Bid invitation demand analysis and local file matching method and system based on large model

Abstract

The invention discloses a bid invitation demand analysis and local file matching method and system based on a large model, and the method comprises the steps: building a local credit file library, classifying credit big data through a semantic model, extracting the data, constructing a knowledge base, and uploading the data to a cloud large model; inputting a bid invitation file to a cloud large model to identify and analyze business demands, extracting and standardizing a demand file in combination with a knowledge base, and packaging and downloading a corresponding file; when the knowledge base is constructed, a graph structure of the knowledge base is used as a classification mode, and a word2vec model is used for constructing a word vector model to capture a semantic relationship; mapping the semantic relationship to a graph entity relationship model, adding relationship data, and screening keyword vectors to construct a decision tree classification model; according to the graph structure, credit big data entity word vectors and classification information are obtained to construct decision training data, the decision training data are imported into a classification model for training, and rapid identification, information classification and qualification analysis of bid invitation texts are realized through the classification model, so that the file matching efficiency is effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of large model identification, and more particularly, to a large model-based bidding demand analysis and local file matching method and system. BACKGROUND

[0002] In the bidding process, extracting and identifying the business conditions and qualification requirements in the bidding documents and matching the localized files are key links. In the prior art, this process mainly relies on manual operation, which not only consumes a lot of time and human resources, but also is prone to information omission or errors due to human negligence, making it difficult to meet the high requirements of efficiency and accuracy of the bidding process. In addition, the tasks of manually checking the number and quality of qualifications, manually identifying qualification requirements, verifying the validity of qualifications, and manually matching and extracting local files are heavy and prone to errors, further increasing the complexity and risk of the bidding work. Therefore, an automated information extraction and matching method is urgently needed to improve work efficiency, reduce errors, and ensure the security and confidentiality of bidding information. SUMMARY

[0003] The present application overcomes the defects of the prior art and provides a large model-based bidding demand analysis and local file matching method and system.

[0004] The first aspect of the present application provides a large model-based bidding demand analysis and local file matching method, comprising: S101: Establishing a local credit file library, classifying text data of credit big data through a semantic model, extracting entity, attribute, and relationship data, and constructing a knowledge base, and uploading the knowledge base to a cloud model; S102: Inputting the bidding documents into the cloud model for business demand recognition analysis, extracting the demand documents in combination with the knowledge base, standardizing the demand documents, obtaining a standard file type description array, and downloading the corresponding file list in combination with the credit file library according to the array; S103: In the construction of the knowledge base, the graph structure of the knowledge base is used as the classification mode of the credit big data, the word data of the knowledge base is constructed into a vector model through a word2vec model, and the semantic relationship between the word vectors is captured; S104: Mapping the semantic relationship between the word vectors to the relationship model between the entities in the graph structure, adding relationship data in the graph based on the semantic relationship, and screening out key word vectors based on the semantic relationship between the word vectors, using the key word vectors as initial decision nodes, and constructing a classification model based on decision trees; S105: According to the graph structure of the knowledge base, the entity word vectors and classification information of the credit big data are obtained, the decision training data is constructed, the decision training data is imported into the classification model, and the classification parameters are adjusted through a heuristic algorithm.

[0005] In the scheme, the S101 is specifically: Establish a local credit file library, obtain credit big data from the credit file library, and convert the credit big data to obtain text big data; Introduce a BERT semantic model, segment the text big data by a preset delimiter, classify the keywords according to various credit types, and extract entity, attribute, and relationship data from the text big data to obtain triple information; In the triple information, the relationship data is specifically set according to the correlation of text data of different credit types; A knowledge base is constructed through the triple, and the knowledge base is uploaded to a cloud big model.

[0006] In the scheme, the S102 is specifically: Input the tender file into the cloud big model and perform business demand analysis to generate business demand keywords; Through the business demand keywords, the demand file is retrieved in the knowledge base to obtain the demand file; According to the file standard in the credit file library, the demand file is standardized to generate a standard file type description array; According to the standard file type description array, a file list is obtained and the file is packaged and downloaded in the credit file library.

[0007] In the scheme, the S103 is specifically: According to the graph structure of the knowledge data in the knowledge base, the classification mode of the credit big data is set; Through the semantic model and the knowledge base, the keywords and the text big data of the credit big data are obtained, the text big data is taken as a corpus, the context relationship of the keywords is learned based on a word2vec model, a vector space based on semantic relationship is constructed according to the keywords, and each keyword is assigned a vector representation through the vector space.

