graphene industry application discovery method based on a large model and knowledge graph analysis, which comprises the following steps: S1, collecting patent literatures, scientific research papers, market analysis reports and industry news related to graphene, and constructing a multi-source heterogeneous data set; s2, preprocessing the multi-source heterogeneous data set; s3, performing information extraction on the processed multi-source heterogeneous data set by utilizing a BERT model, wherein the extracted information comprises technical key points, application fields, market demands and technical trends; s4, constructing a graphene patent and technology trend knowledge graph, and integrating the extracted key information into a unified semantic network; and S5, performing application discovery analysis based on the constructed graphene patent and technology trend knowledge graph, identifying a potential application scene of the graphene material, and providing application optimization suggestions and new market opportunities. According to the method, the analysis efficiency of a research institution is remarkably improved, and a scientific basis is provided for strategic decision making of the research institution in an emerging market.

Description

Technical Field

[0001] The present invention relates to the field of graphene technology, and in particular to a graphene industry application discovery method based on large model and knowledge graph analysis. Background Art

[0002] Graphene is a two-dimensional carbon material with unique physical and chemical properties. Due to its excellent electrical conductivity, thermal conductivity and mechanical strength, it has shown broad application prospects in electronic devices, energy storage, biomedicine and other fields. With the deepening of graphene research, more and more graphene-related technologies and application scenarios are constantly being explored. However, how to systematically discover and analyze the potential applications of graphene in different fields has become an important topic in current research and industrial development.

[0003] At present, the research on the industrial application of graphene mainly relies on the experience of experts and experimental research in specific fields. The research is usually limited to a specific field, such as electronic devices or energy storage. There is a lack of comprehensive analysis of graphene in multiple fields and systematic exploration of cross-field applications. Most of the existing graphene application research adopts traditional data collection and analysis methods, relying on manual screening and collation of existing literature and market data, which is not only time-consuming and labor-intensive, but also prone to missing potential important information. In addition, the existing graphene application analysis methods are mostly based on static data analysis, which cannot dynamically reflect the development trend of graphene technology and the market, and it is difficult to capture emerging opportunities in technological evolution.

[0004] Although existing technologies have promoted the development of the graphene industry to a certain extent, there are still some obvious shortcomings. First, existing analysis methods are mostly single-field research, lacking systematic exploration of cross-field application scenarios of graphene, resulting in limited discovery of potential application scenarios. Second, existing research methods mainly rely on manual screening and manual analysis, which are difficult to cope with the multi-source heterogeneity and massive characteristics of graphene-related data, resulting in low efficiency of data integration and information extraction. Furthermore, existing analyses are mostly based on static and historical data information, which makes it difficult to predict the future development trend of graphene technology and changes in market demand, and cannot provide forward-looking strategic support for decision makers. Therefore, how to propose a graphene industry application discovery method based on large models and knowledge graph analysis, and then comprehensively and systematically discover potential application scenarios of graphene, has become an urgent problem to be solved in the current technical field. Summary of the invention

[0005] One purpose of the present invention is to propose a graphene industry application discovery method based on large models and knowledge graph analysis. The present invention significantly improves the analysis efficiency of research institutions and provides a scientific basis for their strategic decision-making in emerging markets.

[0006] A graphene industry application discovery method based on a large model and knowledge graph analysis according to an embodiment of the present invention includes the following steps:

[0007] S1. Collect graphene-related patent documents, scientific research papers, market analysis reports and industry news to build a multi-source heterogeneous dataset;

[0008] S2, preprocessing of multi-source heterogeneous data sets, including data format standardization and text data structuring;

[0009] S3. Use the BERT model to extract information from the processed multi-source heterogeneous data set. The extracted information includes technical points, application fields, market demand and technological trends. The BERT model identifies key information related to graphene by analyzing the semantics of the text;

[0010] S4. Construct a knowledge graph of graphene patents and technology trends, and integrate the extracted key information into a unified semantic network. The knowledge graph includes the timeline, technical field, creativity, and technology development trend of the patents, and reveals the key trends and potential technology gaps of graphene technology through the correlation analysis between patents and technology trends.

[0011] S5. Conduct application discovery analysis based on the constructed graphene patent and technology trend knowledge graph to identify potential application scenarios of graphene materials and provide application optimization suggestions and new market opportunities.

[0012] Optionally, the S1 includes the following steps:

[0013] S11. Identify and select multiple data sources related to graphene, including patent databases, scientific research literature databases, market analysis platforms, and industry news websites;

[0014] S12. Extract graphene-related data from data sources, including patent documents P, scientific research papers R, market analysis reports M, and industry news N:

[0015] Patent document P: including patent number, applicant, inventor, technical field, application date, publication date and patent abstract;

[0016] Scientific research paper R: including paper title, author, journal, publication date, abstract and keywords;

[0017] Market analysis report M: including report title, issuing agency, report date, analysis content and market trends;

[0018] Industry news: including news title, release time, source and main content;

[0019] S13. Capture data from graphene-related data sources:

[0020] Extract text from patent document P to generate patent dataset D p ;

[0021] Extract text from scientific research paper R to generate scientific research dataset D r ;

[0022] Extract text from market analysis report M to generate market data set D m ;

[0023] Extract text from industry news N to generate news dataset D n .

