Enterprise industry chain association method and system and storable medium

Abstract

The invention discloses an enterprise industry chain association method and system and a storable medium, and the method comprises the steps: defining an industry chain basic atlas which comprises a plurality of industry chain levels, and each industry chain level comprises a plurality of nodes; performing subsequent hierarchical expansion on the industrial chain basic map by using the first large language model to form an industrial chain ultimate map; node words of the industrial chain ultimate graph nodes are expanded through the first large language model, and a professional word library is formed; obtaining text data of different dimensions of an enterprise, and performing enterprise label extraction on the text data; and associating the enterprise label with the industry chain ultimate map, and returning a link related to a node corresponding to the enterprise label. According to the method, the industrial chain of the enterprise can be accurately and quickly associated.

Description

Technical Field

[0001] The present invention relates to the field of big data processing technology, and in particular to an enterprise industrial chain association method, system and storable medium. Background Art

[0002] The industrial chain knowledge graph is a key tool for connecting enterprises and their industries. It displays the relationship between entities such as enterprises, products, and technologies in a structured way.

[0003] Early maps relied on rules and simple algorithms, and were mostly judged by the industry to which the enterprise belonged. This judgment method was relatively general, and the types of enterprises classified by industry were mostly concentrated in listed companies and other key enterprises, while it was impossible to make accurate judgments on small and medium-sized enterprises. In this manual matching method, industry-related researchers find corresponding enterprises based on the company's industry classification, and compile standardized labels for segmented products based on the company's products; determine the upstream, midstream and downstream relationship map and sort out the standardized labels of products contained in the industrial chain. By integrating this information, a clear industrial chain structure diagram is formed. First of all, for enterprises, their business will span multiple industries, and the existing industry classification is unique, which will result in the inability of label classification to map multiple industries; secondly, it takes a long time to manually organize labels, and the classification level is also limited.

[0004] With the development of artificial intelligence and machine learning technologies, the construction of industrial chain knowledge graphs has become more intelligent and automated. However, the industrial chain graphs generated by traditional machine learning methods are relatively generalized and messy, and have certain limitations, mainly reflected in the shallow level and the inability to go deep into the precise relationship between enterprise products and technologies. Therefore, how to improve the efficiency and accuracy of enterprise industrial chain associations has become a technical problem to be solved. Summary of the invention

[0005] In order to overcome the above shortcomings, the purpose of the present invention is to provide a method, system and storable medium for associating an enterprise's industrial chain, which can accurately and quickly associate the enterprise's industrial chain.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is: a method for associating an enterprise industrial chain, comprising: Define a basic graph of the industrial chain, wherein the basic graph of the industrial chain includes a number of industrial chain levels, and each of the industrial chain levels includes a number of nodes; The first language model is used to expand the basic graph of the industrial chain to subsequent levels to form the ultimate graph of the industrial chain; Expanding the node words of the ultimate graph nodes of the industrial chain through the first language model to form a professional vocabulary; Acquire text data of different dimensions of the enterprise, and extract enterprise labels from the text data; The enterprise tag is associated with the ultimate graph of the industrial chain, and the link related to the node corresponding to the enterprise tag is returned.

[0007] The beneficial effects of the present invention are: The basic graph of the industrial chain defines the basic hierarchical structure, provides clear input and constraints for the big model, and then uses the first language big model to extract fixed edge relationships and refine the hierarchy of the industrial chain graph; The integration of professional vocabulary, the ultimate graph of the industrial chain and the collaborative work of multiple models reduces the association errors that may be introduced by the use of generalized vocabulary, and significantly improves the recognition accuracy of professional terms for enterprise products and technologies; The selected large language model is used for automatic association of enterprise industries, reducing the need for manual screening and classification, and greatly improving the efficiency of enterprise industrial chain association.

[0008] Specifically, extracting enterprise tags from the text data includes: Extracting keywords from the text data using the first language model and the second model; Calculating the similarity between the extracted keywords and the words in the professional vocabulary, and obtaining the word in the professional vocabulary that is closest to the text data as a label; The labels obtained by the first language model and the second model are aggregated, and the repeated labels are taken as the final enterprise labels.

[0009] The two large language models can complement each other, and after aggregating the labels of the two, the accuracy of the final extracted enterprise labels can be improved.