[0008] In the scheme, the S104 includes: Entity data is extracted in the knowledge base, and corresponding keywords are associated to obtain entity keywords; The vector representation corresponding to the entity keywords is taken as an analysis object, the similarity between vectors is calculated in the vector space through cosine similarity, and the relationship strength of the entity keywords is evaluated based on the similarity to set a semantic relationship mode; According to the semantic relationship mode, the relationship data in the graph structure is expanded and applied to data updating of the knowledge base; Through the semantic relationship mode, entity keywords with strong correlation characteristics are screened out, the corresponding vector is taken as a keyword vector, and the corresponding classification information and the keyword vector are associated as a classification result; According to the keyword vector, a plurality of initial decision nodes are formed by conditional transformation, and the arrangement of the initial decision nodes is determined by an ID3 heuristic algorithm, and a classification model based on a decision tree is constructed.

[0009] In the scheme, the S105, specifically: The AMIE algorithm is used for knowledge expansion and relationship information addition to the graph structure in the knowledge base. According to the graph structure in the knowledge base, entity information analysis is performed on the credit big data, and corresponding entity word vectors and classification information are obtained. According to the entity word vector and the classification information, a decision training data is constructed, the decision training data is imported into a classification model for classification training, the recall rate of the classification result is evaluated to evaluate the model effect, and the ID3 algorithm is used to adjust the layout and quantity of the decision nodes.

[0010] In the scheme, the local credit file library establishes data connection with the cloud large model through a preset protocol, and the preset protocol includes a transmission encryption and storage encryption protocol.

[0011] The second aspect of the present application also provides a tender demand analysis and local file matching system based on a large model, which comprises a memory, a processor and a data interface, wherein the memory comprises a tender demand analysis and local file matching program based on a large model, and the tender demand analysis and local file matching program based on a large model realizes the following steps when executed by the processor: S101: Establish a local credit file library, classify the credit big data by a semantic model, extract entity, attribute and relationship data, and construct a knowledge base, and upload the knowledge base to a cloud large model; S102: Input the tender file into the cloud large model to perform business demand recognition analysis, extract the demand file in combination with the knowledge base, standardize the demand file, obtain a standard file type description array, and download the file in combination with the credit file library according to the corresponding file list obtained from the array; S103: In the construction of the knowledge base, the graph structure of the knowledge base is used as the classification mode of the credit big data, the word data of the knowledge base is constructed into a vector model by a word2vec model, and the semantic relationship between the word vectors is captured; S104: The semantic relationship between the word vectors is mapped to the relationship model between the entities in the graph structure, and the relationship data in the graph is added based on the semantic relationship, the key word vector is screened out based on the semantic relationship between the word vectors, the key word vector is used as an initial decision node, and a classification model based on a decision tree is constructed; S105: According to the atlas structure in the knowledge base, entity word vectors and classification information of credit big data are obtained, decision training data are constructed, the decision training data are imported into a classification model, and classification parameter adjustment is performed through a heuristic algorithm.

[0012] The third aspect of the present application also provides a computer readable storage medium, wherein a large model-based bidding demand analysis and local file matching program is included, and when the program is executed by a processor, the steps of the large model-based bidding demand analysis and local file matching method according to any one of the above aspects are realized.

[0013] Through the present application, the workload of bidding personnel can be effectively reduced, the effectiveness of automatic identification of bidding demand and qualification is realized, the automatic identification of bidding files and text information is realized, the automatic checking of the number and quality of qualifications is realized, the accuracy of qualification matching is improved, and through the cloud large model, the large model and the local credit file are effectively isolated, and the file security is ensured. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of a large model-based bidding demand analysis and local file matching method of the present application is shown. Figure 2 A classification model construction flowchart of the present application is shown. Figure 3 A block diagram of a large model-based bidding demand analysis and local file matching system of the present application is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. It can be understood that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0016] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the protection scope of the present application is not limited by the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of a large model-based bidding demand analysis and local file matching method of the present application is shown.

[0018] As Figure 1 shown, a flowchart is simplified; The first aspect of the present application provides a large model-based bidding demand analysis and local file matching method, comprising: S101: Establish a local credit file library, classify the credit big data through the semantic model, extract entity, attribute and relationship data, and build a knowledge base, and upload the knowledge base to a cloud big model; S102: Input the tender file into the cloud big model to analyze the business demand, extract the demand file combined with the knowledge base, standardize the demand file, obtain a standard file type description array, and download the file according to the array combined with the credit file library; S103: In the knowledge base construction, the graph structure of the knowledge base is used as the classification mode of the credit big data, the word data of the knowledge base is constructed into a vector model through a word2vec model, and the semantic relationship between the word vectors is captured; S104: The semantic relationship between the word vectors is mapped to the relationship model between the entities in the graph structure, and the relationship data in the graph is added based on the semantic relationship, the key word vectors are selected based on the semantic relationship between the word vectors, and the classification model based on the decision tree is constructed with the key word vectors as the initial decision nodes; S105: According to the graph structure in the knowledge base, the entity word vector and the classification information of the credit big data are obtained, the decision training data is constructed, the decision training data is imported into the classification model, and the classification parameter is adjusted through the heuristic algorithm.

