Enterprise information retrieval method based on artificial intelligence

Abstract

The invention discloses an artificial intelligence-based enterprise information retrieval method, which comprises the following steps of: comprehensively collecting and preprocessing public text data information of an enterprise, and constructing an enterprise relationship network by utilizing an NER model and a deep learning neural network model; then, accurately identifying a retrieval intention of a retrieval enterprise by utilizing a CNN model, and converting the retrieval intention into a feature vector; secondly, organically fusing the enterprise relation network with the requirements of retrieval enterprises, and setting tendency weights for different retrieval intentions according to historical retrieval behaviors, so that the flexibility and practicability of the system are enhanced; besides, by constructing a three-layer multi-task learning model, information sharing and characteristic learning among different retrieval tasks are achieved, and the overall performance of the model is improved; and finally, designing a cooperation feasibility value output node, and calculating and outputting a cooperation feasibility value, thereby providing a visual and quantifiable cooperation decision basis for enterprise retrieval, and effectively improving the accuracy and practicability of enterprise information processing and cooperation evaluation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an enterprise information retrieval method based on artificial intelligence. Background Art

[0002] Enterprise information retrieval is an important way to obtain key data such as business conditions, credit records, legal proceedings, intellectual property rights, etc. At present, traditional enterprise information retrieval mainly relies on online platforms provided by commercial registration agencies, but these platforms have problems such as delayed information updates and complex retrieval interfaces. At the same time, information from commercial credit reporting agencies usually requires payment, which makes it difficult for small businesses or individual industrial and commercial households to obtain comprehensive information. With the development of the Internet, search engine technology has been applied to enterprise information retrieval, but the search results may be uncertain and inaccurate. The query platforms on the market require users to enter clear enterprise information such as name, name or code of legal executives for query, which has a high threshold for use and requires a lot of time to screen information.

[0003] The invention patent with patent application number 202410391523.7 discloses an enterprise information retrieval method and device based on artificial intelligence. It performs entity analysis on enterprise information through a preset named entity recognition model, extracts relationships from the results of the entity analysis through a deep learning neural network model, and constructs an enterprise relationship network; trains the preset convolutional neural network model according to the user's historical retrieval data, and obtains the corresponding user retrieval intent based on the enterprise information retrieval request currently sent by the user and the convolutional neural network model; performs feature fusion on the enterprise relationship network and the user's retrieval intent, and inputs the features obtained after feature fusion into the set multi-task learning model to obtain the enterprise retrieval information output by the multi-task learning model.

[0004] Existing NER and relation extraction technologies still have some problems in practical applications, such as insufficient model generalization ability and low relation extraction accuracy. In addition, when enterprises conduct information retrieval, they often have clear retrieval intentions, such as understanding the financial status of a company, assessing cooperation risks, etc. How to accurately capture and understand the retrieval intentions of enterprises and provide targeted information based on the retrieval intentions is another problem that needs to be solved. Summary of the invention

[0005] This application provides an enterprise information retrieval method based on artificial intelligence. By collecting the public text data information of the enterprise, the NER model and the deep learning neural network model are used to perform entity recognition and relationship extraction to build an enterprise relationship network. At the same time, combined with the retrieval intention characteristics of the enterprise, a multi-task learning model is designed to achieve comprehensive and accurate extraction and analysis of enterprise information, providing strong support for the enterprise's cooperative decision-making.

[0006] This application provides an enterprise information retrieval method based on artificial intelligence, including: S1, collect the public text data of enterprises, perform entity recognition and relationship extraction on enterprise information through NER model and deep learning neural network model, and then build an enterprise relationship network; the enterprise relationship network is to build an enterprise relationship network diagram based on the results of relationship extraction, in which nodes represent enterprises and edges represent the relationship between enterprises; S2, when the searching enterprise sends an enterprise information search request, the request text is converted into a vector representation, input into the trained CNN model, and the search intention feature vector of the searching enterprise is output; S3, merge the feature vectors in the enterprise relationship network and the search intent of the searched enterprise, and set the tendency weight for the search intent of the searched enterprise; S4, construct a multi-task learning model including a task sharing layer, a separate task layer and an output layer; S5, use the labeled dataset to train the multi-task learning model; S6, design a cooperation feasibility value output node in the output layer of the multi-task learning model, input the result after feature fusion into the multi-task model, and obtain the cooperation feasibility value.

[0007] Preferably, the output of the search intent feature vector of the searching enterprise specifically includes: training the CNN model based on the historical search data of the searching enterprise, and when the searching enterprise currently sends each enterprise information search request, inputting the request into the trained CNN model to output the search intent of the searching enterprise.

[0008] Preferably, in said S3, the tendency weight includes: the tendency weight is set for different search intentions according to the historical search intentions of the searched enterprise.