[0010] Specifically, the first language model is an LLM model, and the LLM model extracts the label by: The LLM model extracts keywords from the text data; Calculate the similarity between the extracted keywords and the node words and synonyms in the professional vocabulary to obtain the node words and synonyms that meet the requirements; Compare the similarity with a set threshold, and use the node words or synonyms whose similarity is greater than the set threshold as the most similar professional words; Returning the professional word and the text data to the LLM model, the LLM model determining whether the professional word is a word expressed in the text data; If so, keep the professional term as a label.

[0011] The LLM model is more flexible and is an extension of the pre-trained model. In addition to understanding semantics, it can also generate content based on existing text.

[0012] Furthermore, the second model is a KeyBERT model, and the KeyBERT model specifically extracts the label including: The KeyBERT model is embedded in a professional vocabulary, and weights are set for node words and synonyms in the professional vocabulary; Set the number of words to be extracted and perform keyword extraction; Calculate the similarity between the extracted keywords and the node words and synonyms in the professional vocabulary to obtain the node words and synonyms that meet the requirements; The similarity is compared with a set threshold, and the node words or synonyms with similarity greater than the set threshold are used as labels.

[0013] KeyBERT is more lightweight and uses a smaller text pre-training model. KeyBERT can directly cut words from long texts and extract common keywords. The extracted keywords follow the similarity calculated by the pre-training model.

[0014] Furthermore, the first language model is used to expand the basic graph of the industrial chain to subsequent levels to form the ultimate graph of the industrial chain, which specifically includes: The first language model is used to expand the basic graph of the industrial chain to form an intermediate graph of the industrial chain; The intermediate map of the industrial chain is verified and corrected to form the ultimate map of the industrial chain.

[0015] Specifically, the first language model is used to expand the basic graph of the industrial chain to form an intermediate graph of the industrial chain, which specifically includes: Select all nodes of the last level of the industrial chain in the industrial chain basic graph; Use the first language model to expand the node words of all nodes at the last industrial chain level and output the results; The output results are collected and mapped to relevant nodes.

[0016] Specifically, the node words of the ultimate graph node of the industrial chain are expanded by the first language model to form a professional vocabulary including: Synonyms are expanded through the first language model to identify synonyms and related concepts of node words at each node of the ultimate graph of the industrial chain, and the expanded synonyms are mapped to corresponding nodes. The synonyms and node words form a professional vocabulary.

[0017] Furthermore, the basic map of the industrial chain includes three levels of the industrial chain, and only the upstream and midstream industries are retained in the basic map of the industrial chain.

[0018] The basic graph of the industrial chain provides clear input and constraints for subsequent large models, that is, the relationship between fixed entities and edges. The diversity of downstream application scenarios will lead to generalization and confusion of labels, especially between different industries and markets. Therefore, we focus on retaining the upstream and midstream industries to improve the accuracy and practicality of labels. All basic graphs of the industrial chain do not retain the downstream industries.

[0019] The present invention also discloses an enterprise industrial chain association system, which adopts the above-mentioned enterprise industrial chain association method, and the system includes: A storage module storing a basic graph of the industrial chain; A first expansion module, wherein the first expansion module is used to expand the basic graph of the industrial chain by using the first language model to a subsequent level, so as to form and store the ultimate graph of the industrial chain; A second expansion module, which expands the nodes in the ultimate graph of the industrial chain through the first language model to form a professional vocabulary and store it; A label extraction module, which is used to obtain text data of different dimensions of an enterprise and extract enterprise labels from the text data; The association module is used to associate the enterprise tag with the industrial chain and return the link related to the node corresponding to the enterprise tag.

[0020] The present invention also discloses a computer-readable storage medium, on which instructions are stored. When the instructions are executed by a processor, the execution instructions are used to implement the above-mentioned enterprise industrial chain association method. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 The process of the first embodiment of the present invention Figure 1 ; Figure 2 The process of the first embodiment of the present invention Figure 2 ; Figure 3 This is a system block diagram of Embodiment 2 of the present invention. DETAILED DESCRIPTION

[0022] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby making a clearer and more definite definition of the protection scope of the present invention.

[0023] See attached Figure 1 and attached Figure 2As shown, a method for associating an enterprise industrial chain of the present invention specifically includes: S100. Define a basic graph of the industrial chain, wherein the basic graph of the industrial chain includes several industrial chain levels, and each industrial chain level includes several nodes.

