An industry development multi-dimensional collaborative evaluation method based on an artificial intelligence large model

Abstract

The application relates to the technical field of artificial intelligence, and discloses an industry development multi-dimensional collaborative evaluation method based on an artificial intelligence large model, which positions an industry ecological circle to which a target enterprise belongs by constructing an industry ecological circle classifier, extracts technical posts and market posts in post information of the enterprise by using a large model, respectively uses the large model to perform technical coverage field analysis and industry layout analysis in combination with a technical field category and an industry category, and finally, performs multi-dimensional collaborative evaluation of industry development in combination with industry ecological circle positioning, technical coverage field analysis and industry layout analysis corresponding to all target enterprises, so as to assist researchers in analyzing the industry ecological circle and solve the technical problem of high evaluation cost and low evaluation efficiency in the prior art in which artificial evaluation is performed on the industry ecological circle and enterprises in the industry ecological circle for industry development evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model. BACKGROUND

[0002] In the industrial development policy of each city, the core industries of the local area are classified into several industrial ecosystems, and each industrial ecosystem is further divided into multiple industrial chains. For example, the industrial chains included in the electronic information industry circle include integrated circuits, new displays, intelligent terminals, etc., and the industrial chains included in the digital economy industry circle include high-end software and operating systems, big data and artificial intelligence (including vehicle intelligent control systems), industrial internet, satellite internet and satellite applications, financial technology, etc.

[0003] A company generally belongs to an industrial ecosystem in a local city, and its technology system covers multiple core technologies, and its business is also distributed in multiple industries. At present, researchers manually evaluate the industrial development of industrial ecosystems and their companies based on their own experience and knowledge level, as well as data of companies in the industrial ecosystem, which requires high experience and knowledge reserves of researchers and is difficult to cope with the growing amount of industrial ecosystem data and to perform multi-dimensional evaluation. Therefore, a method for intelligent evaluation from multiple dimensions based on industrial ecosystem data is needed to assist researchers in analyzing industrial ecosystems. SUMMARY

[0004] The purpose of the present application is to provide a method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model, which positions the industrial ecosystem to which the target enterprise belongs by constructing an industrial ecosystem classifier, extracts technical and market positions from the enterprise's job information using a large model, combines technical field categories and industry categories, and uses a large model to analyze the technical coverage field and industry layout, respectively. Finally, the multi-dimensional collaborative evaluation of industrial development is performed by combining the industrial ecosystem positioning, technical coverage field analysis and industry layout analysis of all target enterprises, to assist researchers in analyzing industrial ecosystems and solve the technical problems of high evaluation cost and low evaluation efficiency in the prior art due to manual evaluation of industrial ecosystems and their companies.

[0005] The present application is realized by the following technical solutions:

[0006] The present application discloses a method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model, comprising the following steps:

[0007] According to the brief text of the target enterprise, a general text embedding large model is used to generate a vector representation of the target enterprise and input a pre-trained industrial ecosystem classifier to obtain the industrial ecosystem to which the target enterprise belongs.

[0008] Based on the job descriptions of the target company, a large dialogue model is used to extract technical and marketing positions from the target company's various positions.

[0009] Based on the category to which each technical position belongs in the technical field category set, the analysis results of the technical coverage area corresponding to the target enterprise are obtained by using a large dialogue model and a general text embedding large model. The technical field category set is configured to be generated by a large long text dialogue model after clustering based on the technical position expression vectors corresponding to the technical positions extracted from the job descriptions of each target enterprise.

[0010] Based on industry category sets, introductory texts of target companies, and market positions, a large dialogue model is used to obtain industry layout analysis results for the target companies.

[0011] Based on the analysis results of the target company's technology coverage area, industry layout, and industrial ecosystem, the Gini coefficient is used to calculate the technology coverage and industry layout of each target company in each industrial ecosystem.

[0012] Based on the technological scope and industry layout of each target enterprise in each industrial ecosystem, a visualized multi-dimensional collaborative evaluation result of industrial development is generated for each industrial ecosystem.

[0013] To better realize the present invention, the method for constructing the pre-trained industrial ecosystem classifier further includes the following steps:

[0014] Based on the first prompt word in the preset prompt word list, the dialogue big model is used to generate descriptive text for each industry ecosystem;

[0015] Based on the descriptive text corresponding to each industrial ecosystem, a set of vector representations of industrial ecosystems is generated using a general text embedding model.

[0016] Based on the collection of profiles of leading companies in each industry ecosystem, a set of vector representations of leading companies is generated using a general text embedding model.

[0017] Merge the vector representation set of the industrial ecosystem and the vector representation set of the chain leader enterprises to generate the training dataset for the industrial ecosystem classifier;

[0018] The feedforward neural network is trained using the training dataset to obtain an industrial ecosystem classifier.

[0019] In order to better realize the present application, further, the ecological loss function is used as the loss function of the industrial ecological classifier, the ecological loss function is configured to amplify the difference of the vector expression of different industrial ecological in the classifier space, and to make the vector expression of each chain master enterprise closer to the vector expression of the industrial ecological to which it belongs.