[0019] According to the embodiment of the application, the S101 is specifically: Establish a local credit file library, obtain credit big data from the credit file library, and convert the credit big data into text big data; Introduce a BERT semantic model, segment the text big data into key words through a preset delimiter, classify the key words according to various credit types, and extract entity, attribute and relationship data from the text big data to obtain triple information; In the triple information, the relationship data is specifically set according to the correlation of the text data of different credit types; A knowledge base is constructed through the triple, and the knowledge base is uploaded to a cloud big model.

[0020] Here, it can be understood that the preset delimiter is generally a preset punctuation mark. The plurality of credit types specifically include a plurality of types of credit partial data, such as: Enterprise qualification: business license, production license, industry certification certificate (such as ISO certification), etc., proving the legal operation and professional ability of the enterprise.

[0021] Performance proof: similar project cases in the past three years, contract copies, user evaluation, etc., reflecting the execution ability and experience.

[0022] Personnel configuration: project team member resume, professional title certificate, social security payment proof, to ensure that the personnel have professional qualifications.

[0023] Financial status: audit report, bank credit proof, to show the enterprise's debt paying ability and financial stability.

[0024] Legal documents: legal authorization, joint agreement (if applicable), no illegal record statement, to ensure the legality of the bid.

[0025] In the entity and attribute data, the entity data can be a type entity, such as enterprise qualification, and the attribute can be the corresponding qualification information and credit file type table (including file name, file level, including professional information), such as business license, production license, industry certification, file attribute information, etc. After forming the knowledge data, the corresponding credit big data can be classified and stored based on the graph structure, and the decision tree can be constructed and optimized through the graph structure in the future.

[0026] In the relationship data, the data correlation of different credit types is set, and the data correlation of different credit types is evaluated, for example, there is a certain data correlation between enterprise qualification and performance proof, the data type, data format and part of the data may be consistent, based on which the relationship data between entities (or the same type of data) can be established to evaluate the relationship mode between different entities (or the same type of data).

[0027] According to the embodiment of the present application, the S102 is specifically: Input the bidding file to the cloud large model and perform business demand analysis to generate business demand keywords; Retrieving the demand file in the knowledge base through the business demand keywords to obtain the demand file; Standardizing the demand file according to the file standard in the credit file library to generate a standard file type description array; Obtaining the file list according to the standard file type description array and downloading the file in the credit file library.

[0028] Here, it can be understood that the cloud large model is a model for recognizing and analyzing the bidding text and file matching by using cloud computing power. The file list is set for fast extraction and download of the file, and the standard file type description array corresponds to the file information description of the credit file library, including file type, file size, file described credit type and other information.

[0029] The demand file is a demand file description based on the graph structure in the knowledge base, including entity, attribute and other retrieval data, and the content is the corresponding credit text data.

[0030] According to the embodiment of the present application, the S103 is specifically: According to the graph structure of the knowledge data in the knowledge base, a classification mode of the credit big data is set; Through the semantic model and the knowledge base, keywords and text big data of the credit big data are acquired, the text big data is taken as a corpus, the context relationship of the keywords is learned based on a word2vec model, a vector space based on a semantic relationship is constructed according to the keywords, and each keyword is assigned a vector representation through the vector space.

[0031] It can be understood here that in the graph structure, the classification results of multiple credit texts are set according to the division of entities and attributes, and corresponding classification models are set, so that different types of credit data can be efficiently classified and stored. The keywords and the text big data are data analyzed based on the semantic model in S101. Each keyword corresponds to a vector representation, and the semantic relationship between the keywords can be analyzed through the cosine similarity of the vectors.

[0032] In the embodiment, the semantic relationship is constructed to capture the entity relationship model and update the relationship model of the graph, mine the potential relationship mode of the data, apply two dimensions to analyze the relationship mode of different types of credit data, and construct a decision tree classification model based on the relationship mode analysis, effectively improve the text classification, graph structure knowledge mining and text recognition extraction efficiency of the bidding big data file, and effectively reduce the computing power consumption of the bidding demand recognition and local file matching.