[0009] Preferably, in S6, obtaining the cooperation feasibility value includes: The formula for obtaining the cooperation feasibility value is as follows:

[0010] represents the output value corresponding to the i-th retrieval task, It represents the tendency weight of the i-th search intention feature, and outputs the calculated cooperation feasibility value to the searching enterprise.

[0011] Preferably, the S1 further includes: S11, construct and update the enterprise relationship network diagram, and explicitly add the cooperation relationship between the first-level cooperative enterprise and the second-level cooperative enterprise of the search enterprise to the enterprise relationship network diagram; wherein the first-level cooperative enterprise is the enterprise that checks the cooperation feasibility value with the search enterprise, and the second-level enterprise is the associated enterprise that cooperates with the first-level enterprise; S12, when the search enterprise sends an enterprise information search request, the system first identifies the cooperation relationship between the search enterprise and the first-level cooperative enterprise, and then identifies the cooperation relationship between the first-level cooperative enterprise and the second-level cooperative enterprise through the first-level cooperative enterprise as a jumping point; S13, for the feasibility assessment of the second cooperation between the search enterprise and the first-level cooperative enterprise, the feature fusion method of S6 is continued to be used to fuse the feature vector in the enterprise relationship network with the feature vector of the search intention of the search enterprise; S14, added the identification of cooperative paths to the multi-task learning model and trained the model using the updated dataset; S15, using the upgraded multi-task learning model, calculates the feasibility of the second cooperation between the retrieved enterprise and the secondary cooperative enterprise; S16, setting the second cooperation feasibility value P within a threshold range, evaluating and judging the second cooperation feasibility according to the second cooperation feasibility value P, and pushing secondary cooperation enterprises whose second cooperation feasibility is higher than the threshold to the searching enterprise.

[0012] Preferably, the S14 includes: the cooperation path is the search enterprise→the first-level cooperation enterprise→the second-level cooperation enterprise; and the updated data set is the cooperation information between the search enterprise and the first-level cooperation enterprise, and between the first-level cooperation enterprise and the second-level cooperation enterprise.

[0013] Preferably, the step S15, calculating the feasibility of the second cooperation between the search enterprise and the secondary cooperative enterprise, includes: for the evaluation of the feasibility of the second cooperation between the search enterprise and the secondary cooperative enterprise, it is necessary to integrate the cooperation feature vector between the search enterprise and the primary cooperative enterprise, and the cooperation feature vector between the primary cooperative enterprise and the secondary cooperative enterprise, specifically: let F 检索企业-一级 To retrieve the fusion feature vector of the enterprise and the first-level cooperative enterprise, let F 一级-二级 is the fusion feature vector of the first-level enterprise and the second-level cooperative enterprise, then the fusion feature vector of the core enterprise and the second-level cooperative enterprise is F 检索企业-二级 =αF 检索企业-一级 + β F 一级-二级 , where α and β are weight coefficients, which respectively represent the importance of the cooperative relationship between the search enterprise and the first-level cooperative enterprise, and between the first-level cooperative enterprise and the second-level cooperative enterprise; the second cooperation feasibility value P 检索企业-二级 The calculation formula is:

[0014] represents the output value corresponding to the i-th evaluation task, represents the weight of the i-th evaluation task.

[0015] Preferably, in S14, the newly added cooperation path in S14 further includes: S141, when the searching enterprise conducts multiple enterprise searches, the search results with the second cooperation feasibility value P greater than the threshold are added as new cooperation paths to the network diagram to form multiple cooperation paths; multiple cooperation paths = (enterprise search → first-level cooperation enterprise i → second-level cooperation enterprise i → ... → n-level cooperation enterprise), each cooperation path starts from the searching enterprise and extends step by step through the first-level cooperation enterprises; S142, identifying multi-level related enterprises based on multiple cooperation paths of the searched enterprises; S143, for the relationship between adjacent levels in the multi-level associated enterprises, calculate the cooperation feature vector between the adjacent levels and the search enterprise, and merge the cooperation feature vectors between the search enterprise and the multi-level associated enterprises; S144, determining the multi-level cooperation degree between the search enterprise and the multi-level related enterprises, sorting the multi-level cooperation degrees in descending order, and calculating the cooperation values ​​between the search enterprise and the multi-level cooperation degree enterprises in sequence; S145, adding the cooperation value between the retrieval enterprise and the multi-level cooperation enterprise to the upgraded multi-task learning model, and outputting the cooperation value between the retrieval enterprise and the multi-level cooperation enterprise.

[0016] Preferably, in S144, the cooperation value includes: The formula for cooperation value is as follows:

[0017] To retrieve the cooperation feature vector between an enterprise and its i-th level related enterprise, is the multi-level cooperation degree of a company in the i-th level of related enterprises, To retrieve the cooperation value between the enterprise and the i-th level related enterprise.