[0020] In order to better realize the present application, further, the ecological loss function is composed of a cross-entropy loss function and an industrial ecological discriminant enhancement loss function, and its expression is L=L ce +L ge , wherein L ce is the cross-entropy loss function, L ge is the industrial ecological discriminant enhancement loss function.

[0021] The expression of the cross-entropy loss function L ce is , wherein y is a one-hot encoded true label vector, is a probability distribution vector predicted by a feedforward neural network.

[0022] The expression of the industrial ecological discriminant enhancement loss function L ge is , wherein, f i is the vector expression of the industrial ecological , f j is the vector expression of the industrial ecological , q i,j is the jth chain master enterprise in the ith industrial ecological, is the Euclidean distance to calculate the similarity of two vectors, is the third layer feature vector extracted by the feedforward neural network.

[0023] In order to better realize the present application, further, the method for constructing the technical field category set comprises:

[0024] According to the second prompt word in the preset prompt word list and the collected job description of each target enterprise, a technical position is extracted from each target enterprise using a dialogue large model;

[0025] A general text embedding large model is used to generate a technical position expression vector corresponding to each technical position;

[0026] All technical position expression vectors are clustered to obtain a plurality of clusters representing a technical field category as a technical field vector set;

[0027] According to a third prompt word in the preset prompt word list, the super-long text dialogue large model is used to summarize and induce the large-scale post descriptions corresponding to each cluster in the technical field vector set, to generate the technical field category name corresponding to each cluster, and to generate the technical field category set corresponding to each cluster.

[0028] To better implement the present application, further, based on the categories to which each technical post belongs in the technical field category set, the method for obtaining the technical coverage field analysis result corresponding to the target enterprise using the dialogue large model and the general text embedding large model comprises:

[0029] According to the collected post descriptions of each target enterprise and a second prompt word in the preset prompt word list, the dialogue large model is used to extract technical posts from each post of each target enterprise;

[0030] The general text embedding large model is used to generate technical post expression vectors corresponding to each technical post;

[0031] The most similar calculation is performed by calculating the similarity of each technical post expression vector and each technical field category in the technical field category set, respectively, to obtain the technical field category with the highest similarity to each technical post expression vector in the technical field category set, so as to determine the technical field category corresponding to each technical post in the target enterprise;

[0032] The technical field categories corresponding to all technical posts in the target enterprise are taken as the technical coverage field analysis result corresponding to the target enterprise.

[0033] To better implement the present application, further, based on the industry category set, the brief text of the target enterprise and the market post, the method for obtaining the industry layout analysis result corresponding to the target enterprise using the dialogue large model comprises:

[0034] According to the collected post descriptions of each target enterprise and a fourth prompt word in the preset prompt word list, the dialogue large model is used to extract market posts from each post of each target enterprise;

[0035] According to the market post, the brief text of the target enterprise and a fifth prompt word in the preset prompt word list, the dialogue large model is used to obtain a plurality of industries in which the target enterprise is located as the industry layout analysis result corresponding to the target enterprise, wherein the fifth prompt word is constructed from the industry category set.

[0036] To better implement the present application, further, the calculation expression of the technical field coverage degree is wherein U is the number of technical fields covered in the technical coverage field analysis result of the target enterprise, K is the number of technical posts of the target enterprise, K mK is the number of technical posts of the mth technical field involved in the target enterprise n K is the number of technical posts of the nth technical field involved in the target enterprise

[0037] The calculation expression of the industry layout degree is , wherein V is the number of industries laid out in the industry layout analysis result of the target enterprise, D is the number of market posts of the target enterprise, D m D is the number of market posts of the mth industry laid out by the target enterprise n D is the number of market posts of the nth industry laid out by the target enterprise.

[0038] In order to better realize the present application, further, the visualized multi-dimensional collaborative evaluation result of industrial development includes a technical-industry joint distribution ITSD evaluation result graph of an industrial ecosystem and an industrial development trend IDT evaluation graph;

[0039] The technical-industry joint distribution ITSD evaluation result graph of the industrial ecosystem is configured with the technical field coverage degree as the vertical coordinate and the industry layout degree as the horizontal coordinate, and is divided into four quadrants according to the midpoints of the horizontal coordinates and the midpoints of the vertical coordinates, and the target enterprises in the same quadrant are divided into the same type of enterprises;

[0040] The industrial development trend IDT evaluation graph is configured with time as the horizontal coordinate and the number of at least one type of enterprises in the technical-industry joint distribution ITSD evaluation result graph of the industrial ecosystem as the vertical coordinate.