[0033] According to the embodiment of the application, the S104 comprises: Entity data is extracted in the knowledge base, and corresponding keywords are associated to obtain entity keywords; The vector representation corresponding to the entity keywords is taken as an analysis object, the similarity between the vectors is calculated through the cosine similarity in the vector space, the relationship strength of the entity keywords is evaluated based on the similarity, and a semantic relationship mode is set; According to the semantic relationship mode, the relationship data in the graph structure is expanded and applied to the data update of the knowledge base; Through the semantic relationship mode, entity keywords with strong association characteristics are screened out, the corresponding vectors are taken as keyword vectors, the corresponding classification information is associated with the keyword vectors as a classification result; According to the keyword vector, a plurality of initial decision nodes are formed through conditional transformation, the arrangement of the initial decision nodes is determined through an ID3 heuristic algorithm, and a classification model based on a decision tree is constructed.

[0034] Here it can be understood that the corresponding classification information is the entity classification information corresponding to the entity keyword with strong association characteristics. In the semantic relationship mode, the relationship strength between each entity keyword and other entity keywords is stored. The entity keyword with strong association characteristics has a strong semantic relationship with multiple entity keywords. The entity keyword with strong association characteristics plays an important role in text recognition and tender data association, so the strong association entity keyword is screened to set the decision condition of the decision tree, and subsequent real-time input tender demand text can be recognized and quickly classified by file type. Through the classification result, efficient tender demand file extraction and document construction are further realized, and automatic demand analysis and efficient local file matching are realized.

[0035] The decision tree model can be established in the cloud large model.

[0036] The semantic relationship mode is relationship information based on word2vec model semantic evaluation, which is used to analyze the semantic relationship between entity data (different credit file types). In the original knowledge graph structure, the relationship model of the file association level is included, the semantic relationship mode is used to expand the relationship data of the semantic level, and the potential association mode between entities and credit types can be effectively mined through the relationship mode of two dimensions, so as to improve the tender text recognition ability and classification effect based on the knowledge base. In addition, in the construction of the decision tree model, the semantic association analysis is used to set the strong association entity keyword, that is, the credit file type with high semantic association characteristics is screened out and set as a decision node, so as to improve the recognition accuracy and efficiency of tender text big data and real-time demand text, and further classify and efficiently match the tender files.

[0037] According to the embodiment of the present application, the S105, specifically: The graph structure in the knowledge base is expanded and relationship information is added through the AMIE algorithm; According to the graph structure in the knowledge base, entity information analysis is performed on the credit big data, and corresponding entity word vectors and classification information are obtained; According to the entity word vector and the classification information, a decision training number is constructed, the decision training data is imported into a classification model for classification training, the recall rate of the classification result is evaluated to evaluate the model effect, and the ID3 algorithm is used to adjust the layout and number of decision nodes.

[0038] Here it can be understood that the knowledge base includes corresponding graph structure, and the graph structure includes corresponding entity, attribute data and relationship mode information. The entity word vector is obtained through the word2vec model.

[0039] It is worth mentioning that in the existing bidding document identification and document extraction process, it is difficult to quickly and effectively classify and match the bidding credit data, and the semantic analysis keywords are often matched based on historical analysis results, lacking data relationship mining and identification analysis of credit text in combination with semantic relationship mode, data type association and other dimensions, resulting in low efficiency of bidding text identification, and lacking effective training data for text semantic model identification, making it difficult to efficiently identify large bidding files of multiple file types or large data, resulting in low bidding file automatic identification and extraction capability. The application can effectively solve the above problems.

[0040] According to the embodiment of the application, the local credit file library establishes data connection with the cloud big model through a preset protocol.

[0041] It can be understood here that the user interaction terminal is used for visualizing the bidding file identification result, the classification result, the standard file type specification array and other information. A preset encryption protocol can be used in the data connection process to ensure data security.

[0042] According to the embodiment of the application, the preset protocol includes transmission encryption and storage encryption protocol.

[0043] It should be noted that the storage between the cloud big model and the localized file implements the transmission encryption and storage encryption process, and the protocol can include TLS (Transport Layer Security Protocol) and SSL (Secure Socket Layer Protocol), AES (Advanced Encryption Standard), etc. to ensure the security of the bidding file identification and extraction process.

[0044] According to the embodiment of the application, it further comprises: In one working cycle, the input bidding file is input to the cloud big model to generate real-time bidding text, and the business demand identification analysis is performed according to the real-time bidding text, and the demand file and the real-time bidding text are identified and classified according to the classification model, and the first word vector and the first classification result after classification are generated in combination with the word2vec model; The first word vector is collected to form a real-time data set, and the GAN model is used to generate and simulate data and perform adversarial training based on the generator and the discriminator with the real-time data set as the real data; The trained GAN model generates a certain amount of simulated word vectors and imports them into the word2vec model for rationality verification and data filtering to obtain the training word vector; The training word vector and the first classification result are imported into the classification model as a decision training set for real-time training and model updating.