[0024] In this embodiment, the basic graph of the industrial chain in step 1 is manually defined by manually collecting relevant industry research reports, which usually contain the basic framework of each industry, from which the basic graph of the industrial chain can be defined. Manually defining the basic graph of the industrial chain provides clear input and constraints for the subsequent large model, that is, fixing the relationship between entities and edges.

[0025] In one embodiment, the basic graph of the industrial chain includes three industrial chain levels, namely the first level of the industrial chain, the second level of the industrial chain and the third level of the industrial chain. Usually, manually defining three industrial chain levels can meet the requirements of subsequent level expansion.

[0026] In one embodiment, only the upstream and midstream of the industry are retained in the basic map of the industrial chain, that is, only the upstream and midstream nodes are included in the secondary level of the industry. The basic framework of the industry research report usually includes the raw materials and equipment of the upstream industry, the production, processing and manufacturing of the midstream, and the application scenarios and sales channels of the downstream. However, the upstream and midstream nodes often have more standardized production processes and quality control standards, which makes label extraction and analysis more consistent and comparable. The diversity of downstream application scenarios will lead to generalization and confusion of labels, especially between different industries and markets. Therefore, in this embodiment, we focus on retaining the upstream and midstream of the industry to improve the accuracy and practicality of the labels, and all basic maps of the industrial chain do not retain the downstream of the industry.

[0027] For example, the first level of the industrial chain is integrated circuits, and the second level of the industrial chain includes two nodes, one is upstream and the other is midstream. The upstream includes semiconductor materials such as silicon wafers and electronic special gases, as well as key equipment such as photolithography machines and etching machines, while the midstream links cover the design, manufacturing, and packaging and testing processes of integrated circuits, which are all nodes of the third level of the industrial chain.

[0028] S200: Use the first language model to expand the basic graph of the industrial chain to subsequent levels to form the ultimate graph of the industrial chain.

[0029] Step S200 specifically includes: S21. Use the first language model to expand the basic graph of the industrial chain to form the intermediate graph of the industrial chain.

[0030] S22. Verify and correct the intermediate map of the industrial chain to form the ultimate map of the industrial chain.

[0031] In this embodiment, the powerful learning and information extraction capabilities of the large model can be used to further expand and enhance the graph to achieve a more detailed industrial chain graph.

[0032] Step S21 is specifically as follows: S211, select all nodes of the last industrial chain level in the industrial chain basic map; S212, using the first largest language model to expand all nodes in step S211, and output the result; S213: Collect the output results in step S212, and map the output results to relevant nodes.

[0033] In one embodiment, the first large language model is an LLM large model, which is set through prompt, the subject is the node name, and the relationship is set to expand by product category, core technology, and component.

[0034] For example, select the node semiconductor materials in the basic map of the industrial chain: integrated circuits-upstream-semiconductor materials, and input the integrated circuit semiconductor materials, categories, and core technologies in the LLM model. The LLM model outputs the results. Semiconductor materials include first-generation semiconductor materials (silicon and germanium), second-generation semiconductor materials (gallium arsenide and indium phosphide), third-generation semiconductor materials (gallium nitride and silicon carbide), and fourth-generation semiconductor materials (gallium oxide). According to the results, silicon and germanium materials are defined as the fifth level of the industrial chain. Subsequently, the model is used to expand the main categories of silicon and germanium materials, and the output results are germanium single crystals, silicon wafers, silicon rods (silicon ingots), silicon powder, etc.

[0035] Step S22 is completed manually to integrate and optimize the intermediate graph of the industrial chain expanded by the large language model to form the final ultimate graph of the industrial chain, such as removing duplicate items, so that it has a clear structure and a reasonable layout, which is easy to understand and analyze, and ensures the accuracy and completeness of all information.

[0036] S300, through the first language model, the node words of the ultimate graph nodes of the industrial chain are expanded to form a professional vocabulary.

[0037] The expanded professional vocabulary formed through step S300 not only enhances the understanding and description of each level of the industrial chain, but also improves the accuracy of the graph in the professional field.

[0038] In step 300, the nodes in the ultimate graph of the industrial chain are expanded by the first language model, specifically including: Synonyms are expanded through the first language model to identify the synonyms and related concepts of the node words at each node of the ultimate graph of the industrial chain, and the expanded synonyms are mapped to the corresponding nodes. Synonyms and node words form a professional vocabulary.