[0041] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0042] The use of artificial intelligence large model technology for accurate classification of the industrial ecosystem to which the enterprise belongs, extraction of key data such as technical posts and market posts, and accurate analysis results of technical coverage fields and industry layout greatly improves the efficiency of individual analysis of each enterprise in the industrial ecosystem. Through multi-dimensional collaborative evaluation of the industrial ecosystem (spatial dimension), technical field coverage degree (innovation dimension) and industry layout degree (economic dimension), the comprehensive analysis result of a certain industrial ecosystem is accurately obtained. BRIEF DESCRIPTION OF DRAWINGS

[0043] The present application is further illustrated in combination with the following drawings and examples, and all the conceptual innovations of the present application should be regarded as the disclosed content and the protection scope of the present application.

[0044] Figure 1 FIG. 1 is a flowchart of embodiment 1 of the industrial development multi-dimensional collaborative evaluation method based on the artificial intelligence large model in the present application.

[0045] Figure 2A flowchart of a preferred implementation of embodiment 1 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0046] Figure 3 A flowchart of constructing training data for an industrial ecosystem classifier in embodiment 2 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0047] Figure 4 A schematic diagram of an industrial ecosystem classifier in embodiment 2 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0048] Figure 5 A flowchart of positioning an enterprise in an industrial ecosystem in embodiment 2 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0049] Figure 6 A flowchart of constructing a technology field category in embodiment 3 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0050] Figure 7 A flowchart of analyzing the technology coverage of an enterprise in embodiment 3 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0051] Figure 8 A flowchart of analyzing the industry layout of an enterprise in embodiment 3 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0052] Figure 9 A schematic diagram of the technology-industry joint distribution ITSD in embodiment 3 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application.

[0053] Figure 10 A schematic diagram of the industrial development trend IDT in embodiment 3 of the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model in the present application. DETAILED DESCRIPTION

[0054] Embodiment 1

[0055] As shown in the following table, the method for multi-dimensional collaborative evaluation of industrial development based on a large artificial intelligence model includes the following steps: Figure 1

[0056] ​According to the brief text of the target enterprise, a general text embedding large model is used to generate a vector representation of the target enterprise and input a pre-trained industry ecosystem classifier to obtain the industry ecosystem to which the target enterprise belongs;

[0057] According to the job descriptions of the target enterprise, a dialogue large model is used to extract technical positions and market positions from the target enterprise.

[0058] Based on the categories to which each technical position belongs in the technical field category set, a dialogue large model and a general text embedding large model are used to obtain the technical coverage field analysis result corresponding to the target enterprise, wherein the technical field category set is configured to be generated by a super-long text dialogue large model after clustering according to the technical position expression vectors corresponding to the technical positions extracted from the job descriptions of each target enterprise.

[0059] Based on the industry category set, the brief text of the target enterprise and the market position, a dialogue large model is used to obtain the industry layout analysis result corresponding to the target enterprise.

[0060] According to the technical coverage field analysis result, the industry layout analysis result and the industry ecosystem to which the target enterprise belongs, the Gini coefficient is used to calculate the technical field coverage and industry layout of each target enterprise in each industry ecosystem.

[0061] According to the technical field coverage and industry layout of each target enterprise in each industry ecosystem, a visual industry development multidimensional synergy evaluation result corresponding to each industry ecosystem is generated.

[0062] Specifically, in this embodiment, if the industrial development multidimensional synergy evaluation of A city is to be performed, the brief text data of each enterprise and the job description text data corresponding to each enterprise on each recruitment software in the case of positioning A city are collected, and the enterprises that have collected data are the target enterprises.

[0063] In the embodiment, for each type of artificial intelligence large model, a selection can be made according to the demand. In a preferred embodiment, the dialogue large model is GLM-4, and the general text embedding large model is BGE-M3. GLM-4 has the characteristics of deep semantic understanding and 128K context task decomposition, and can better handle the diversity of job description expression and judge the subdivision field in combination with industry common sense. BGE-M3 has the characteristics of multi-vector fusion architecture, 8192 Token long text support, and cross-language semantic alignment. Its multi-vector fusion architecture can generate dense vectors, sparse vectors and multi-vectors simultaneously, avoiding the loss of key information by a single vector. The 8192 Token long text support can retain the key paragraphs of the full text and avoid information loss caused by Token truncation. The cross-language semantic alignment can process 100+ languages in a unified vector space, allowing consistent representation of multilingual terms in a multinational enterprise introduction and avoiding retrieval bias caused by incompatible English and Chinese vectors.

[0064] In addition, in the embodiment, the prompt words required by the large model are set in the same prompt word list for calling.

[0065] Due to the diversity of technical fields, the technical fields involved by cities and enterprises differ greatly. In the embodiment, the technical field categories involved by the industry of the city are obtained by analysis using large models and clustering tools.

[0066] In the embodiment, the technical coverage analysis result includes the types and quantities of technical fields covered by the enterprise and the number of technical positions in each field. The industry layout analysis result includes the types and quantities of industries in which the enterprise is located and the number of market positions in each industry.