[0024] S14. Obtain the patent dataset D p , scientific research dataset D r 、Market Dataset D m and news dataset D n Integrate and construct multi-source heterogeneous datasets D clean .

[0025] Optionally, S3 includes the following steps:

[0026] S31. For multi-source heterogeneous datasets D clean Perform data preprocessing, including word segmentation, part-of-speech tagging, stop word removal, and text vectorization of text data, and represent the preprocessed graphene-related text data as a vector set X;

[0027] S32. Build and train an improved BERT model specific to the graphene industry;

[0028] S33, inputting the preprocessed vector set X into the improved BERT model for deep semantic analysis, the improved BERT model encodes the input graphene-related text data through a multi-layer Transformer structure to generate a context-related text representation H;

[0029] S34. Use the multi-task classifier MTC in the improved BERT model to classify the text representation H, extract information related to graphene technology key points, application fields, market demand and technology trends, and generate the corresponding information label set T:

[0030] Technical Points Label Collection T tech By analyzing the patent dataset D p Generation of technical details and innovations in

[0031] Application field tag set T app By analyzing the scientific research dataset D r Application scenarios and technical implementation generation in

[0032] Market demand label set T market By analyzing the market data set D m Generation of market trends and demand forecasts in

[0033] Technology Trend Label Collection T trend By analyzing the news dataset D n Generate industry dynamics and technology development direction;

[0034] S35, the extracted technical key points label set T tech , application field label set T app , Market demand label set T market and technology trend label collection T trend Summarize and generate a semantic vector set V of graphene-related information, which represents the technical key points, application prospects, market demand and technological development trends of the graphene industry.

[0035] Optionally, the S32 includes the following steps:

[0036] S321. Select dataset D specific to the graphene industry train As a training data set, the data set D train Including patent dataset D p , scientific research dataset D r 、Market Dataset D m and news dataset D n A subset of

[0037] S322. Pre-train a word embedding model in the field of graphene, which is obtained by training on patent datasets, scientific research datasets, market datasets, and news datasets in the field of graphene. The word embedding model is represented by E graphene :

[0038]

[0039] Among them, w t represents the target word, w c represents the context words, T is the total number of words in the graphene-related field, and C t is the target word w t The context window of

[0040] S323. Based on the BERT model, a word embedding layer for the graphene field is added to embed the pre-trained word embedding model E graphene As the input of the BERT model, by adding a word embedding layer in the graphene field, the BERT model can capture domain-specific information when processing graphene-related data:

[0041] H0=Egraphene +P pos ;

[0042] Among them, H0 is the initial input representation, P pos is positional embedding;

[0043] S324, improve the pre-trained BERT model, use domain adaptive fine-tuning technology to fine-tune the initial weight W0 of the BERT model through the training data set, and generate a model weight W specific to the graphene field graphene :

[0044]

[0045] Where η is the learning rate, B is the batch size, is the loss function of the b-th batch, Represents the b-th batch of training data;

[0046] S325. Input the training dataset into the improved BERT model containing the domain-specific word embedding layer, encode the text data through the multi-layer Transformer encoder, and generate the context-related text representation H:

[0047] H=Transformer(H0,W graphene );

[0048] Among them, Transformer is a multi-layer self-attention mechanism that processes the input text representation H0 to generate a representation H containing graphene-specific contextual information;

[0049] S326, introduce a multi-task learning mechanism, integrate different tasks of the graphene industry into the joint learning framework, and use the multi-task loss function L multi Perform joint optimization:

[0050]

[0051] in, They represent the loss functions of the i-th task in technical point identification, application field classification, market demand forecasting and technology trend analysis, respectively, and α i is the weight coefficient of each task;

[0052] S327, Improved BERT model BERT graphene Verification and testing are carried out to make the BERT model accurate and robust in identifying graphene technology key points, application areas, market demands and technological trends.

[0053] Optionally, S4 includes the following steps:

[0054] S41. Extracting graphene-related technical key point label set T from the semantic vector set V tech , application field label set T app , Market demand label set T market and technology trend label collection T trend , as the basic node for building the knowledge graph of graphene patents and technology trends:

[0055] V={v1,v2,,v n}, where v i =BERT graphene (T tech ,T app ,T market ,T trend );

[0056] Among them, V represents the semantic vector set of graphene-related information, v i is the i-th vector in the set;

[0057] S42. Define the node types of the graphene patent and technology trend knowledge graph, including patent node N p , Technical field node N tech 、Application field node N app 、Market demand node N market and technology trend node N trend , the node is defined as:

[0058] Where V ij =Attention(v j ,v k );

[0059] Among them, N i represents the embedding representation of the i-th node, V ij is the attention weight between nodes, by focusing on the semantic vector v j and v k The relationship between is generated by embedding function f embed The vector representation is converted into specific node types in the graph, and the nodes together form the skeleton of the graphene patent and technology trend knowledge graph;

[0060] S43. By analyzing the semantic relationship between nodes, define the edge types between nodes, including technical association edges E tech , apply the associated edge E app 、Market demand related edge E market and technology trend correlation edge E trend , and calculate the weight W of each edge e, by calculating the edge weights and determining the association strength between nodes, the technical association and application relationship are formed in the graphene patent and technology trend knowledge graph:

[0061]

[0062] Among them, W eij Represents node N i and node N j The edge weight between i ,N j ) is node N i and node N j The semantic similarity between i ,N j ) represents node N i and node N j The strength of the relationship between |V il -V jl | is node N i and node N j The embedding difference of

[0063] S44. Construct a graphene patent and technology trend knowledge graph KG based on the types of nodes and edges and their weights graphene , which represents the patent technology evolution path, application field expansion, market demand changes and technology trend evolution in the graphene industry. The relationship between all nodes and edges constitutes the structure of the entire graphene patent and technology trend knowledge graph, which represents the graphene technology evolution path and its expansion in different application fields and market demands:

[0064]

[0065] Among them, KG graphene It is a graphene graphene patent and technology trend knowledge map constructed through node N i and node N j The edge weight between The product of is represented by;

[0066] S45. Constructed graphene patent and technology trend knowledge graph KG graphene Perform timing analysis to identify the timestamps of each technology node Timestamps identify the time sequence of each technology node in the graphene patent and technology trend knowledge graph, forming a timeline of graphene technology, revealing the evolution and trend of graphene technology, and providing key moments in the timestamp during the technology development process:

[0067]

[0068] S46. Using the graphene patent and technology trend knowledge graph KG graphene Conduct technology correlation and trend analysis to identify technology gaps in patent landscape and unmet market needs:

[0069]

[0070] Among them, Gap tech Indicates the size of the technology gap, freq(N j ) is node N j The frequency of occurrence in the knowledge graph.

[0071] Optionally, S5 includes the following steps:

[0072] S51. Knowledge graph based on graphene patents and technology trends KG graphene , by analyzing patent node N p , Technical field node N tech 、Application field node N app 、Market demand node N market and technology trend node N trend The correlation between them can identify the potential application scenarios of graphene materials in different technical fields;

[0073] S52, define application discovery function F app , according to node N i and node N j The semantic similarity between i ,N j ) and the strength of association Calculate the application opportunity score S app :

[0074]

[0075] Among them, S app Represents the potential application opportunities of graphene materials in specific technical fields, with a weight of Reflects the strength of association between nodes;

[0076] S53. Score based on application opportunity app , identify graphene materials in the field of technology tech and application areas app Among the potential application scenarios, select the scenarios with higher scores as the key application directions;

[0077] S54. Conduct technical feasibility analysis on the identified potential application scenarios, evaluate the practical feasibility of graphene materials in different application scenarios based on the physical and chemical properties of graphene materials and market demand, and generate application optimization suggestions. opt,Optimization suggestions include the matching degree between material properties and application scenarios, the urgency of market demand, and the difficulty of technical implementation;

[0078] S55, based on application optimization suggestions R opt , by constructing a market opportunity matrix, new opportunities for graphene materials in different markets are identified. The market opportunity matrix is ​​defined as:

[0079]

[0080] Among them, M opportunity represents market opportunities, R opt (N i ) is the application scenario score of the optimization suggestion, M potential (N i ) represents the potential index of each application scenario in the market;

[0081] S56. Display the final application analysis results and market opportunity matrix through visualization tools to help decision makers understand the application prospects and new market opportunities of graphene materials.

[0082] The beneficial effects of the present invention are:

[0083] (1) The present invention introduces an improved BERT model to perform semantic analysis and information extraction on multi-source heterogeneous data sets, thereby solving the problem of low efficiency of manual screening and manual analysis in the prior art. By using domain adaptive fine-tuning technology to improve the BERT model, the model is made more suitable for specific data in the graphene field, and the accuracy and depth of semantic information extraction are improved, ensuring the efficient integration and structured processing of multi-source heterogeneous data such as graphene-related patents, scientific research literature, market analysis and industry news.

[0084] (2) The present invention realizes dynamic correlation analysis of graphene technology by constructing a patent and technology trend knowledge graph specific to the graphene field. It not only shows the patent evolution path of graphene technology, but also reveals the key trends and potential technological gaps in the development of graphene technology through semantic correlation analysis of technical fields, application fields, market demands and technological trends. The dynamic knowledge graph construction method helps decision makers understand the evolution process of graphene technology in real time and discover potential technological innovations and market opportunities.

[0085] (3) The present invention integrates multiple application fields of graphene technology into a joint learning framework for analysis through a multi-task learning mechanism. By constructing an application discovery function for graphene materials and utilizing the semantic similarity and association strength in the knowledge graph, the present invention systematically identifies potential application scenarios of graphene in multiple technical fields. This not only enables the discovery of optimized applications in traditional fields, but also predicts the application prospects of graphene in emerging fields, thereby providing more comprehensive application development guidance for enterprises and research institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0087] Figure 1 A flowchart of a graphene industry application discovery method based on a large model and knowledge graph analysis proposed by the present invention;

[0088] Figure 2 The present invention proposes a flowchart for constructing a graphene patent and technology trend knowledge graph in a graphene industry application discovery method based on a large model and knowledge graph analysis. DETAILED DESCRIPTION

[0089] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0090] refer to Figure 1-2 , a graphene industry application discovery method based on large model and knowledge graph analysis, comprising the following steps:

[0091] S1. Collect graphene-related patent documents, scientific research papers, market analysis reports and industry news to build a multi-source heterogeneous dataset;

[0092] S2, preprocessing of multi-source heterogeneous data sets, including data format standardization and text data structuring;

[0093] S3. Use the BERT model to extract information from the processed multi-source heterogeneous data sets. The extracted information includes technical points, application fields, market demand and technological trends. The BERT model identifies key information related to graphene by analyzing the semantics of the text.