[0039] Synonyms include Chinese synonyms, English synonyms, English abbreviations, and phrases with similar expressions.

[0040] In this step, for each node word, its expression in different languages, Chinese and English, is identified, and the English abbreviation and full name of the node are expanded to form a professional vocabulary.

[0041] Step 400: Obtain text data of different dimensions of the enterprise and extract enterprise labels from the text data.

[0042] In this embodiment, enterprise tag extraction specifically includes: S41, extracting keywords from the text data using the first language model and the second model; S42, calculating the similarity between the extracted keywords and the professional vocabulary, and obtaining the words in the professional vocabulary that are closest to the text data as labels; S43. Summarize the labels obtained by the first language model and the second model, and take the repeated labels as the final enterprise labels.

[0043] From the perspective of the precise matching algorithm, the keywords currently extracted from the text to be processed contain some interference words and generalized words, which will reduce the accuracy of the association, and it is necessary to increase the accuracy of the extraction of enterprise product / technology labels. To solve the problem of matching errors, this embodiment uses multiple different large language models to improve the ability to understand the text content, and performs similarity calculations with the professional vocabulary at each level in the ultimate graph of the industrial chain generated in the third step to filter out interference words and improve the accuracy of the association.

[0044] In this embodiment, the text data of an enterprise can be obtained from the public data on the Internet through crawlers. The text data of different dimensions include the business scope of the enterprise, patent abstracts, official website introduction, news information, etc. By extracting and screening multi-dimensional data labels, the interference of false data can be effectively avoided.

[0045] The largest language model is the LLM model. When the LLM model extracts labels, it specifically includes: S411a. Use the LLM model to extract keywords from text data.

[0046] S411b, calculating similarity between the extracted keywords and the node words and synonyms in the professional vocabulary to obtain node words and synonyms that meet the requirements.

[0047] The similarity calculation of text usually converts the text into a vector representation, and then uses mathematical methods to calculate the distance or angle between these vectors to measure the similarity between the texts. The cosine similarity algorithm, the Euclidean distance algorithm, etc. can be used. When the LLM model performs similarity calculation, it can directly compare the similarity with the first threshold and retain the node words and synonyms that meet the requirements. There may be multiple node words and synonyms that meet the requirements.

[0048] S411c, comparing the similarity with a set threshold, and taking the node words or synonyms with similarity greater than the set threshold as the most similar professional words.

[0049] The similarity in this step refers to the similarity between the node words and synonyms that meet the requirements in step S411b and the keywords.

[0050] In the professional vocabulary, node words and synonyms are included. Keywords are calculated for similarity with node words and synonyms respectively. The comprehensive score method is used to combine the similarity score results based on node words and synonyms. Different weights are set for node words and synonyms, and a comprehensive score is formed through weighted average. Only when the comprehensive score exceeds this set threshold, it is selected. This method can reduce the impact of inaccurate professional term recognition to a certain extent and improve the accuracy and robustness of similarity calculation.

[0051] S411c, returning the extracted professional words and text data to the LLM model, and the LLM model determines whether the professional words are words expressed in the text data.

[0052] S411d. If yes, keep this professional word as a label; if no, do not use this professional word as a label.

[0053] The LLM model is more flexible and is an extension based on the pre-trained model. Based on the understanding of semantics, it can also generate content based on existing texts, that is, generate content-related keywords in this embodiment.

[0054] The second model is the KeyBERT model. When the KeyBERT model extracts labels, it specifically includes: S412a, the KeyBERT model is embedded in a professional vocabulary, and weights are set for node words and synonyms in the professional vocabulary.

[0055] Usually, a larger weight is set for the node words to increase the model's exposure to these terms through this step.

[0056] S412b, set the number of words to be extracted and perform keyword extraction.

[0057] To avoid the broadness of the extracted words, the extracted words are combined into short phrases.

[0058] S412c, calculating similarity between the extracted keywords and the node words and synonyms in the professional vocabulary to obtain node words and synonyms that meet the requirements.

[0059] S412d, comparing the similarity with a set threshold, and taking the node words or synonyms with similarity greater than the set threshold as labels.