[0067] The technical field coverage and industry layout degree are calculated by the Gini coefficient, which can quantify the distribution of the enterprise's technology and the distribution of the enterprise's industry layout, respectively. The greater the value of the technical field coverage, the wider the coverage of the enterprise in different technical fields, and the smaller the value, the more obvious the focus of the enterprise's technical innovation. The greater the value of the industry layout degree, the wider the layout of the enterprise in different industries, and the smaller the value, the more concentrated the enterprise in the industry layout. By further dividing and classifying the enterprises with similar technical field coverage and industry layout degree values, different types of enterprises can be further obtained. For example, in a graph, the vertical coordinate is set as the technical field coverage, and the horizontal coordinate is set as the industry layout degree. The enterprises in the same quadrant are divided into the same type of enterprise by manually dividing the quadrants, such as dividing them into four quadrants according to the midpoint of the horizontal coordinate and the midpoint of the vertical coordinate. The enterprises in the same cluster are divided into the same type of enterprise by circling multiple clusters in the graph according to the distance between the points. After dividing the different types of enterprises, the number of each type of enterprise in a certain period can be displayed by taking time as the horizontal coordinate and the number of each type of enterprise as the vertical coordinate.

[0068] It is obvious that there is no order between the classification of the target enterprise's industrial ecosystem and the extraction of technical positions and market positions for technical coverage field analysis and industry layout analysis. Technical positions can be extracted separately to obtain technical coverage field analysis results, and market positions can be extracted separately to obtain industry layout analysis results. In a preferred embodiment, after inputting the data of the target enterprise, the three are processed in parallel, and the results are output respectively, and then the calculation of the technical field coverage and the industry layout is performed, as shown in Figure 2 An optional embodiment of the multi-dimensional collaborative evaluation method for industrial development based on artificial intelligence large models, the input is the introduction text data of each enterprise on each recruitment software and the corresponding job description text data of each enterprise in a city, including the following steps:

[0069] Step S1a, positioning of the target enterprise's industrial ecosystem; specifically, according to the introduction text of the target enterprise, using a general text embedding large model, generating a vector expression of the target enterprise and inputting a pre-trained industrial ecosystem classifier, obtaining the industrial ecosystem to which the target enterprise belongs;

[0070] Step S1b, target enterprise technology coverage field analysis; specifically, according to the job descriptions of the target enterprise, using a dialogue large model, extracting technical positions from the various positions of the target enterprise, based on the categories to which each technical position belongs in the technical field category set, using the dialogue large model and the general text embedding large model to obtain the target enterprise's corresponding technical coverage field analysis results, wherein the technical field category set is configured to be generated by a super-long text dialogue large model after clustering according to the technical position expression vectors corresponding to the technical positions extracted from the job descriptions of each target enterprise;

[0071] Step S1c, target enterprise industry layout analysis; specifically, according to the job descriptions of the target enterprise, using a dialogue large model, extracting market positions from the various positions of the target enterprise, based on the industry category set, the introduction text of the target enterprise and the market positions, using the dialogue large model to obtain the target enterprise's corresponding industry layout analysis results;

[0072] Step S2, calculation of the technical field coverage and industry layout of each industrial ecosystem; specifically, according to the target enterprise's technical coverage field analysis results, industry layout analysis results and the industrial ecosystem to which it belongs, using the Gini coefficient, respectively calculating the technical field coverage and industry layout of each target enterprise in each industrial ecosystem;

[0073] Step S3, visualizing the multi-dimensional collaborative evaluation results of the industrial development of each industrial ecosystem; specifically, generating the visualized multi-dimensional collaborative evaluation results of the industrial development of each industrial ecosystem according to the technology field coverage and industry layout degree of each target enterprise in each industrial ecosystem.

[0074] Embodiment 2

[0075] This embodiment is further optimized on the basis of the above-mentioned embodiment 1. In this embodiment, the method for constructing the pre-trained industrial ecosystem classifier comprises the following steps:

[0076] According to the first prompt word in the preset prompt word list, the dialog large model is used to generate description text corresponding to each industrial ecosystem;

[0077] According to the description text corresponding to each industrial ecosystem, a general text embedding large model is used to generate an industrial ecosystem vector expression set;

[0078] According to the chain master enterprise profile set in each industrial ecosystem, a general text embedding large model is used to generate a chain master enterprise vector expression set;

[0079] Merge the industrial ecosystem vector expression set and the chain master enterprise vector expression set to generate a training data set of the industrial ecosystem classifier;

[0080] The training data set is used to train the feedforward neural network to obtain the industrial ecosystem classifier.

[0081] Further, an ecosystem loss function is used as the loss function of the industrial ecosystem classifier, and the ecosystem loss function is configured to amplify the difference between the vector expressions of different industrial ecosystems in the classifier space, and to make the vector expressions of each chain master enterprise closer to the vector expression of the industrial ecosystem to which it belongs.

[0082] Further, the ecosystem loss function is composed of a cross-entropy loss function and an industrial ecosystem discrimination enhancement loss function, and its expression is L = L ce + L ge , wherein L ce is the cross-entropy loss function, and L ge is the industrial ecosystem discrimination enhancement loss function.