[0018] Preferably, it includes: S61, obtaining the cooperation value between the search enterprise and the multi-level cooperation degree enterprises, and calibrating the cooperation feasibility value; S62, calibrating and calculating the cooperation value and the cooperation feasibility value to obtain a calibrated cooperation feasibility value; wherein the calibrated cooperation feasibility value=R(cooperation feasibility value, cooperation value), R represents the selection of a function type, and weighted addition can be selected for calculation; S63, adding the calibrated cooperation feasibility value to the upgraded multi-task learning model; S64, using the updated multi-task learning model, outputs a calibrated cooperation feasibility value.

[0019] One or more technical solutions provided in this application have at least the following technical effects or advantages: By comprehensively and deeply collecting and preprocessing the public text data information of enterprises, a solid foundation is laid for subsequent entity recognition and relationship extraction, greatly improving the accuracy and comprehensiveness of information processing. At the same time, the advanced CNN model is used to accurately identify the search intent of the search enterprise, so that the system can more accurately capture the needs of the search enterprise and provide highly personalized information services. By organically integrating the enterprise relationship network with the needs of the search enterprise and setting reasonable tendency weights, the model can more flexibly adapt to the preferences and needs of different search enterprises, enhancing the adaptability and practicality of the system. In addition, the constructed multi-task learning model realizes information sharing and feature learning between different search tasks, which not only improves the overall performance and efficiency of the model, but also enables the system to perform well on multiple search tasks at the same time. Finally, by designing the cooperation feasibility value output node and calculating the cooperation feasibility value, an intuitive and quantifiable basis for cooperation decision-making is provided for the search enterprise, which greatly reduces the risk and uncertainty of decision-making and significantly improves the accuracy and practicality of enterprise information processing and cooperation evaluation.

[0020] By introducing the jump point of cooperative relationships, the scope of the enterprise cooperation network is effectively expanded, allowing enterprises to discover more potential partners. At the same time, the solution improves the accuracy and reliability of cooperation possibility assessment by integrating multi-dimensional feature vectors and upgrading the multi-task learning model. This refined evaluation method not only takes into account direct cooperative relationships, but also explores indirect cooperation opportunities, providing enterprises with more diversified cooperation options. In addition, the solution can also intelligently push enterprises with a higher possibility of cooperation, providing strong support for enterprises' cooperation decisions, greatly improving cooperation efficiency and success rate.

[0021] By searching multiple companies and building a multi-level network of related companies, the solution can comprehensively and intuitively display the complex relationship between the searched companies and their potential cooperative companies. At the same time, by calculating the cooperation feature vector and multi-level cooperation degree, the solution can quantitatively evaluate the cooperation closeness and cooperation potential between the searched companies and multi-level related companies, providing strong data support for the cooperation decision-making of the companies. In addition, the solution further optimizes the selection process of cooperative companies through the calculation of cooperation value, which helps to improve the cooperation efficiency and cooperation success rate of the companies.

[0022] By obtaining the cooperation value between the search enterprise and multi-level cooperation enterprises and calibrating the original cooperation feasibility value, the accuracy and comprehensiveness of the cooperation evaluation are significantly improved. The calibrated cooperation feasibility value is integrated into the upgraded multi-task learning model, so that the model can make fuller use of multi-source information for more accurate learning and prediction. This technical solution not only optimizes the evaluation process of cooperation potential, but also provides enterprises with a more reliable basis for cooperation decision-making, which helps enterprises to more accurately identify cooperation opportunities and optimize cooperation networks in a complex and changing market environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The present invention is a flowchart of an enterprise information retrieval method based on artificial intelligence. DETAILED DESCRIPTION

[0024] To facilitate the understanding of the present invention, the present application will be described more comprehensively below with reference to the relevant drawings; the drawings show preferred embodiments of the present invention, but the present invention can be implemented in many different forms and is not limited to the embodiments described herein; on the contrary, the purpose of providing these embodiments is to enable a more thorough and comprehensive understanding of the disclosed content of the present invention.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by technicians in the technical field to which the present invention belongs; the terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention; the term "and / or" used herein includes any and all combinations of one or more related listed items.

[0026] Embodiment 1: Figure 1 It is a flowchart of an enterprise information retrieval method based on artificial intelligence according to an embodiment of the present invention.

[0027] like Figure 1 As shown, an enterprise information retrieval method based on artificial intelligence includes the following steps: S1, collects the public text data of enterprises, performs entity recognition and relationship extraction on enterprise information through NER model and deep learning neural network model, and then builds the enterprise relationship network.

[0028] Among them, the public information of enterprises includes but is not limited to the company name, legal representative, registered capital, establishment time, business scope, company reports, news articles, financial statements, etc., to ensure that the data set is diverse and extensive to improve the generalization ability of the model; corporate entities such as company names, person names, etc.; corporate relationships are the relationships between entities, and the extracted relationships are cooperation, competition, investment, and holdings between enterprises; the corporate relationship network is to construct an enterprise relationship network diagram based on the results of relationship extraction. The nodes in the network diagram represent enterprises, and the edges represent the relationships between enterprises.