[0094] S4. Construct a knowledge graph of graphene patents and technology trends, and integrate the extracted key information into a unified semantic network. The knowledge graph includes the timeline, technical field, creativity, and technology development trend of the patents, and reveals the key trends and potential technology gaps of graphene technology through the correlation analysis between patents and technology trends.

[0095] S5. Conduct application discovery analysis based on the constructed graphene patent and technology trend knowledge graph to identify potential application scenarios of graphene materials and provide application optimization suggestions and new market opportunities.

[0096] In this implementation, S1 includes the following steps:

[0097] S11. Identify and select multiple data sources related to graphene, including patent databases, scientific research literature databases, market analysis platforms, and industry news websites;

[0098] S12. Extract graphene-related data from data sources, including patent documents P, scientific research papers R, market analysis reports M, and industry news N:

[0099] Patent document P: including patent number, applicant, inventor, technical field, application date, publication date and patent abstract;

[0100] Scientific research paper R: including paper title, author, journal, publication date, abstract and keywords;

[0101] Market analysis report M: including report title, issuing agency, report date, analysis content and market trends;

[0102] Industry news: including news title, release time, source and main content;

[0103] S13. Capture data from graphene-related data sources:

[0104] Extract text from patent document P to generate patent dataset D p ;

[0105] Extract text from scientific research paper R to generate scientific research dataset D r ;

[0106] Extract text from market analysis report M to generate market data set D m ;

[0107] Extract text from industry news N to generate news dataset D n .

[0108] S14. Obtain the patent dataset D p , scientific research dataset D r、Market Dataset D m and news dataset D n Integrate and construct multi-source heterogeneous datasets D clean .

[0109] In this implementation, S3 includes the following steps:

[0110] S31. For multi-source heterogeneous datasets D clean Perform data preprocessing, including word segmentation, part-of-speech tagging, stop word removal, and text vectorization of text data, and represent the preprocessed graphene-related text data as a vector set X;

[0111] S32. Build and train an improved BERT model specific to the graphene industry;

[0112] S33, inputting the preprocessed vector set X into the improved BERT model for deep semantic analysis, the improved BERT model encodes the input graphene-related text data through a multi-layer Transformer structure to generate a context-related text representation H;

[0113] S34. Use the multi-task classifier MTC in the improved BERT model to classify the text representation H, extract information related to graphene technology key points, application fields, market demand and technology trends, and generate the corresponding information label set T:

[0114] Technical Points Label Collection T tech By analyzing the patent dataset D p Generation of technical details and innovations in

[0115] Application field tag set T app By analyzing the scientific research dataset D r Application scenarios and technical implementation generation in

[0116] Market demand label set T market By analyzing the market data set D m Generation of market trends and demand forecasts in

[0117] Technology Trend Label Collection T trend By analyzing the news dataset D n Generate industry dynamics and technology development direction;

[0118] S35, the extracted technical key points label set T tech , application field label set T app , Market demand label set T market and technology trend label collection T trendSummarize and generate a semantic vector set V of graphene-related information, which represents the technical key points, application prospects, market demand and technological development trends of the graphene industry.

[0119] In this implementation, S32 includes the following steps:

[0120] S321. Select dataset D specific to the graphene industry train As the training data set, data set D train Including patent dataset D p , scientific research dataset D r 、Market Dataset D m and news dataset D n A subset of

[0121] S322. Pre-train a word embedding model in the field of graphene, which is obtained by training on patent datasets, scientific research datasets, market datasets, and news datasets in the field of graphene. The word embedding model is represented by E graphene :

[0122]

[0123] Among them, w t represents the target word, w c represents the context words, T is the total number of words in the graphene-related field, and C t is the target word w t The context window of

[0124] S323. Based on the BERT model, a word embedding layer for the graphene field is added to embed the pre-trained word embedding model E graphene As the input of the BERT model, by adding a word embedding layer in the graphene field, the BERT model can capture domain-specific information when processing graphene-related data:

[0125] H0=E graphene +P pos ;

[0126] Among them, H0 is the initial input representation, P pos is positional embedding;

[0127] S324, improve the pre-trained BERT model, use domain adaptive fine-tuning technology to fine-tune the initial weight W0 of the BERT model through the training data set, and generate a model weight W specific to the graphene field graphene :

[0128]

[0129] Where η is the learning rate, B is the batch size, is the loss function of the b-th batch, Represents the b-th batch of training data;

[0130] S325. Input the training dataset into the improved BERT model containing the domain-specific word embedding layer, encode the text data through the multi-layer Transformer encoder, and generate the context-related text representation H:

[0131] H=Transformer(H0,W graphene );

[0132] Among them, Transformer is a multi-layer self-attention mechanism that processes the input text representation H0 to generate a representation H containing graphene-specific contextual information;

[0133] S326, introduce a multi-task learning mechanism, integrate different tasks of the graphene industry into the joint learning framework, and use the multi-task loss function L multi Perform joint optimization:

[0134]

[0135] in, They represent the loss functions of the i-th task in technical point identification, application field classification, market demand forecasting and technology trend analysis, respectively, and α i is the weight coefficient of each task;

[0136] S327, Improved BERT model BERT graphene Verification and testing are carried out to make the BERT model accurate and robust in identifying graphene technology key points, application areas, market demands and technological trends.