[0083] The expression of the cross-entropy loss function L ce is , wherein y is a one-hot encoded real label vector, is a probability distribution vector predicted by the feedforward neural network.

[0084] The industrial ecosystem discrimination enhancement loss function L geThe expression of the vector representation of the i-th industrial ecosystem is wherein, f i is the industrial ecosystem The vector representation of the i-th industrial ecosystem is f j is the industrial ecosystem The vector representation of the i-th industrial ecosystem is i,j is the j-th chain master enterprise in the i-th industrial ecosystem, is the Euclidean distance used to calculate the similarity between two vectors, is the third layer feature vector extracted by the feedforward neural network.

[0085] In this embodiment, in order to accurately classify enterprises according to their industrial ecosystems, the vector representation of the industrial ecosystem and the chain master enterprise of the industrial ecosystem is used to expand the training data samples and better represent the entire industrial ecosystem.

[0086] At the same time, as shown in Table 1, in this embodiment, the prompt words required by the large model are set in the same prompt word list for calling. The first to fifth prompt words in this and subsequent embodiments correspond to prompt word 1 to prompt word 5 in Table 1, respectively.

[0087] Table 1-Prompt word list

[0088]

[0089] Specifically, as shown in Figure 3 for the i-th industrial ecosystem , first use the dialogue large model GLM-4 to describe it in words through prompt word 1 in Table 1, and then use the general text embedding large model BGE-M3 to generate a 1024-dimensional vector f i for each industrial ecosystem description to complete the vector representation of the industrial ecosystem;

[0090] Similarly, as shown in Figure 3 for the i-th industrial ecosystem , obtain a set of N chain master enterprise profiles , and use the general text embedding large model BGE-M3 to obtain a set of chain master enterprise vector representations containing N 1024-dimensional vectors to complete the vector representation of the chain master enterprise of the industrial ecosystem; This is because using only the vector representation of the industrial ecosystem as the training set for the industrial ecosystem classifier, on the one hand, the training data is insufficient, and on the other hand, the representation is also difficult to well represent the entire industrial ecosystem, therefore, this embodiment also uses the vector representation of the chain master enterprise as the training set;

[0091] Finally, the vector representation of the industrial ecosystem and the vector representation set of chain master enterprises in the industrial ecosystem are merged, and the training data of the industrial ecosystem is constructed which contains N+1 vectors, each vector has a dimension of 1024, to complete the construction of the training data of the industrial ecosystem classifier.

[0092] In this embodiment, as shown in Figure 4 , an industrial ecosystem classifier is constructed using a general feedforward neural network, which contains 4 layers, the first layer has 1024 neurons for reading in training data with a dimension of 1024, the second and third layers have 256 and 64 neurons respectively, and the last layer has M neurons for outputting the category of the industrial ecosystem;

[0093] In general classification problems, only the cross-entropy loss function is used as the loss function of the classifier, that is , while in the data set constructed in this embodiment, the vector representations and of two different industrial ecosystems f i and f j have stronger differences, while the differences between the vector representations f i,n and f i,m of two chain master enterprises are relatively weak, so this embodiment further constructs an industrial ecosystem discrimination enhancement loss , which makes the difference between the two industrial ecosystems in the classifier space as large as possible, and makes the vector representation of each chain master enterprise in the classifier space closer to the vector representation of the industrial ecosystem it belongs to. In this loss function, is used to make the vector representations f i and f j of different industrial ecosystems and have greater differences in the classifier space, is used to make the vector representation of each chain master enterprise q i,j closer to the vector representation of the industrial ecosystem it belongs to in the classifier space f i ;

[0094] In addition, the training of the industrial ecosystem classifier uses the commonly used stochastic gradient descent method SGD (Stochastic Gradient Descent) method for training. ​​

[0095] In the classification using the industrial ecosystem classifier, as shown in Figure 5 for a company to be classified into an industrial ecosystem, its profile is a, and a 1024-dimensional vector q is extracted using the general text embedding large model BGE-M3, which is input into the trained industrial ecosystem classifier to obtain the industrial ecosystem to which the company belongs, and the positioning of the industrial ecosystem to which the company belongs is completed.

[0096] By constructing the training data set to train the industrial ecosystem classifier, and constructing the corresponding ecosystem loss function, the industrial ecosystem to which the target company belongs can be more accurately determined.

[0097] Example 3:

[0098] This embodiment is further optimized based on the above-mentioned embodiment 2. In this embodiment, the method for constructing the technical field category set comprises:

[0099] According to the second prompt word in the preset prompt word list and the collected job description of each target enterprise, the technical post is extracted from each target enterprise using the dialogue large model;

[0100] The general text embedding large model is used to generate a technical post expression vector corresponding to each technical post;

[0101] All technical post expression vectors are clustered to obtain a plurality of clusters representing a technical field category as a technical field vector set;

[0102] According to the third prompt word in the preset prompt word list, the large-scale job description corresponding to each cluster in the technical field vector set is summarized using the super-long text dialogue large model, and the technical field category name corresponding to each cluster is generated. And generate a technical field category set corresponding to each cluster.