[0029] Specifically, collect basic information such as company name, legal representative, registered capital, establishment time, business scope, and public text data such as company reports, news articles, and financial statements to ensure that the data set covers companies of different industries and sizes to improve the generalization ability of the model; clean the collected data, remove irrelevant information, and segment the text data into words and sentences for subsequent processing. Tag the text with parts of speech to help the NER model better identify entities. Select the NER model, input the preprocessed data into the NER model for training, and identify entities such as company names and person names. Label the relationships of the identified entities, such as cooperation, competition, investment, and holding, and use the graph convolutional network (GCN) deep learning neural network model to train the labeled data and extract the relationships between companies. Based on the results of relationship extraction, construct a corporate relationship network diagram, with nodes representing companies and edges representing relationships between companies (cooperative relationships, competitive relationships, litigation relationships, supply chain relationships, investment relationships, and holding relationships, etc.). The training parameters of the NER model are learning rate 0.001, batch size 32, and training rounds 10; the training parameters of the relation extraction model are learning rate 0.0005, batch size 64, and training rounds 20.

[0030] S2, when the searching enterprise sends an enterprise information search request to the system, the request text is converted into a vector representation, input into the trained CNN model, and the search intent feature vector of the searching enterprise is output.

[0031] Specifically, the CNN model is trained based on the historical search data of the searching enterprise. When the searching enterprise currently sends a search request for information of each enterprise, the request is input into the trained CNN model to output the search intention of the searching enterprise.

[0032] For example, when searching for a company in the past, the searcher enters financial information to understand the company's loan situation, and enters corporate structure information to understand the company's business structure. Model training is performed based on searching for company-financial information-loan situation, searching for company-corporate structure-business structure. When the searched company enters financial information next time, the search intention of the searched company is to understand the loan situation. When the searched company enters corporate structure, the search intention of the searched company is to understand the business structure.

[0033] S3, fuses the feature vectors in the enterprise relationship network and the search intent of the searched enterprise, and sets the tendency weight for the search intent of the searched enterprise.

[0034] Among them, the tendency weight is set for different search intentions according to the historical search intentions of the search enterprise. The tendency weight reflects the degree of importance or preference of the search enterprise on specific aspects of the partner enterprise when selecting a partner. When the search enterprise particularly values ​​a certain aspect of the characteristics or capabilities of the partner enterprise, the tendency weight of this aspect will be correspondingly higher. For example, for search enterprises that frequently conduct risk assessments, the weight of risk assessment is set higher; the feature vectors in the enterprise relationship network and the search intention of the search enterprise are fused; according to the historical search behavior and preferences of the search enterprise, the tendency weight is set for different search intentions, and it is updated regularly.

[0035] S4, build a multi-task learning model including task sharing layer, individual task layer and output layer.

[0036] Among them, the task sharing layer is used to share information between different retrieval tasks and extract common features between tasks; the company's public information and enterprise information retrieval requests are divided into four categories: risk assessment, strategic planning, investment decision-making and recruitment management. A separate task layer is set for each task such as risk assessment, strategic planning, investment decision-making and recruitment management to learn the unique features of each task; the output layer is set as the output node corresponding to each retrieval task.

[0037] S5, use the labeled dataset to train the multi-task learning model.

[0038] Specifically, we prepare labeled datasets containing enterprise relationship networks, enterprise search intent, risk assessment, strategic planning, investment decision-making, and recruitment management, input the fused feature vectors into the multi-task learning model for training, and optimize the model parameters through supervised learning.

[0039] S6, design a cooperation feasibility value output node in the output layer of the multi-task learning model, input the result after feature fusion into the multi-task model, and obtain the cooperation feasibility value.

[0040] Among them, the calculation formula of the cooperation feasibility value S is:

[0041] represents the output value corresponding to the i-th retrieval task (enterprise relationship network, retrieval enterprise retrieval intention, risk assessment, strategic planning, investment decision and recruitment management), The inclination weight of the i-th search intent feature is represented, and the calculated cooperation feasibility value is output to the searching enterprise as the basis for cooperation decision-making.

[0042] It should be noted that the system includes: The data collection and preprocessing module is responsible for collecting the company's public text data information and performing preprocessing work such as cleaning, word segmentation, and part-of-speech tagging to provide a high-quality data foundation for subsequent entity recognition and relationship extraction.

[0043] The entity recognition and relationship extraction module (including the NER model and the GCN model) is used to use the NER model to identify entities such as company names and person names, and then use the GCN model to extract the relationships between companies, such as cooperation, competition, investment, and holding, and finally construct a corporate relationship network diagram.

[0044] The retrieval enterprise retrieval intention recognition module (CNN model) is used to convert the enterprise information retrieval request sent by the retrieval enterprise into a vector representation through the CNN model, and output the retrieval intention feature vector of the retrieval enterprise for subsequent fusion with the enterprise relationship network.