[0045] It can be understood here that in the multi-type and high data amount bidding document text recognition and decision classification process, it is difficult to train and update the classification model, and the classification result often deviates from the expectation, and for the training data of the classification model, it is difficult to perform model updating and adaptive training based on the existing local credit file library. Based on this, the present application acquires the bidding text data periodically, constructs real data based on the corresponding word vector, simulates the word vector data through GAN, generates a corresponding decision training set for model training, and improves the adaptability of the decision tree classification effect.

[0046] Figure 2 The classification model construction flowchart of the present application is shown.

[0047] Figure 3 The block diagram of the bidding demand analysis and local file matching system based on a large model of the present application is shown.

[0048] The second aspect of the present application also provides a bidding demand analysis and local file matching system based on a large model, which comprises a memory 101, a processor 103 and a data interface 102, the data interface is used for receiving and transmitting data, and the analysis result data is sent to other data terminals, the memory comprises a bidding demand analysis and local file matching program based on a large model, and the bidding demand analysis and local file matching program based on a large model realizes the following steps when executed by the processor: S101: Establish a local credit file library, classify the credit big data through a semantic model, extract entity, attribute and relationship data, and construct a knowledge base, and upload the knowledge base to a cloud large model; S102: Input the bidding document to the cloud large model for business demand recognition analysis, extract the demand document in combination with the knowledge base, standardize the demand document, obtain a standard file type description array, obtain a corresponding file list in combination with the credit file library according to the array, and download the file package; S103: In the knowledge base construction, the graph structure of the knowledge base is used as the classification mode of the credit big data, the word data of the knowledge base is constructed into a vector model through a word2vec model, and the semantic relationship between the word vectors is captured; S104: The semantic relationship between the word vectors is mapped to the relationship model between the entities in the graph structure, the relationship data in the graph is added based on the semantic relationship, the key word vectors are screened out based on the semantic relationship between the word vectors, the key word vectors are used as initial decision nodes, and a classification model based on a decision tree is constructed; S105: The entity word vectors and classification information of the credit big data are obtained according to the graph structure of the knowledge base, the decision training data is constructed, the decision training data is imported into the classification model, and the classification parameter is adjusted through a heuristic algorithm.

[0049] According to the embodiment of the present application, the S101, in particular: Establish a local credit file library, obtain credit big data from the credit file library, and convert the credit big data to text big data; Introduce a BERT semantic model, segment the text big data by a preset delimiter, classify the keywords according to various credit types, and extract entity, attribute, and relationship data from the text big data to obtain triple information; In the triple information, the relationship data is specifically set according to the relevance of text data of different credit types; Build a knowledge base through the triple, and upload the knowledge base to a cloud big model.

[0050] Here, it can be understood that the preset delimiter is generally a preset punctuation mark. The various credit types specifically include various types of data of credit partial data, such as: Enterprise qualification: business license, production license, industry certification certificate (such as ISO certification), etc., proving the legal operation and professional ability of the enterprise.

[0051] Performance proof: similar project cases in the past three years, contract copies, user evaluations, etc., reflecting the execution ability and experience.

[0052] Personnel allocation: project team member resumes, professional title certificates, social security payment proofs, to ensure that the personnel have professional qualifications.

[0053] Financial status: audit report, bank credit proof, to show the debt repayment ability and financial stability of the enterprise.

[0054] Legal documents: legal person's authorization, joint agreement (if applicable), no illegal record statement, to ensure the legality of the bid.

[0055] In the entity and attribute data, the entity data can be a type entity, such as enterprise qualification, and the attribute can be the corresponding qualification information and credit file type table (including file name, file level, including professional information), such as business license, production license, industry certification certificate, file attribute information, etc. After forming the knowledge data, the corresponding credit big data can be classified and stored based on the graph structure, and the decision tree can be constructed and optimized through the graph structure in the future.

[0056] In the relationship data specifically set according to the relevance of text data of different credit types, the data correlation of various credit types is specifically evaluated, for example, there is a certain data correlation between enterprise qualification and performance proof, the file data type, data format, and part of the data may be consistent, based on which the relationship data between entities can be established, and the relationship mode between different entities (or same type data) can be evaluated.