[0137] In this implementation, S4 includes the following steps:

[0138] S41. Extracting graphene-related technical key point label set T from the semantic vector set V tech , application field label set T app , Market demand label set T market and technology trend label collection T trend , as the basic node for building the knowledge graph of graphene patents and technology trends:

[0139] V = {v1, v2, v n}, where v i =BERT graphene (T tech , T app , T market , T trend);

[0140] Among them, V represents the semantic vector set of graphene-related information, v i is the i-th vector in the set;

[0141] S42. Define the node types of the graphene patent and technology trend knowledge graph, including patent node N p , Technical field node N tech 、Application field node N app 、Market demand node N market and technology trend node N trend , the node is defined as:

[0142] Where V ij =Attention(v j , v k );

[0143] Among them, N i represents the embedding representation of the i-th node, V ij is the attention weight between nodes, by focusing on the semantic vector v j and v k The relationship between is generated by embedding function f embed The vector representation is converted into specific node types in the graph, and the nodes together form the skeleton of the graphene patent and technology trend knowledge graph;

[0144] S43. By analyzing the semantic relationship between nodes, define the edge types between nodes, including technical association edges E tech , apply the associated edge E app 、Market demand related edge E market and technology trend correlation edge E trend , and calculate the weight W of each edge e , by calculating the edge weights and determining the association strength between nodes, the technical association and application relationship are formed in the graphene patent and technology trend knowledge graph:

[0145]

[0146] in, Represents node N i and node N j The edge weight between i , N j ) is node N i and node N j The semantic similarity between i , N j ) represents node N i and node Nj The strength of the relationship between |V il -V jl | is node N i and node N j The embedding difference of

[0147] S44. Construct a graphene patent and technology trend knowledge graph KG based on the types of nodes and edges and their weights graphene , which represents the patent technology evolution path, application field expansion, market demand changes and technology trend evolution in the graphene industry. The relationship between all nodes and edges constitutes the structure of the entire graphene patent and technology trend knowledge graph, which represents the graphene technology evolution path and its expansion in different application fields and market demands:

[0148]

[0149] Among them, KG graphene It is a graphene graphene patent and technology trend knowledge map constructed through node N i and node N j The edge weight between The product of is represented by;

[0150] S45. Constructed graphene patent and technology trend knowledge graph KG graphene Perform timing analysis to identify the timestamps of each technology node Timestamps identify the time sequence of each technology node in the graphene patent and technology trend knowledge graph, forming a timeline of graphene technology, revealing the evolution and trend of graphene technology, and providing key moments in the timestamp during the technology development process:

[0151]

[0152] S46. Using the graphene patent and technology trend knowledge graph KG graphene Conduct technology correlation and trend analysis to identify technology gaps in patent landscape and unmet market needs:

[0153]

[0154] Among them, Gap tech Indicates the size of the technology gap, freq(N j ) is node N j The frequency of occurrence in the knowledge graph.

[0155] In this implementation, S5 includes the following steps:

[0156] S51. Knowledge graph based on graphene patents and technology trends KG graphene, by analyzing patent node N p , Technical field node N tech 、Application field node N app 、Market demand node N market and technology trend node N trend The correlation between them can identify the potential application scenarios of graphene materials in different technical fields;

[0157] S52, define application discovery function F app , according to node N i and node N j The semantic similarity between i , N j ) and the strength of association Calculate the application opportunity score S app :

[0158]

[0159] Among them, S app Represents the potential application opportunities of graphene materials in specific technical fields, with a weight of Reflects the strength of association between nodes;

[0160] S53. Score based on application opportunity app , identify graphene materials in the field of technology tech and application areas app Among the potential application scenarios, select the scenarios with higher scores as the key application directions;

[0161] S54. Conduct technical feasibility analysis on the identified potential application scenarios, evaluate the practical feasibility of graphene materials in different application scenarios based on the physical and chemical properties of graphene materials and market demand, and generate application optimization suggestions. opt ,Optimization suggestions include the matching degree between material properties and application scenarios, the urgency of market demand, and the difficulty of technical implementation;

[0162] S55, based on application optimization suggestions R opt , by constructing a market opportunity matrix, new opportunities for graphene materials in different markets are identified. The market opportunity matrix is ​​defined as:

[0163]

[0164] Among them, M opportunity represents market opportunities, R opt (N i ) is the application scenario score of the optimization suggestion, M potential (N i ) represents the potential index of each application scenario in the market;

[0165] S56. Display the final application analysis results and market opportunity matrix through visualization tools to help decision makers understand the application prospects and new market opportunities of graphene materials.