[0103] Based on the category to which each technical post belongs in the technical field category set, the method for obtaining the technical coverage field analysis result corresponding to the target enterprise using the dialogue large model and the general text embedding large model comprises:

[0104] According to the second prompt word in the preset prompt word list and the collected job description of each target enterprise, the technical post is extracted from each target enterprise using the dialogue large model;

[0105] The general text embedding large model is used to generate a technical post expression vector corresponding to each technical post;

[0106] The most similar calculation is performed by calculating the similarity of each technical post expression vector and each technical field category in the technical field category set respectively, and the technical field category with the highest similarity of each technical post expression vector in the technical field category set is obtained to determine the technical field category corresponding to each technical post in the target enterprise;

[0107] The technical field categories corresponding to all technical posts in the target enterprise are taken as the technical coverage field analysis result corresponding to the target enterprise.

[0108] The method for obtaining the industry layout analysis result corresponding to the target enterprise based on the industry category set, the brief text of the target enterprise and the market post using the dialogue large model includes:

[0109] According to the fourth prompt word in the preset prompt word list and the collected post description of each target enterprise, the market post is extracted from each post of each target enterprise using the dialogue large model;

[0110] According to the market post, the brief text of the target enterprise and the fifth prompt word in the preset prompt word list, the dialogue large model is used to obtain a plurality of industries in the layout of the target enterprise as the industry layout analysis result corresponding to the target enterprise, wherein the fifth prompt word is constructed from the industry category set.

[0111] The calculation expression of the technical field coverage degree is , wherein U is the number of technical fields covered in the technical coverage field analysis result of the target enterprise, K is the number of technical posts of the target enterprise, K m is the number of technical posts of the target enterprise involved in the mth technical field, K n is the number of technical posts of the target enterprise involved in the nth technical field;

[0112] The calculation expression of the industry layout degree is , wherein V is the number of industries in the industry layout analysis result of the target enterprise, D is the number of market posts of the target enterprise, D m is the number of market posts of the target enterprise in the mth industry, D n is the number of market posts of the target enterprise in the nth industry.

[0113] Further, the visualized multi-dimensional collaborative evaluation result of industrial development includes the technical-industry joint distribution ITSD evaluation result graph of the industrial ecosystem and the industrial development trend IDT evaluation graph;

[0114] The technical-industry joint distribution ITSD evaluation result graph of the industrial ecosystem is configured with the technical field coverage degree as the vertical coordinate and the industry layout degree as the horizontal coordinate, is divided into four quadrants according to the midpoint of the horizontal coordinate and the midpoint of the vertical coordinate, and target enterprises in the same quadrant are classified into the same type of enterprise.

[0115] The industrial development trend IDT evaluation graph is configured with time as the horizontal coordinate and the number of at least one type of enterprise in the technical-industry joint distribution ITSD evaluation result graph of the industrial ecosystem as the vertical coordinate.

[0116] Specifically, due to the diversity of technical fields, the technical fields involved by various cities and various enterprises differ greatly, and the embodiment obtains the technical field categories involved by the industry of the city through analysis by a large model and clustering tools, as shown in Figure 6 , first, for each post of each enterprise collected, a dialogue large model GLM-4 is used to determine whether each post is a technical post through prompt word 2 in Table 1, for each technical post, a general text embedding large model BGE-M3 is used to generate a post expression vector with a dimension of 1024, and for all technical post expression vectors of enterprises, a density clustering method DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is used for clustering to obtain L clusters, and noise points not included in the clusters are ignored, each cluster represents a technical field category, and the post vector mean ti in the i th cluster is the vector expression of the i th technical field category, and the set of L vectors is ; at the same time, clustering forms a large number of post vectors and their corresponding post descriptions in each cluster, and therefore a large model KIMI supporting super-long text dialogue is used to summarize and induce a suitable technical field category name through prompt word 3 in Table 1 for each large-scale post description in each cluster, and the construction of the technical field category is completed.

[0117] In the embodiment, the industry category uses the national economic industry category set in China in 2017.

[0118] When analyzing the technical coverage of enterprises, as shown in Figure 7 , for each technical post of an enterprise, a dialogue large model GLM-4 is used to determine whether the post is a technical post through prompt word 2 in Table 1, for each technical post, a general text embedding large model BGE-M3 is used to generate a post expression vector t, and a most similar calculation is used to determine the expression of the technical field , The expression of the technical field The most similar technical field to the post expression vector t can be calculated from the L technical fields, wherein, , The similarity of two vectors is calculated as the Euclidean distance, and by calculating the technical field involved in each post of the enterprise, the U technical fields covered by the enterprise can be finally obtained.

[0119] When performing enterprise industry layout analysis, as shown in Figure 8 , first, for each post of the enterprise, use the dialogue large model GLM-4 to determine whether each post is a market post by using the prompt word 4 in Table 1, and then use the dialogue large model GLM-4 to finally obtain V industries of the enterprise layout by using the prompt word 5 in Table 1.