[0057] According to the embodiment of the present application, S102, in particular: Input the tender file to the cloud large model and perform business demand analysis to generate business demand keywords; Through the business demand keywords, search the demand file in the knowledge base to obtain the demand file; According to the file standard in the credit file library, standardize the demand file to generate a standard file type description array; According to the standard file type description array, obtain the file list and perform file packaging and downloading in the credit file library.

[0058] Here it can be understood that the cloud large model is a model for recognizing tender text and analyzing file matching using cloud computing power. The file list is set for fast extraction and downloading of files, and the standard file type description array corresponds to the file information description of the credit file library, including file type, file size, and file credit type information.

[0059] The demand file is a demand file description based on the graph structure in the knowledge base, including entity, attribute, and retrieval data, and the content is the corresponding credit text data.

[0060] According to the embodiment of the present application, S103, in particular: According to the graph structure of the knowledge data in the knowledge base, set the classification mode of the credit big data; Through the semantic model and the knowledge base, obtain the keywords and text big data of the credit big data, use the text big data as a corpus, learn the context relationship of the keywords based on the word2vec model, construct a vector space based on the semantic relationship according to the keywords, and assign a vector to each keyword through the vector space. representation.

[0061] Here it can be understood that in the graph structure, the classification results of multiple credit texts are set according to the division of entities and attributes, and the corresponding classification model is set, which can efficiently classify and store different types of credit data. The keywords and text big data are the data analyzed based on the semantic model in S101. Each keyword corresponds to a vector representation, and the semantic relationship between keywords can be analyzed through the cosine similarity of vectors.

[0062] In the embodiment, constructing semantic relationship can capture entity relationship model and update graph relationship model, mine potential relationship patterns of data, apply two dimensions to analyze relationship patterns of different types of credit data, and construct a decision tree classification model based on relationship pattern analysis, effectively improve the text classification, graph structure knowledge mining, and text recognition and extraction efficiency of tender big data files, and effectively reduce the computing power consumption of tender demand recognition and local file matching.

[0063] According to the embodiment of the present application, the S104 comprises: Extracting entity data in the knowledge base and associating corresponding keywords to obtain entity keywords; Taking the vector representation corresponding to the entity keyword as the analysis object, the similarity between vectors is calculated in the vector space through the cosine similarity, and the relationship strength of the entity keyword is evaluated based on the similarity, and the semantic relationship mode is set; According to the semantic relationship mode, the relationship data in the graph structure is expanded and applied to the data update of the knowledge base; Through the semantic relationship mode, the entity keywords with strong correlation characteristics are screened out, and the corresponding vector is taken as the keyword vector, and the corresponding classification information is associated with the keyword vector as the classification result; According to the keyword vector, a plurality of initial decision nodes are formed by conditional transformation, and the arrangement of the initial decision nodes is determined by the ID3 heuristic algorithm to construct a classification model based on the decision tree.

[0064] It can be understood here that the corresponding classification information is the entity classification information corresponding to the entity keyword with strong correlation characteristics. In the semantic relationship mode, the relationship strength between each entity keyword and other entity keywords is stored. The entity keyword with strong correlation characteristics has strong semantic relationship with multiple entity keywords. The entity keyword with strong correlation characteristics plays an important role in text recognition and tender data association. Therefore, the strong correlation entity keyword is screened to set the decision condition of the decision tree, and the real-time input tender demand text can be recognized and quickly classified by the file type in the subsequent. Through the classification result, efficient tender demand file extraction and document construction are further realized, and automatic demand analysis and efficient local file matching are realized.

[0065] The decision tree model can be established in the cloud large model.

[0066] The semantic relationship mode is the relationship information based on the word2vec model for semantic evaluation, which is used to analyze the semantic relationship between entity data (different credit file types). In the original knowledge graph structure, the relationship model of the file association level is included. Through the semantic relationship mode, the relationship data of the semantic level is expanded. Through the relationship mode of two dimensions, the potential association mode between entities and credit types can be effectively mined, so as to improve the tender text recognition ability and classification effect based on the knowledge base. In addition, in the construction of the decision tree model, the present application utilizes semantic association analysis to set the entity keyword with strong correlation, that is, to screen out the credit file type with high semantic correlation characteristics and set it as a decision node, so as to improve the recognition accuracy and efficiency of the tender text big data and real-time demand text, and further perform tender file classification and efficient matching.

[0067] According to the embodiment of the present application, the S105, in particular: The knowledge in the knowledge base is expanded and the relationship information is added through the AMIE algorithm. According to the graph structure in the knowledge base, the entity information of the credit data is analyzed and the corresponding entity word vector and classification information are obtained. According to the entity word vector and the classification information, the decision training data is constructed, the decision training data is imported into the classification model for classification training, the recall rate of the classification result is evaluated to evaluate the model effect, and the ID3 algorithm is used to adjust the layout and the number of the decision nodes.