[0166] Embodiment 1:

[0167] In early 2024, in order to verify the effect of the present invention, a well-known graphene research and development company located in Shenzhen, China, was applied to explore the application potential of graphene in emerging technology fields. The company has rich experience in the research and development of graphene materials, but has certain difficulties in dealing with massive data analysis and cross-domain application scenario identification. For this reason, they chose to adopt the method of the present invention to analyze graphene-related patent literature, scientific research papers, market analysis reports and industry news collected from 2019 to 2023. The goal is to discover the potential applications of graphene in quantum computing and flexible electronic devices in a short period of time, and optimize existing market strategies.

[0168] In January 2024, the company extracted graphene technology and market data for the past five years from global graphene-related databases, with a total data volume of 120TB, including 1.5 million patent documents, 850,000 scientific research papers, 5,000 market analysis reports and 100,000 industry news. Through the method of the present invention, the multi-source heterogeneous data is first preprocessed, and the unstructured text data is converted into structured semantic units, laying the foundation for the subsequent knowledge graph construction.

[0169] During the data processing stage, the method of the present invention completed the preprocessing of all data in just 15 days, while the traditional method took 45 days. During the processing, the system performed strict anomaly detection on each data source and found that 7.5% of the data had format problems. After automated repair, 100% of the data integration was finally achieved.

[0170] In February 2024, the company began to use the improved BERT model to analyze the processed data set, focusing on the field of quantum computing. By analyzing nearly 100,000 related patent documents, the company found an interesting phenomenon: since 2021, research on the application of graphene in quantum dot devices and superconducting quantum computing has gradually increased, but market reports show that the technology conversion rate in this field is still low.

[0171] Through further knowledge graph construction and analysis, the system identified a series of key technical nodes, including "improved stability of quantum dot materials" and "low-temperature application of superconducting quantum computing". Based on the technical nodes, the system generated application scenario suggestions, showing that graphene is expected to improve computing stability and efficiency in superconducting quantum computing.

[0172] In the comparative analysis, the traditional method failed to effectively capture the emerging technology nodes, resulting in the identification of only some relatively conventional application scenarios. On the contrary, the method of the present invention accurately identified the trend that the market demand for quantum computing will increase at an average annual growth rate of 42.5% in the next three years, and put forward targeted R&D suggestions.

[0173] In March of the same year, the company applied the method of the present invention to conduct an in-depth analysis of the field of flexible electronics. Through the construction and analysis of the knowledge graph, the system discovered the wide application potential of graphene in flexible OLED displays and smart wearable devices. Through time series analysis, the system identified that the growth rate of flexible electronics-related patents between 2019 and 2023 reached 37.8%, and predicted that this field will continue to expand at an average annual growth rate of 30.2% in the next five years.

[0174] Through quantitative analysis, the system also discovered technological gaps in the field of flexible electronics, especially in the research and development of transparent conductive film materials. To verify this finding, the system conducted a regression analysis of market demand data from 2019 to 2023. The results showed that the predicted market demand value for transparent conductive films is 15.7%, significantly higher than the average level in other fields (10.3%).

[0175] Through the method of the present invention, the company quickly adjusted its market strategy, increased its investment in the research and development of transparent conductive films, and reached cooperation agreements with a number of smart device manufacturers. It is expected that the products will be commercialized within the next two years.

[0176] Table 1 below is a specific comparison data between the method of the present invention and the traditional method in the field of quantum computing and flexible electronics, demonstrating the superiority of the method of the present invention:

[0177] Table 1 Specific comparative data between the method of the present invention and the traditional method in the field of quantum computing and flexible electronics

[0178] project Method of the present invention Traditional methods Data processing time (days) 15.5 45.2 Data integration success rate (%) 100.0 92.5 Number of application scenarios identified (quantum computing) 12.8 5.6 Number of application scenarios identified (flexible electronics) 15.3 7.2 Forecasted market demand growth rate (quantum computing, %) 42.5 27.8 Forecasted market demand growth rate (flexible electronics, %) 30.2 18.7 Key technology node identification rate (%) 97.4 74.6 Market demand forecast accuracy (%) 89.3 65.1

[0179] It can be seen from the data in Table 1 that the method of the present invention is far superior to the traditional method in terms of data processing efficiency, accuracy of application scenario identification and accuracy of market demand forecasting. The data proves the feasibility and effectiveness of the present invention in practical applications, especially in the application discovery of the emerging technology field of graphene, which has significant advantages.

[0180] The present invention introduces an improved BERT model to perform semantic analysis and information extraction on multi-source heterogeneous data sets, thereby solving the problem of low efficiency of manual screening and manual analysis in the prior art. By improving the BERT model using domain adaptive fine-tuning technology, the model is made more suitable for specific data in the graphene field, and the accuracy and depth of semantic information extraction are improved, ensuring the efficient integration and structured processing of multi-source heterogeneous data such as graphene-related patents, scientific research literature, market analysis and industry news.

[0181] The present invention realizes dynamic correlation analysis of graphene technology by constructing a patent and technology trend knowledge graph specific to the graphene field. It not only shows the patent evolution path of graphene technology, but also reveals the key trends and potential technological gaps in the development of graphene technology through semantic correlation analysis of technical fields, application fields, market demands and technological trends. The dynamic knowledge graph construction method helps decision makers understand the evolution process of graphene technology in real time and discover potential technological innovations and market opportunities.