[0045] The feature fusion and tendency weight setting module is used to fuse the feature vectors in the enterprise relationship network and the search intent of the searching enterprise, and set tendency weights for different search intents according to the historical search intent of the searching enterprise to reflect the preference and importance of the searching enterprise to different tasks.

[0046] The multi-task learning model construction and training module is used to build a multi-task learning model that includes a task sharing layer, a separate task layer, and an output layer. It optimizes the model parameters through supervised learning so that the model can handle multiple retrieval tasks at the same time, such as risk assessment, strategic planning, investment decision-making, recruitment management, etc.

[0047] The cooperation feasibility value calculation and output module is used to design the cooperation feasibility value output node in the output layer of the multi-task learning model, calculate the cooperation feasibility value according to the fused feature vector and tendency weight, and output it to the retrieval enterprise as the basis for cooperation decision-making.

[0048] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: By comprehensively and deeply collecting and preprocessing the public text data information of enterprises, a solid foundation is laid for subsequent entity recognition and relationship extraction, greatly improving the accuracy and comprehensiveness of information processing. At the same time, the advanced CNN model is used to accurately identify the search intent of the search enterprise, so that the system can more accurately capture the needs of the search enterprise and provide highly personalized information services. By organically integrating the enterprise relationship network with the needs of the search enterprise and setting reasonable tendency weights, the model can more flexibly adapt to the preferences and needs of different search enterprises, enhancing the adaptability and practicality of the system. In addition, the constructed multi-task learning model realizes information sharing and feature learning between different search tasks, which not only improves the overall performance and efficiency of the model, but also enables the system to perform well on multiple search tasks at the same time. Finally, by designing the cooperation feasibility value output node and calculating the cooperation feasibility value, an intuitive and quantifiable basis for cooperation decision-making is provided for the search enterprise, which greatly reduces the risk and uncertainty of decision-making and significantly improves the accuracy and practicality of enterprise information processing and cooperation evaluation.

[0049] Example 2: In Example 1, although it has been possible to achieve comprehensive processing of enterprise information and accurate identification of the search intent of the retrieval enterprise, and calculate the cooperation feasibility value based on this information, it has certain limitations in the expansion and deep mining of cooperative relationships. Specifically, Example 1 mainly focuses on direct cooperative relationships, that is, the feasibility of cooperation between the retrieval enterprise and the intended cooperative enterprise, but fails to fully consider the possibility of expanding the cooperative network through the intended cooperative enterprise. Therefore, this embodiment proposes to solve the limitations of Example 1 in the expansion of cooperative relationships, by introducing intermediate companies as jumping points, to achieve deep mining and expansion of the cooperative network, and provide more and better cooperation opportunities for the retrieval enterprise.

[0050] In some embodiments, when building an enterprise relationship network, step S1 further includes: S11, construct and update the enterprise relationship network diagram, and explicitly add the cooperative relationship between the first-level cooperative enterprises and the second-level cooperative enterprises of the searched enterprise into the enterprise relationship network diagram.

[0051] Among them, the first-level cooperative enterprise is the enterprise that checks the feasible value of cooperation with the searching enterprise (the intended enterprise that wants to cooperate), and the second-level enterprise is the affiliated enterprise that has cooperation with the first-level enterprise.

[0052] S12, when the searching enterprise sends an enterprise information searching request, the system first identifies the cooperative relationship between the searching enterprise and the primary cooperative enterprise, and then identifies the cooperative relationship between the primary cooperative enterprise and the secondary cooperative enterprise through the primary cooperative enterprise as a jumping point.

[0053] S13, for the second cooperation feasibility assessment between the searching enterprise and the first-level cooperative enterprise, the feature fusion method of S6 is continued to be used to fuse the feature vector in the enterprise relationship network with the feature vector of the searching enterprise's search intention.

[0054] Among them, for the feasibility assessment of the second cooperation between the search enterprise and the second-level cooperative enterprise, it is necessary to integrate the cooperation feature vectors between the search enterprise and the first-level cooperative enterprise, and the cooperation feature vectors between the first-level cooperative enterprise and the second-level cooperative enterprise. Specifically, let F 检索企业-一级 To retrieve the fusion feature vector of the enterprise and the first-level cooperative enterprise, let F 一级-二级 is the fusion feature vector of the first-level enterprise and the second-level cooperative enterprise, then the fusion feature vector of the core enterprise and the second-level cooperative enterprise is F 检索企业-二级 =αF 检索企业-一级 + β F 一级-二级 , where α and β are weight coefficients, which respectively represent the importance of the cooperative relationship between the retrieval enterprise and the first-level cooperative enterprise, and between the first-level cooperative enterprise and the second-level cooperative enterprise.

[0055] S14, adds the identification of cooperative paths to the multi-task learning model and trains the model using the updated dataset.

[0056] The cooperation path is search enterprise → first-level cooperation enterprise → second-level cooperation enterprise, and the updated data set is the cooperation information between the search enterprise and the first-level cooperation enterprise, and between the first-level cooperation enterprise and the second-level cooperation enterprise.