[0068] It can be understood here that the knowledge base includes corresponding graph structure, and the graph structure includes corresponding entity, attribute data and relationship mode information, etc. The entity word vector is obtained through the word2vec model.

[0069] According to the embodiment of the present application, the local credit file library establishes data connection with the cloud model through a preset protocol.

[0070] It can be understood here that the user interaction terminal is used for visualizing the information such as the bidding file recognition result, the classification result, the standard file type specification array, etc. The preset encryption protocol can be used in the data connection process to ensure the data security.

[0071] According to the embodiment of the present application, the preset protocol includes a transmission encryption and storage encryption protocol.

[0072] The third aspect of the present application further provides a computer readable storage medium, wherein the computer readable storage medium includes a large model based bidding demand analysis and local file matching program, and when the large model based bidding demand analysis and local file matching program is executed by a processor, the steps of the large model based bidding demand analysis and local file matching method according to any one of the above are realized.

[0073] The application discloses a tender demand analysis and local file matching method and system based on a large model, which comprises the following steps: establishing a local credit file library, classifying credit big data by using a semantic model, extracting data to construct a knowledge base, and uploading the cloud large model; inputting the tender file into the cloud large model to identify and analyze the business demand, combining the knowledge base to extract and standardize the demand file, packing and downloading the corresponding file; when constructing the knowledge base, taking the graph structure as the classification mode, using the word2vec model to construct the word vector model to capture the semantic relationship; mapping the semantic relationship to the graph entity relationship model, adding relationship data, and screening the keyword vector to construct the decision tree classification model; obtaining the credit big data entity word vector and classification information to construct decision training data according to the graph structure, importing the classification model for training, and realizing the rapid identification, information classification and qualification analysis of the tender text through the classification model, thereby effectively improving the file matching efficiency.

[0074] In the above embodiments, the implementation can be achieved by software, hardware, firmware or any combination thereof, in whole or in part. When implemented by software, the implementation can be achieved in the form of a computer program product in whole or in part. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application can be generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, data user line) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as digital versatile disc, or semiconductor media (such as solid state disk), etc. In various embodiments of the present application, if there is no special description and logical conflict, the terms and / or descriptions of different embodiments are consistent and can be referred to each other, and the technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationship.

[0075] In the embodiments of the present application, "at least one" means one or more, and "multiple" means two or two more. The "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent: A exists alone, A and B exist together, and B exists alone, wherein A and B can be singular or plural. In the textual description of the embodiments of the present application, the character " / " generally represents that the front and rear associated objects are in an "or" relationship. In the present application, "first", "second", and various number designations are only for the convenience of description and do not limit the scope of the embodiments of the present application. For example, to distinguish different messages, etc., and are not used to describe a specific order or sequence.

[0076] It can be understood that the various number designations involved in the embodiments of the present application are only for the convenience of description and do not limit the scope of the embodiments of the present application. The size of the serial number of the above processes does not mean the order of execution, and the execution order of the processes should be determined by its function and inherent logic.

[0077] Finally, it should be noted that: the above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this, any change or replacement within the technical scope disclosed in the present application should be covered within the protection scope of the present application.

Claims

1. A large model-based bidding requirement analysis and local file matching method, characterized in that, Comprise: S101: Establish a local credit file library, classify the credit big data through the semantic model, extract entity, attribute and relationship data, and build a knowledge base, and upload the knowledge base to the cloud big model; S102: Input the bidding file to the cloud big model for business demand recognition analysis, extract the demand file combined with the knowledge base, and standardize the demand file to obtain a standard file type description array, and obtain the corresponding file list according to the array and package and download the file in the credit file library; S103: In the knowledge base construction, the graph structure of the knowledge base is used as the classification mode of the credit big data, the word data of the knowledge base is constructed into a vector model through the word2vec model, and the semantic relationship between the word vectors is captured; S104: The semantic relationship between the word vectors is mapped to the relationship model between the entities in the graph structure, and the relationship data in the graph is added based on the semantic relationship, and the key word vectors are selected based on the semantic relationship between the word vectors, and the key word vectors are used as the initial decision nodes to construct the classification model based on the decision tree; S105: According to the graph structure of the knowledge base, the entity word vector and the classification information of the credit big data are obtained, the decision training data is constructed, the decision training data is imported into the classification model, and the classification parameter is adjusted through the heuristic algorithm.