[0060] Step S412d is the same as S411b. The professional vocabulary includes node words and synonyms. The keywords calculate the similarity of node words and synonyms respectively. The similarity score results based on node words and synonyms are combined by the comprehensive score method. Different weights are set for node words and synonyms, and a comprehensive score is formed by weighted average. Only when the comprehensive score exceeds this threshold, it is selected. Through this method, the impact of inaccurate professional terminology recognition can be reduced to a certain extent, and the accuracy and robustness of similarity calculation can be improved.

[0061] KeyBERT is more lightweight and uses a smaller text pre-training model. KeyBERT can directly cut words from long texts and extract common keywords. The extracted keywords follow the similarity calculated by the pre-training model.

[0062] In this embodiment, KeyBERT and LLM models can complement each other. After summarizing the labels of the two, the repeated labels are taken as the final enterprise labels. The integration of professional vocabulary, industrial chain map and the collaborative work of multiple models reduces the association errors that may be introduced by the use of generalized vocabulary, and significantly improves the recognition accuracy of professional terms for enterprise products and technologies. Step S400 can achieve the standardization of enterprise labels, that is, the node words in the industrial chain are used as enterprise labels.

[0063] S500: associate the enterprise tag with the ultimate graph of the industrial chain, and return the link related to the node corresponding to the enterprise tag.

[0064] Enterprise labels are derived from the professional vocabulary in the final graph of the industrial chain, which can avoid association errors when associating enterprises and industries. By directly matching the extracted enterprise labels with each node of the industrial chain, the role of the enterprise in the industrial chain can be determined, and the enterprise's product or technology labels can be accurately linked, returning the links and field labels related to the corresponding nodes.

[0065] For example, for "Yantuoxin (Suzhou) Integrated Circuit Co., Ltd.", the industry chain association is performed. When only the LLM model is used to extract enterprise labels, the association results obtained through different text data collected from the network are as follows: 1. Information Data The text data is: On September 27, 2023, A Co., Ltd. (hereinafter referred to as "A") learned that the company has completed the first round of financing, which was led by X1 Capital, followed by many investment institutions with industrial backgrounds such as X2 Investment, X3 Chuangxin, and X4 Investment, indicating the industry's recognition of A's strength. A was registered and established in April 2023 in the Artificial Intelligence Industrial Park, a carrier under a certain company. Its headquarters is located in B, with development centers in C and D, and a complete technical support laboratory in E. A is the first high-tech optoelectronic chip company in China with a global layout. It is committed to the research and development and industrialization of low-cost, high-performance optoelectronic chips. It has formed a full-chain product system covering 1.25G to 100G, from high-performance laser drivers at the transmitting end to high-sensitivity transimpedance amplifiers at the receiving end, including TIA, CDR, laser driver chips, etc., which are widely used in broadband fiber access, data centers, telecommunications and 5G base stations and other scenarios.

[0066] The keyword extraction results are: ['A', 'financing', 'X1', 'optoelectronic chip', 'R&D', 'industrialization', 'global layout', 'a company', 'artificial intelligence industrial park'].

[0067] The node words (node) and synonyms (syn) that meet the requirements are: {'syn': 'photoelectric converter', 'node': 'optoelectronic chip'}, {'syn': 'optoelectronic device chip', 'node': 'optoelectronic chip'}, {'syn': 'optoelectronic chip', 'node': 'optoelectronic chip'}, {'syn': 'special chip', 'node': 'ASIC'}, {'syn': 'customized integrated circuit', 'node': 'ASIC'}.

[0068] The final label is: optoelectronic chip.

[0069] The associated results are: integrated circuits-midstream-chip manufacturing-analog integrated circuits-optoelectronic chips.

[0070] 2. Patent Data The text data is: The present invention provides a chip mounting system and mounting method thereof, wherein the chip mounting system includes: a wafer film, which carries the chip; a wafer moving mechanism, which is used to adjust the initial position and angle of the wafer film; a pin mechanism, which is located below the wafer film, and is used to lift the chip upward to separate the chip from the wafer film; a wafer recognition camera, which is located above the pin mechanism and coaxial with the pin mechanism, and is used to identify and locate the position of the chip; a mounting substrate, which is located above the wafer recognition camera; a mounting device includes a positive mounting mechanism, a flip-down mechanism and a mounting mechanism, wherein the positive mounting mechanism is used to realize positive mounting of the chip, the flip-down mechanism is used to realize flip-down mounting of the chip, and the mounting mechanism is installed on a two-dimensional moving mechanism, and is used to pick up the chip from the positive mounting mechanism or the flip-down mechanism and mount it to the position to be mounted on the mounting substrate. The present invention is compatible with the flip-down mounting process and the positive mounting process, and can also improve the mounting efficiency and positioning accuracy of the chip.