[0182] Through a multi-task learning mechanism, the present invention integrates multiple application fields of graphene technology into a joint learning framework for analysis. By constructing an application discovery function for graphene materials and utilizing the semantic similarity and association strength in the knowledge graph, the present invention systematically identifies potential application scenarios of graphene in multiple technical fields. It can not only discover optimized applications in traditional fields, but also predict the application prospects of graphene in emerging fields, thereby providing more comprehensive application development guidance for enterprises and research institutions.

[0183] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A graphene industry application discovery method based on large model and knowledge graph analysis, characterized in that: The steps include: S1. Collect graphene-related patent documents, scientific research papers, market analysis reports and industry news to build a multi-source heterogeneous dataset; S2, preprocessing of multi-source heterogeneous data sets, including data format standardization and text data structuring; S3. Use the BERT model to extract information from the processed multi-source heterogeneous data set. The extracted information includes technical points, application fields, market demand and technological trends. The BERT model identifies key information related to graphene by analyzing the semantics of the text; S4. Construct a knowledge graph of graphene patents and technology trends, and integrate the extracted key information into a unified semantic network. The knowledge graph includes the timeline, technical field, creativity, and technology development trend of the patents, and reveals the key trends and potential technology gaps of graphene technology through the correlation analysis between patents and technology trends. S5. Conduct application discovery analysis based on the constructed graphene patent and technology trend knowledge graph to identify potential application scenarios of graphene materials and provide application optimization suggestions and new market opportunities.

2. According to claim 1, a graphene industry application discovery method based on large model and knowledge graph analysis is characterized in that: The S1 comprises the following steps: S11. Identify and select multiple data sources related to graphene, including patent databases, scientific research literature databases, market analysis platforms, and industry news websites; S12. Extract graphene-related data from data sources, including patent documents P, scientific research papers R, market analysis reports M, and industry news N: Patent document P: including patent number, applicant, inventor, technical field, application date, publication date and patent abstract; Scientific research paper R: including paper title, author, journal, publication date, abstract and keywords; Market analysis report M: including report title, issuing agency, report date, analysis content and market trends; Industry news N: including news title, release time, source and main content; S13. Capture data from graphene-related data sources: Extract text from patent document P to generate patent dataset D p ; Extract text from scientific research paper R to generate scientific research dataset D r ; Extract text from market analysis report M to generate market data set D m ; Extract text from industry news N to generate news dataset D n . S14. Obtain the patent dataset D p , scientific research dataset D r 、Market Dataset D m and news dataset D n Integrate and construct multi-source heterogeneous datasets D clean .

3. The method for discovering graphene industry applications based on large models and knowledge graph analysis according to claim 1, characterized in that: The S3 comprises the following steps: S31. For multi-source heterogeneous datasets D clean Perform data preprocessing, including word segmentation, part-of-speech tagging, stop word removal, and text vectorization of text data, and represent the preprocessed graphene-related text data as a vector set X; S32. Build and train an improved BERT model specific to the graphene industry; S33, inputting the preprocessed vector set X into the improved BERT model for deep semantic analysis, the improved BERT model encodes the input graphene-related text data through a multi-layer Transformer structure to generate a context-related text representation H; S34. Use the multi-task classifier MTC in the improved BERT model to classify the text representation H, extract information related to graphene technology key points, application fields, market demand and technology trends, and generate the corresponding information label set T: Technical Points Label Collection T tech By analyzing the patent dataset D p Generation of technical details and innovations in Application field tag set T app By analyzing the scientific research dataset D r Application scenarios and technical implementation generation in Market demand label set T market By analyzing the market data set D m Generation of market trends and demand forecasts in Technology Trend Label Collection T trend By analyzing the news dataset D n Generate industry dynamics and technology development direction; S35, the extracted technical key points label set T tech , application field label set T app , Market demand label set T market and technology trend label collection T trend Summarize and generate a semantic vector set V of graphene-related information, which represents the technical key points, application prospects, market demand and technological development trends of the graphene industry.

4. The method for discovering graphene industry applications based on large models and knowledge graph analysis according to claim 1 is characterized in that: The S32 comprises the following steps: S321. Select dataset D specific to the graphene industry train As a training data set, the data set D train Including patent dataset D p , scientific research dataset D r 、Market Dataset D m and news dataset D n A subset of S322. Pre-train a word embedding model in the field of graphene, which is obtained by training on patent datasets, scientific research datasets, market datasets, and news datasets in the field of graphene. The word embedding model is represented by E graphene : Among them, w t represents the target word, w c represents the context words, T is the total number of words in the graphene-related field, and C t is the target word w t The context window of S323. Based on the BERT model, a word embedding layer for the graphene field is added to embed the pre-trained word embedding model E graphene As the input of the BERT model, by adding a word embedding layer in the graphene field, the BERT model can capture domain-specific information when processing graphene-related data: H0=E graphene +P pos ; Among them, H0 is the initial input representation, P pos is positional embedding; S324, improve the pre-trained BERT model, use domain adaptive fine-tuning technology to fine-tune the initial weight W0 of the BERT model through the training data set, and generate a model weight W specific to the graphene field graphene : Where η is the learning rate, B is the batch size, is the loss function of the b-th batch, Represents the b-th batch of training data; S325. Input the training dataset into the improved BERT model containing the domain-specific word embedding layer, encode the text data through the multi-layer Transformer encoder, and generate the context-related text representation H: H=Transformer(H0,W graphene ); Among them, Transformer is a multi-layer self-attention mechanism that processes the input text representation H0 to generate a representation H containing graphene-specific contextual information; S326, introduce a multi-task learning mechanism, integrate different tasks of the graphene industry into the joint learning framework, and use the multi-task loss function L multi Perform joint optimization: in, They represent the loss functions of the i-th task in technical point identification, application field classification, market demand forecasting and technology trend analysis, respectively, and α i is the weight coefficient of each task; S327, Improved BERT model BERT graphene Verification and testing are carried out to make the BERT model accurate and robust in identifying graphene technology key points, application areas, market demands and technological trends.