[0120] In this embodiment, after obtaining the technical coverage field analysis result, industry layout analysis result and the industry ecological circle to which the target enterprise belongs, the multi-dimensional coordination evaluation of the industry development of the industry ecological circle is performed, including the evaluation by using the industry-technology-sector distribution ITSD (Industry-Technology-Sector Distribution) evaluation method under the industry ecological circle, and the evaluation by using the industrial development trend IDT (Industrial development trend) evaluation method.

[0121] When the industry-technology-sector distribution ITSD evaluation method is used for evaluation, first, the industry ecological circle classifier is used to obtain the industry ecological circle to which each enterprise belongs, and then the Gini coefficient is used to calculate the technical field coverage and industry layout of each enterprise in each industry ecological circle. Finally, for each industry ecological circle, the technical field coverage TEC and the industry layout SEC of each enterprise in the industry ecological circle are calculated and plotted in a graph, and the industry-technology-sector distribution ITSD evaluation result graph of the industry ecological circle can be obtained, as shown in Figure 9 In this embodiment, the enterprises with TEC and SEC both less than 0.5 are divided into small and medium-sized enterprises; the enterprises with TEC and SEC both greater than 0.5 are large-scale head enterprises; the enterprises with TEC less than 0.5 and SEC greater than 0.5 are bottom technology enterprises, which cover few technical fields but involve many industries; the enterprises with TEC greater than 0.5 and SEC less than 0.5 are specialized, sophisticated and innovative enterprises, which focus on fewer industries through a wider technical field (here, the specialized, sophisticated and innovative enterprises are not the “specialized, sophisticated and innovative” enterprises in the project declaration, but are only used to call the enterprises with TEC greater than 0.5 and SEC less than 0.5); of course, referring to the description in Embodiment 1, the coordinates can be selected according to the actual situation of the industry ecological circle, such as TEC and SEC both less than 0.6, etc.

[0122] Further, for a certain industrial ecosystem, the number of small and micro enterprises, bottom technology enterprises, large-scale head enterprises and specialized, specialized and new enterprises at different times is counted, and a change graph is drawn, so that an industrial development trend IDT evaluation graph can be obtained, such as Figure 10 shown, is an example of the number development trend of small and micro enterprises and specialized, specialized and new enterprises of a certain industrial ecosystem.

[0123] By constructing the technology field category set, the present embodiment can avoid misjudgment caused by large differences in technology fields involved by various cities and enterprises, and more accurately analyze the technology coverage fields of enterprises, so as to obtain more accurate multi-dimensional collaborative evaluation results of industrial development.

[0124] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Any simple modification or equivalent change made on the basis of the technical essence of the present application to the above embodiment falls within the protection scope of the present application.

Claims

1. An artificial intelligence large model-based multi-dimensional collaborative evaluation method for industrial development, characterized in that, The method comprises the following steps: According to the brief text of the target enterprise, a general text embedding large model is used to generate a vector expression of the target enterprise and input a pre-trained industry ecosystem classifier to obtain the industry ecosystem to which the target enterprise belongs; According to the job descriptions of the target enterprise, a dialogue large model is used to extract technical positions and market positions from the target enterprise; Based on the categories to which each technical position belongs in the technical field category set, a dialogue large model and a general text embedding large model are used to obtain the technical coverage field analysis result corresponding to the target enterprise, wherein the technical field category set is configured to be generated by a super-long text dialogue large model after clustering according to the technical position expression vectors corresponding to the technical positions extracted from the job descriptions of each target enterprise; Based on the industry category set, the brief text of the target enterprise and the market position, a dialogue large model is used to obtain the industry layout analysis result corresponding to the target enterprise; According to the technical coverage field analysis result, the industry layout analysis result and the industry ecosystem to which the target enterprise belongs, a Gini coefficient is used to calculate the technical field coverage and the industry layout degree of each target enterprise in each industry ecosystem, respectively; According to the technical field coverage and the industry layout degree of each target enterprise in each industry ecosystem, a visual industry development multi-dimensional coordination evaluation result corresponding to each industry ecosystem is generated; An ecological circle loss function is used as the loss function of the industry ecosystem classifier, and the ecological circle loss function is configured to amplify the difference between the vector expressions of different industry ecosystems in the classifier space, and to make the vector expression of each chain master enterprise closer to the vector expression of the industry ecosystem to which it belongs; The ecological loss function is composed of a cross-entropy loss function and an industrial ecological circle discrimination enhancement loss function, and an expression thereof is wherein, is a cross-entropy loss function, is an industrial ecological circle discrimination enhancement loss function; the cross-entropy loss function is expressed as wherein, is a one-hot encoded true label vector, is a probability distribution vector predicted by the feedforward neural network; The industrial ecosystem discrimination enhancement loss function The expression is ,in, For the industrial ecosystem Vector representation, For the industrial ecosystem Vector representation, For the j-th leading enterprise in the i-th industrial ecosystem, Calculate the similarity between two vectors using Euclidean distance. This is the feature vector extracted from the third layer of the feedforward neural network.