[0071] The keyword extraction results are: ['chip mounting system', 'wafer film', 'ejector mechanism', 'wafer recognition camera', 'mounting substrate', 'upright mounting mechanism', 'flip mounting mechanism', 'two-dimensional moving mechanism', 'mounting efficiency', 'positioning accuracy'].

[0072] The node words (node) and synonyms (syn) that meet the requirements are: [{'syn': 'Chip packaging machine', 'node':'Chip bonding machine'}, {'syn': 'Chip alignment machine', 'node': 'Chip bonding machine'}, {'syn': 'Chip welding machine', 'node': 'Chip bonding machine'},{'syn': 'Semiconductor testing equipment', 'node': 'Wafer testing equipment'}].

[0073] The final label is: chip bonding machine.

[0074] The associated results are: integrated circuit-upstream-IC packaging equipment-chip bonding machine.

[0075] 3. Company official website The text data is: The main products of A Integrated Circuit Co., Ltd. cover multiple fields of optoelectronic chips, mainly including: 1.25G to 100G full-chain optoelectronic chip product system: The company has formed optoelectronic chip products covering from 1.25G to 100G, which include high-performance laser drivers at the transmitting end and high-sensitivity transimpedance amplifiers at the receiving end. TIA (transimpedance amplifier): As a key component at the receiving end, TIA chips play an important role in the process of optoelectronic signal conversion. CDR (clock and data recovery) chips: This type of chip is used for clock synchronization and data recovery in high-speed data communications.

[0076] Laser driver chip: The chip used to drive the laser is crucial to ensure the stable operation of the laser. Super transimpedance amplifier: The company launched the world's first Super transimpedance amplifier for gigabit networks at a recent optical exhibition, marking the company's technological breakthrough in this field.

[0077] The keyword extraction results are: ['A', 'integrated circuit', 'optoelectronic chip', '1.25G to 100G', 'laser driver', 'transimpedance amplifier', 'clock and data recovery', 'laser driver chip', 'Super transimpedance amplifier'].

[0078] The node words (node) and synonyms (syn) that meet the requirements are: {'syn': 'optoelectronic chip', 'node': 'optoelectronic chip'}, {'syn': 'photonic chip', 'node': 'optoelectronic chip'}, {'syn': 'optoelectronic chip','node': 'optoelectronic chip'},{'syn': 'amplifier chip', 'node': 'amplifier chip'}, {'syn': 'amplifier', 'node': 'amplifier chip'}, {'syn': 'fiber laser amplifier', 'node': 'fiber laser'}].

[0079] The final labels are: optoelectronic chip, amplifier chip, fiber laser.

[0080] The associated results are: integrated circuit-midstream-chip manufacturing-analog integrated circuit-optoelectronic chip, integrated circuit-midstream-chip manufacturing-analog integrated circuit-amplifier chip, integrated circuit-midstream-chip manufacturing-analog integrated circuit-fiber laser.

[0081] Embodiment 2 In one embodiment, see the attached Figure 3 As shown, the present invention also discloses an enterprise industry chain association system, including a storage module, a first expansion module, a second expansion module, a label extraction module and an association module.

[0082] The storage module stores the basic graph of the industrial chain, and the first expansion module is used to use the first language model to perform subsequent hierarchical expansion on the basic graph of the industrial chain to form and store the ultimate graph of the industrial chain. The second expansion module expands the nodes in the ultimate graph of the industrial chain through the first language model to form and store a professional vocabulary. The label extraction module is used to obtain text data of different dimensions of the enterprise and extract enterprise labels from the text data. The association module is used to associate enterprise labels with the industrial chain and return the links related to the nodes corresponding to the enterprise labels.

[0083] Embodiment 3 In one embodiment, the present invention also discloses that the present application also provides a computer-readable storage medium, on which instructions are stored, and when the execution instructions are executed by a processor, they are used to implement the enterprise industrial chain association method in Example 1.