5. The method for discovering graphene industry applications based on large models and knowledge graph analysis according to claim 1, characterized in that: The S4 comprises the following steps: S41. Extracting graphene-related technical key point label set T from the semantic vector set V tech , application field label set T app , Market demand label set T market and technology trend label collection T trend , as the basic node for building the knowledge graph of graphene patents and technology trends: V={v1,v2,,v n }, where v i =BERT graphene (T tech ,T app ,T market ,T trend ); Among them, V represents the semantic vector set of graphene-related information, v i is the i-th vector in the set; S42. Define the node types of the graphene patent and technology trend knowledge graph, including patent node N p , Technical field node N tech 、Application field node N app 、Market demand node N market and technology trend node N trend , the node is defined as: Where V ij =Attention(v j ,v k ); Among them, N i represents the embedding representation of the i-th node, V ij is the attention weight between nodes, by focusing on the semantic vector v j and v k The relationship between is generated by embedding function f embed The vector representation is converted into specific node types in the graph, and the nodes together form the skeleton of the graphene patent and technology trend knowledge graph; S43. By analyzing the semantic relationship between nodes, define the edge types between nodes, including technical association edges E tech , apply the associated edge E app 、Market demand related edge E market and technology trend correlation edge E trend , and calculate the weight W of each edge e , by calculating the edge weights and determining the association strength between nodes, the technical association and application relationship are formed in the graphene patent and technology trend knowledge graph: in, Represents node N i and node N j The edge weight between i ,N j ) is node N i and node N j The semantic similarity between i ,N j ) represents node N i and node N j The strength of the relationship between |V il -V jl | is node N i and node N j 's embedding differences; S44. Construct a graphene patent and technology trend knowledge graph KG based on the types of nodes and edges and their weights graphene , which represents the patent technology evolution path, application field expansion, market demand changes and technology trend evolution in the graphene industry. The relationship between all nodes and edges constitutes the structure of the entire graphene patent and technology trend knowledge graph, which represents the graphene technology evolution path and its expansion in different application fields and market demands: Among them, KG graphene It is a graphene graphene patent and technology trend knowledge map constructed through node N i and node N j The edge weight between The product of is represented by; S45. Constructed graphene patent and technology trend knowledge graph KG graphene Perform timing analysis to identify the timestamps of each technology node Timestamps identify the time sequence of each technology node in the graphene patent and technology trend knowledge graph, forming a timeline of graphene technology, revealing the evolution and trend of graphene technology, and providing key moments in the timestamp during the technology development process: S46. Using the graphene patent and technology trend knowledge graph KG graphene Conduct technology correlation and trend analysis to identify technology gaps in patent landscape and unmet market needs: Among them, Gap tech Indicates the size of the technology gap, freq(N j ) is node N j The frequency of occurrence in the knowledge graph.

6. The method for discovering graphene industry applications based on large models and knowledge graph analysis according to claim 1, characterized in that: The S5 comprises the following steps: S51. Knowledge graph based on graphene patents and technology trends KG graphene , by analyzing patent node N p , Technical field node N tech 、Application field node N app 、Market demand node N market and technology trend node N trend The correlation between them can identify the potential application scenarios of graphene materials in different technical fields; S52, define application discovery function F app , according to node N i and node N j The semantic similarity between i ,N j ) and the strength of association Calculate the application opportunity score S app : Among them, S app Represents the potential application opportunities of graphene materials in a specific technical field, with a weight of W eij Reflects the strength of association between nodes; S53. Score based on application opportunity app , identify graphene materials in the field of technology tech and application areas app Among the potential application scenarios, select the scenarios with higher scores as the key application directions; S54. Conduct technical feasibility analysis on the identified potential application scenarios, evaluate the practical feasibility of graphene materials in different application scenarios based on the physical and chemical properties of graphene materials and market demand, and generate application optimization suggestions. opt ,Optimization suggestions include the matching degree between material properties and application scenarios, the urgency of market demand, and the difficulty of technical implementation; S55, based on application optimization suggestions R opt , by constructing a market opportunity matrix, new opportunities for graphene materials in different markets are identified. The market opportunity matrix is ​​defined as: Among them, M opportunity represents market opportunities, R opt (N i ) is the application scenario score of the optimization suggestion, M potential (N i ) represents the potential index of each application scenario in the market; S56. Display the final application analysis results and market opportunity matrix through visualization tools to help decision makers understand the application prospects and new market opportunities of graphene materials.

Images