[0057] S15, using the upgraded multi-task learning model, calculate the feasibility of the second cooperation between the search enterprise and the secondary cooperative enterprise. The second cooperation feasibility value is calculated by weighted average. P , for the search enterprise and the secondary cooperative enterprise, the second cooperation feasibility value P 检索企业-二级 The calculation formula is:

[0058] represents the output value corresponding to the i-th evaluation task, represents the weight of the i-th evaluation task.

[0059] S16, setting the second cooperation feasibility value P within a threshold range, evaluating and judging the second cooperation feasibility according to the second cooperation feasibility value P, and pushing secondary cooperation enterprises whose second cooperation feasibility is higher than the threshold to the searching enterprise.

[0060] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: By introducing the jump point of cooperative relationships, the scope of the enterprise cooperation network is effectively expanded, allowing enterprises to discover more potential partners. At the same time, the solution improves the accuracy and reliability of cooperation possibility assessment by integrating multi-dimensional feature vectors and upgrading the multi-task learning model. This refined evaluation method not only takes into account direct cooperative relationships, but also explores indirect cooperation opportunities, providing enterprises with more diversified cooperation options. In addition, the solution can also intelligently push enterprises with a higher possibility of cooperation, providing strong support for enterprises' cooperation decisions, greatly improving cooperation efficiency and success rate.

[0061] Embodiment 3: In Embodiment 2, when searching for enterprise information, the user may not be satisfied with the cooperative enterprises obtained in the initial search, so a second, third or even multiple searches are required. Multiple searches will establish multiple cooperation paths and form multi-level associated enterprises. This embodiment aims to propose a method for calculating the cooperation value of multi-level associated enterprises to evaluate and recommend enterprises with high cooperation value.

[0062] In some embodiments, in step S14, adding a new cooperation path further includes: S141, when searching for enterprises for multiple times, the search results with the second cooperation feasibility value P greater than the threshold are added to the network graph as new cooperation paths to form multiple cooperation paths.

[0063] Among them, multiple cooperation paths = (enterprise search → first-level cooperation enterprise i → second-level cooperation enterprise i → ... → n-level cooperation enterprise). Each cooperation path starts from the search enterprise and extends step by step through the first-level cooperation enterprise, second-level cooperation enterprise, etc.

[0064] S142, identifying multi-level related enterprises based on multiple cooperation paths of the searched enterprises.

[0065] Among them, multi-level affiliated enterprises are {first-level affiliated enterprises, second-level affiliated enterprises, third-level affiliated enterprises...n-level affiliated enterprises}; first-level affiliated enterprises (enterprises with cooperation intention directly retrieved by the searching enterprise), second-level affiliated enterprises (a collection of enterprises cooperating with first-level affiliated enterprises), third-level affiliated enterprises (a collection of enterprises cooperating with second-level affiliated enterprises), and so on.

[0066] For example, company A, as a search company, searches for three first-level cooperative companies with cooperation intentions, namely B, C, and D. These three companies constitute first-level related companies. Company A cooperates with company B, Company A and Company B cooperate with company C, and Company C, Company A, and Company D cooperate with company D. Therefore, companies A, B, C, and D constitute second-level related companies.

[0067] S143, for the relationship between adjacent levels in the multi-level associated enterprises, calculate the cooperation feature vector between the adjacent levels and the search enterprise, and merge the cooperation feature vectors of the search enterprise and the multi-level associated enterprises.

[0068] S144, determining the multi-level cooperation degree between the search enterprise and the multi-level related enterprises, sorting the multi-level cooperation degrees in descending order, and calculating the cooperation values ​​between the search enterprise and the multi-level cooperation degree enterprises in sequence.

[0069] Among them, the multi-level cooperation degree is the number of cooperations between a company in the n-level associated enterprises and different companies in the n-1 associated enterprises; for example, in the example in S141, because companies B, C, and D all cooperate with company A, and companies B, C, and D are the first-level cooperative enterprises of company A, the multi-level cooperation degree between company A and company A is 3.

[0070] The formula for calculating the cooperation value is:

[0071] To retrieve the cooperation feature vector between an enterprise and its i-th level related enterprise, is the multi-level cooperation degree of a company in the i-th level of related enterprises, To retrieve the cooperation value between the enterprise and the i-th level related enterprise.

[0072] S145, adding the cooperation value between the retrieval enterprise and the multi-level cooperation enterprise to the upgraded multi-task learning model, and outputting the cooperation value between the retrieval enterprise and the multi-level cooperation enterprise.

[0073] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: By searching multiple companies and building a multi-level network of related companies, the solution can comprehensively and intuitively display the complex relationship between the searched companies and their potential cooperative companies. At the same time, by calculating the cooperation feature vector and multi-level cooperation degree, the solution can quantitatively evaluate the cooperation closeness and cooperation potential between the searched companies and multi-level related companies, providing strong data support for the cooperation decision-making of the companies. In addition, the solution further optimizes the selection process of cooperative companies through the calculation of cooperation value, which helps to improve the cooperation efficiency and cooperation success rate of the companies.