[0084] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0085] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the present application may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0086] The above implementation modes are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with this technology to understand the content of the present invention and implement it, and they cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for associating an enterprise industrial chain, characterized by: include: Define a basic graph of the industrial chain, wherein the basic graph of the industrial chain includes a plurality of industrial chain levels, and each of the industrial chain levels includes a plurality of nodes; Using the first language model to expand the basic graph of the industrial chain to subsequent levels, forming the ultimate graph of the industrial chain; Expanding the node words of the ultimate graph nodes of the industrial chain through the first language model to form a professional vocabulary; Acquire text data of different dimensions of the enterprise, and extract enterprise labels from the text data; The enterprise tag is associated with the ultimate graph of the industrial chain, and the link related to the node corresponding to the enterprise tag is returned.

2. The enterprise industrial chain association method according to claim 1, characterized in that: Extracting enterprise tags from the text data specifically includes: Extracting keywords from the text data using the first language model and the second model; Calculate the similarity between the extracted keywords and the words in the professional vocabulary, and obtain the word in the professional vocabulary that is closest to the text data as a label; The labels obtained by the first language model and the second model are aggregated, and the repeated labels are taken as the final enterprise labels.

3. The enterprise industrial chain association method according to claim 2, characterized in that: The first language model is an LLM model, and the LLM model specifically extracts the label by: The LLM model extracts keywords from the text data; Calculate the similarity between the extracted keywords and the node words and synonyms in the professional vocabulary to obtain the node words and synonyms that meet the requirements; Compare the similarity with a set threshold, and use the node words or synonyms whose similarity is greater than the set threshold as the most similar professional words; Returning the professional word and the text data to the LLM model, the LLM model determining whether the professional word is a word expressed in the text data; If so, keep the professional term as a label.

4. The enterprise industrial chain association method according to claim 2, characterized in that: The second model is a KeyBERT model, and the KeyBERT model specifically extracts the label including: The KeyBERT model is embedded in a professional vocabulary, and weights are set for node words and synonyms in the professional vocabulary; Set the number of words to be extracted and perform keyword extraction; Calculate the similarity between the extracted keywords and the node words and synonyms in the professional vocabulary to obtain the node words and synonyms that meet the requirements; The similarity is compared with a set threshold, and the node words or synonyms with similarity greater than the set threshold are used as labels.

5. The enterprise industrial chain association method according to claim 1, characterized in that: The first language model is used to expand the basic graph of the industrial chain to subsequent levels to form the ultimate graph of the industrial chain, which specifically includes: The first language model is used to expand the basic graph of the industrial chain to form an intermediate graph of the industrial chain; The intermediate map of the industrial chain is verified and corrected to form the ultimate map of the industrial chain.

6. The enterprise industrial chain association method according to claim 5, characterized in that: The first language model is used to expand the basic graph of the industrial chain to form an intermediate graph of the industrial chain, which specifically includes: Select all nodes of the last level of the industrial chain in the industrial chain basic graph; Use the first language model to expand the node words of all nodes at the last industrial chain level and output the results; The output results are collected and mapped to relevant nodes.

7. The enterprise industrial chain association method according to claim 1 is characterized by: The node words of the ultimate graph node of the industrial chain are expanded by the first language model to form a professional vocabulary, which specifically includes: Synonyms are expanded through the first language model to identify synonyms and related concepts of node words at each node of the ultimate graph of the industrial chain, and the expanded synonyms are mapped to corresponding nodes. The synonyms and node words form a professional vocabulary.

8. The enterprise industrial chain association method according to claim 1, characterized in that: The basic map of the industrial chain includes three levels of the industrial chain, and only the upstream and midstream industries are retained in the basic map of the industrial chain.

9. An enterprise industrial chain association system, characterized by: The enterprise industrial chain association method according to any one of claims 1 to 8 is adopted, and the system includes: A storage module storing a basic graph of the industrial chain; A first expansion module, wherein the first expansion module is used to expand the basic graph of the industrial chain by using the first language model to a subsequent level, so as to form and store the ultimate graph of the industrial chain; A second expansion module, which expands the nodes in the ultimate graph of the industrial chain through the first language model to form a professional vocabulary and store it; A label extraction module, which is used to obtain text data of different dimensions of an enterprise and extract enterprise labels from the text data; The association module is used to associate the enterprise tag with the industrial chain and return the link related to the node corresponding to the enterprise tag.

10. A computer-readable storage medium, characterized in that: Instructions are stored on a computer-readable storage medium, and when the execution instructions are executed by a processor, they are used to implement the enterprise industrial chain association method described in any one of claims 1-9.

Images