2. The method of claim 1, wherein the method is based on a large model. The S101, specifically: Establish a local credit file library, obtain credit big data from the credit file library, and convert the credit big data into text big data; Introduce the BERT semantic model, segment the text big data by the preset delimiter, classify the keywords according to various credit types, and extract entity, attribute and relationship data from the text big data to obtain triple information; In the triple information, the relationship data is specifically set according to the correlation of the text data of different credit types; A knowledge base is constructed by triplets, and the knowledge base is uploaded to the cloud big model.

3. The method of claim 1, wherein the method is based on a large model for bidding requirement analysis and local file matching. The S102, specifically: Input the bidding file to the cloud big model and perform business demand analysis to generate business demand keywords; Retrieval of demand file in knowledge base through business demand keywords, obtain demand file; According to the file standard in the credit file library, the demand file is standardized to generate a standard file type description array; According to the standard file type description array, obtain the file list and package and download the file in the credit file library.

4. The method of claim 1, wherein the method is based on a large model for bidding requirement analysis and local file matching. The S103 specifically: According to the graph structure of the knowledge data in the knowledge base, set the classification mode of the credit big data; Through the semantic model and the knowledge base, obtain the keywords and text big data of the credit big data, use the text big data as a corpus, learn the context relationship of the keywords based on the word2vec model, construct a vector space based on the semantic relationship according to the keywords, and assign a vector to each keyword through the vector space.

5. The method according to claim 4, wherein, The S104, comprising: Extract entity data in the knowledge base and associate corresponding keywords to obtain entity keywords; The vector corresponding to the entity keyword is taken as an analysis object, the similarity between vectors is calculated in a vector space through cosine similarity, and the relationship strength of the entity keyword is evaluated based on the similarity, and a semantic relationship mode is set; According to the semantic relationship mode, the relationship data in the graph structure is expanded and applied to the data updating of the knowledge base; Through the semantic relationship mode, the entity keywords with strong correlation characteristics are screened out, the corresponding vector is taken as a keyword vector, and the corresponding classification information is associated with the keyword vector as a classification result; According to the keyword vector, a plurality of initial decision nodes are formed by conditional transformation, and the arrangement of the initial decision nodes is determined by the ID3 heuristic algorithm to construct a classification model based on a decision tree.

6. The large model-based bidding requirement analysis and local file matching method according to claim 5, characterized in that, The S105, specifically: The AMIE algorithm is used to expand the knowledge and add relationship information to the graph structure in the knowledge base; According to the graph structure in the knowledge base, the entity information of the credit big data is analyzed to obtain the corresponding entity word vector and classification information; According to the entity word vector and the classification information, a decision training data is constructed, the decision training data is imported into the classification model for classification training, the recall rate of the classification result is used to evaluate the model effect, and the ID3 algorithm is used to adjust the layout and quantity of the decision nodes.

7. The method according to claim 1, wherein, The local credit file library establishes a data connection with the cloud big model through a preset protocol.

8. The large model-based bidding requirement analysis and local file matching method according to claim 7, characterized in that, The preset protocol includes a transmission encryption and storage encryption protocol. 9.A large model-based bidding requirement analysis and local file matching system, characterized in that, The system includes a memory, a processor, and a data interface, and the memory includes a big model-based bidding demand analysis and local file matching program, which is executed by the processor to realize the following steps: S101: Establish a local credit file library, classify the text data of the credit big data through a semantic model, extract entity, attribute, and relationship data, and construct a knowledge base, and upload the knowledge base to a cloud big model; S102: Input the bidding file into the cloud big model for commercial demand recognition analysis, extract the demand file in combination with the knowledge base, standardize the demand file, obtain a standard file type description array, and download the file package according to the array and the credit file library; S103: In the knowledge base construction, the graph structure of the knowledge base is taken as the classification mode of the credit big data, the word data of the knowledge base is constructed into a vector model through a word2vec model, and the semantic relationship between the word vectors is captured; S104: The semantic relationship between the word vectors is mapped to the relationship model between the entities in the graph structure, and the relationship data in the graph is added based on the semantic relationship, the keyword vector is screened out based on the semantic relationship between the word vectors, the keyword vector is taken as an initial decision node, and a classification model based on a decision tree is constructed; S105: According to the graph structure in the knowledge base, the entity word vector and the classification information of the credit big data are obtained, the decision training data is constructed, the decision training data is imported into the classification model, and the classification parameter is adjusted through the heuristic algorithm.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium comprises a large model-based bidding requirement analysis and local file matching program, and when the large model-based bidding requirement analysis and local file matching program is executed by the processor, the steps of the large model-based bidding requirement analysis and local file matching method according to any one of claims 1 to 8 are realized.

Images