[0074] Example 4: In Example 1, the evaluation of the possibility of cooperation is mainly based on the information obtained from the initial search, which can often only reflect the basic situation of the cooperating enterprise and the direct cooperation potential, but it is difficult to fully and deeply reveal the deep-seated value and potential risks of the cooperative relationship. Due to the limitations of the initial search, multi-level affiliated enterprises that have indirect cooperative relationships with the searched enterprise, as well as the cooperation opportunities and value brought by these enterprises, may be ignored. Therefore, in order to further improve the accuracy of the calculation of the cooperation feasibility value of Example 1, this solution proposes to use the cooperation value of the multi-level affiliated enterprises in Example 3 to calibrate the cooperation feasibility value of Example 1.

[0075] In some embodiments, in step S6, obtaining the cooperation feasibility value further includes: S61, obtaining the cooperation value between the search enterprise and the multi-level cooperation enterprises, and calibrating the cooperation feasibility value.

[0076] S62, calibrating and calculating the cooperation value and the cooperation feasibility value to obtain a calibrated cooperation feasibility value.

[0077] Among them, the calibrated cooperation feasibility value = R (cooperation feasibility value, cooperation value), R represents the selection of function type, and weighted addition can be selected for calculation.

[0078] S63, adding the calibrated cooperation feasibility value to the upgraded multi-task learning model.

[0079] S64, using the updated multi-task learning model, outputs a calibrated cooperation feasibility value.

[0080] The technical solutions in the above embodiments of the present application have at least the following technical effects or advantages: By obtaining the cooperation value between the search enterprise and multi-level cooperation enterprises and calibrating the original cooperation feasibility value, the accuracy and comprehensiveness of the cooperation evaluation are significantly improved. The calibrated cooperation feasibility value is integrated into the upgraded multi-task learning model, so that the model can make fuller use of multi-source information for more accurate learning and prediction. This technical solution not only optimizes the evaluation process of cooperation potential, but also provides enterprises with a more reliable basis for cooperation decision-making, which helps enterprises to more accurately identify cooperation opportunities and optimize cooperation networks in a complex and changing market environment.

[0081] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An enterprise information retrieval method based on artificial intelligence, characterized in that: include: S1, collect the public text data of enterprises, perform entity recognition and relationship extraction on enterprise information through NER model and deep learning neural network model, and then build an enterprise relationship network; the enterprise relationship network is to build an enterprise relationship network diagram based on the results of relationship extraction, in which nodes represent enterprises and edges represent the relationship between enterprises; S2, when the searching enterprise sends an enterprise information search request, the request text is converted into a vector representation, input into the trained CNN model, and the search intention feature vector of the searching enterprise is output; S3, merge the feature vectors in the enterprise relationship network and the search intent of the searched enterprise, and set the tendency weight for the search intent of the searched enterprise; S4, construct a multi-task learning model including a task sharing layer, a separate task layer and an output layer; S5, use the labeled dataset to train the multi-task learning model; S6, design a cooperation feasibility value output node in the output layer of the multi-task learning model, input the result after feature fusion into the multi-task model, and obtain the cooperation feasibility value.

2. The enterprise information retrieval method based on artificial intelligence according to claim 1, characterized in that: The output of the search intention feature vector of the search enterprise specifically includes: training the CNN model according to the historical search data of the search enterprise, and when the search enterprise currently sends each enterprise information search request, the request is input into the trained CNN model to output the search intention of the search enterprise.

3. The enterprise information retrieval method based on artificial intelligence according to claim 1, characterized in that: In S3, the tendency weight includes: the tendency weight is set for different search intentions according to the historical search intentions of the searched enterprise.

4. The enterprise information retrieval method based on artificial intelligence according to claim 1, characterized in that: In S6, obtaining the cooperation feasibility value includes: The formula for obtaining the cooperation feasibility value is as follows: represents the output value corresponding to the i-th retrieval task, It represents the tendency weight of the i-th search intention feature, and outputs the calculated cooperation feasibility value to the searching enterprise.