2. The multi-dimensional collaborative evaluation method for industrial development based on artificial intelligence large model according to claim 1, characterized in that, The method for constructing the pre-trained industry ecosystem classifier comprises the following steps: According to a first prompt word in a preset prompt word list, a dialogue large model is used to generate description text corresponding to each industry ecosystem; According to the description text corresponding to each industry ecosystem, a general text embedding large model is used to generate an industry ecosystem vector expression set; According to the brief set of chain master enterprises in each industry ecosystem, a general text embedding large model is used to generate a chain master enterprise vector expression set; The industry ecosystem vector expression set and the chain master enterprise vector expression set are merged to generate a training data set of the industry ecosystem classifier; A feedforward neural network is trained using the training data set to obtain the industry ecosystem classifier.

3. The multi-dimensional collaborative evaluation method for industrial development based on artificial intelligence large model according to claim 1, characterized in that, The method for constructing the technical field category set comprises: According to the job descriptions of each target enterprise collected and a second prompt word in a preset prompt word list, a dialogue large model is used to extract technical positions from each target enterprise; A general text embedding large model is used to generate technical position expression vectors corresponding to each technical position; All technical position expression vectors are clustered to obtain a plurality of clusters representing a technical field category as a technical field vector set; According to the third prompt word in the preset prompt word list, the large-scale post description corresponding to each cluster in the technical field vector set is summarized using the ultra-long text dialogue large model, and the technical field category name corresponding to each cluster is generated, and the technical field category set corresponding to each cluster is generated.

4. The multi-dimensional collaborative evaluation method for industrial development based on artificial intelligence large model according to claim 1, characterized in that, Based on the categories to which each technical post belongs in the technical field category set, a method for obtaining a technical coverage field analysis result corresponding to a target enterprise using a dialogue large model and a general text embedding large model includes: According to the collected job descriptions of each target enterprise and the second prompt word in the preset prompt word list, the dialogue large model is used to extract technical posts from each target enterprise; The general text embedding large model is used to generate technical post expression vectors corresponding to each technical post; The most similar calculation is performed by calculating the similarity of each technical post expression vector and each technical field category in the technical field category set, respectively, to obtain the technical field category with the highest similarity to itself in the technical field category set for each technical post expression vector, so as to determine the technical field category corresponding to each technical post in the target enterprise. The technical field categories corresponding to all technical posts in the target enterprise are used as the technical coverage field analysis result corresponding to the target enterprise.

5. The multi-dimensional collaborative evaluation method for industrial development based on artificial intelligence large model according to claim 1, characterized in that, Based on the industry category set, the brief text of the target enterprise, and the market post, a method for obtaining an industry layout analysis result corresponding to a target enterprise using a dialogue large model includes: According to the collected job descriptions of each target enterprise and the fourth prompt word in the preset prompt word list, the dialogue large model is used to extract market posts from each target enterprise; According to the market post, the brief text of the target enterprise, and the fifth prompt word in the preset prompt word list, the dialogue large model is used to obtain multiple industries in which the target enterprise is located as the industry layout analysis result corresponding to the target enterprise, wherein the fifth prompt word is constructed from the industry category set.

6. The multi-dimensional collaborative evaluation method for industrial development based on the artificial intelligence large model according to claim 1, characterized in that: The calculation expression of the technical field coverage degree is Wherein, U is the number of technical fields covered in the technical field coverage analysis result of the target enterprise, K is the number of technical positions of the target enterprise, Km is the number of technical positions of the target enterprise involved in the mth technical field, Kn is the number of technical positions of the target enterprise involved in the nth technical field. The calculation expression of the industry layout degree is Wherein, V is the number of industries in the layout analysis result of the target enterprise, D is the number of market positions of the target enterprise, is the number of market positions of the mth industry in which the target enterprise is located, is the number of market positions of the nth industry in which the target enterprise is located.

7. The multi-dimensional collaborative evaluation method for industrial development based on the artificial intelligence large model according to claim 6, characterized in that: The visualized multi-dimensional collaborative evaluation result for industrial development includes an ITSD evaluation result graph of the technical-industry joint distribution of the industrial ecosystem and an IDT evaluation graph of the industrial development trend; The ITSD evaluation result graph of the technical-industry joint distribution of the industrial ecosystem is configured with the technical field coverage degree as the vertical coordinate and the industry layout degree as the horizontal coordinate, and is divided into four quadrants according to the midpoint of the horizontal coordinate and the midpoint of the vertical coordinate, and target enterprises in the same quadrant are classified into the same category; The IDT evaluation graph of the industrial development trend is configured with time as the horizontal coordinate and the number of at least one category of enterprises in the ITSD evaluation result graph of the technical-industry joint distribution of the industrial ecosystem as the vertical coordinate.

Images