5. The enterprise information retrieval method based on artificial intelligence according to claim 1, characterized in that: Said S1 further comprises: S11, construct and update the enterprise relationship network diagram, and explicitly add the cooperation relationship between the first-level cooperative enterprise and the second-level cooperative enterprise of the search enterprise to the enterprise relationship network diagram; wherein the first-level cooperative enterprise is the enterprise that checks the cooperation feasibility value with the search enterprise, and the second-level enterprise is the associated enterprise that cooperates with the first-level enterprise; S12, when the search enterprise sends an enterprise information search request, the system first identifies the cooperation relationship between the search enterprise and the first-level cooperative enterprise, and then uses the first-level cooperative enterprise as a jumping point to identify the cooperation relationship between the first-level cooperative enterprise and the second-level cooperative enterprise; S13, for the feasibility assessment of the second cooperation between the search enterprise and the first-level cooperative enterprise, the feature fusion method of S6 is continued to be used to fuse the feature vector in the enterprise relationship network with the feature vector of the search intention of the search enterprise; S14, added the identification of cooperative paths to the multi-task learning model and trained the model using the updated dataset; S15, using the upgraded multi-task learning model, calculates the feasibility of the second cooperation between the retrieved enterprise and the secondary cooperative enterprise; S16, setting the second cooperation feasibility value P within a threshold range, evaluating and judging the second cooperation feasibility according to the second cooperation feasibility value P, and pushing secondary cooperation enterprises whose second cooperation feasibility is higher than the threshold to the searching enterprise.

6. The enterprise information retrieval method based on artificial intelligence as claimed in claim 5, characterized in that: The S14 includes: the cooperation path is the search enterprise→the first-level cooperation enterprise→the second-level cooperation enterprise; the updated data set is the cooperation information between the search enterprise and the first-level cooperation enterprise, and between the first-level cooperation enterprise and the second-level cooperation enterprise.

7. The enterprise information retrieval method based on artificial intelligence as claimed in claim 5, characterized in that: The step S15, calculating the feasibility of the second cooperation between the search enterprise and the second-level cooperative enterprise, includes: for the evaluation of the feasibility of the second cooperation between the search enterprise and the second-level cooperative enterprise, it is necessary to integrate the cooperation feature vector between the search enterprise and the first-level cooperative enterprise, and the cooperation feature vector between the first-level cooperative enterprise and the second-level cooperative enterprise, specifically: let F 检索企业-一级 To retrieve the fusion feature vector of the enterprise and the first-level cooperative enterprise, let F 一级-二级 is the fusion feature vector of the first-level enterprise and the second-level cooperative enterprise, then the fusion feature vector of the core enterprise and the second-level cooperative enterprise is F 检索企业-二级 =αF 检索企业-一级 + β F 一级-二级 , where α and β are weight coefficients, which respectively represent the importance of the cooperative relationship between the search enterprise and the first-level cooperative enterprise, and between the first-level cooperative enterprise and the second-level cooperative enterprise; the second cooperation feasibility value P 检索企业-二级 The calculation formula is: represents the output value corresponding to the i-th evaluation task, represents the weight of the i-th evaluation task.

8. The enterprise information retrieval method based on artificial intelligence as claimed in claim 7, characterized in that: In the above S14, the newly added cooperation path in S14 also includes: S141, when the searching enterprise conducts multiple enterprise searches, the search results with the second cooperation feasibility value P greater than the threshold are added as new cooperation paths to the network diagram to form multiple cooperation paths; multiple cooperation paths = (enterprise search → first-level cooperation enterprise i → second-level cooperation enterprise i → ... → n-level cooperation enterprise), each cooperation path starts from the searching enterprise and extends step by step through the first-level cooperation enterprises; S142, identifying multi-level related enterprises based on multiple cooperation paths of the searched enterprises; S143, for the relationship between adjacent levels in the multi-level associated enterprises, calculating the cooperation feature vector between the adjacent levels and the search enterprise, and fusing the cooperation feature vectors between the search enterprise and the multi-level associated enterprises; S144, determining the multi-level cooperation degree between the search enterprise and the multi-level related enterprises, sorting the multi-level cooperation degrees in descending order, and calculating the cooperation values ​​between the search enterprise and the multi-level cooperation degree enterprises in sequence; S145, adding the cooperation value between the retrieval enterprise and the multi-level cooperation enterprise to the upgraded multi-task learning model, and outputting the cooperation value between the retrieval enterprise and the multi-level cooperation enterprise.

9. The enterprise information retrieval method based on artificial intelligence as claimed in claim 5, characterized in that: In S144, the value of cooperation includes: The formula for cooperation value is as follows: To retrieve the cooperation feature vector between an enterprise and its i-th level related enterprise, is the multi-level cooperation degree of a company in the i-th level of related enterprises, To retrieve the cooperation value between the enterprise and the i-th level related enterprise.

10. The enterprise information retrieval method based on artificial intelligence according to claim 8, characterized in that: include: S61, obtaining the cooperation value between the search enterprise and the multi-level cooperation degree enterprises, and calibrating the cooperation feasibility value; S62, calibrating and calculating the cooperation value and the cooperation feasibility value to obtain a calibrated cooperation feasibility value; wherein the calibrated cooperation feasibility value=R(cooperation feasibility value, cooperation value), R represents the selection of a function type, and weighted addition can be selected for calculation; S63, adding the calibrated cooperation feasibility value to the upgraded multi-task learning model; S64, using the updated multi-task learning model, outputs a calibrated cooperation feasibility value.

Images