An intelligent management method and system for bidding information based on artificial intelligence

Abstract

This application provides an artificial intelligence-based bidding information intelligent management method and system, which obtains bidding document data by collecting all public bidding documents, and then constructs a label reference tree based on the bidding document data; selects a bidding document, extracts bidding word features and bidding sentence features, and then obtains the bidding confidence features of the bidding document; obtains the label decoupling chain of the bidding document based on the bidding confidence features; obtains the bidding information of the target company, obtains the matching credibility between the bidding document and the bidding information through the label decoupling chain, repeats the above steps, and continues to obtain the matching credibility between the remaining bidding documents and the bidding information; when matching the target company's bid, pushes the matching bidding document to the target company based on all matching credibility. The above scheme can realize the hierarchical label classification of bidding documents, thereby improving the accuracy of pushing bidding information of the target company.

Description

Technical Field

[0001] The present application relates to the field of information management technology, and more specifically, to an artificial intelligence-based intelligent bidding information management method and system. Background Art

[0002] Information management refers to the systematic, standardized and effective management of all aspects of information collection, processing, storage, transmission, retrieval and utilization. Intelligent information management is a method that combines traditional information management with advanced technologies such as artificial intelligence, machine learning, big data analysis, and natural language processing to improve the automation, intelligence and efficiency of information management.

[0003] Intelligent management of bidding information is a method of applying advanced technologies such as artificial intelligence and big data to the bidding process to improve efficiency, transparency and decision-making accuracy. The bidding information retrieval and analysis in intelligent management of bidding information can use big data technology to analyze the collected bidding data from previous years, quickly retrieve relevant bidding information, and help companies accurately find suitable bidding projects. However, in the existing technology, the intelligent management system of bidding information can only identify the key entities in the bidding documents, and then simply classify the bidding information according to the identified key entities to realize the push of bidding information. It does not conduct in-depth analysis and classification of the collected bidding information, resulting in low accuracy and lack of practicality of the pushed bidding information. Therefore, how to realize the hierarchical label classification of bidding documents to improve the accuracy of bidding information push for target companies is a difficult problem faced by the industry. Summary of the Invention

[0004] This application provides an artificial intelligence-based intelligent bidding information management method and system, which can realize hierarchical label classification of bidding documents, thereby improving the accuracy of bidding information push to target companies.

[0005] In a first aspect, the present application provides an artificial intelligence-based intelligent bidding information management method, comprising:

[0006] Collecting all public bidding documents to obtain bidding document data, and constructing a label reference tree for classifying all bidding documents in the bidding document data based on bidding key information in the bidding document data;

[0007] A tender document is selected from the collected tender documents, tender word features and tender sentence features are extracted from the selected tender document, and feature mutual information between the tender word features and the tender sentence features is determined. Based on the feature mutual information, the tender word features and the tender sentence features are confidence-fused to obtain a tender confidence feature of the tender document.

[0008] performing label decoupling on the label reference tree according to the bidding confidence feature of the bidding document and all labels of the bidding document to obtain a label decoupling chain of the bidding document in the label reference tree;

[0009] Obtain the target company's bidding information, perform label matching on the bidding information through the label decoupling chain, obtain the matching credibility between the bidding document and the bidding information, repeat the above steps to continue to obtain the matching credibility between the remaining bidding documents and the bidding information;

[0010] When matching bids to target companies, matching bidding documents are pushed to the target companies based on all matching credibility.

[0011] In some embodiments, extracting bidding word features and bidding sentence features from the selected bidding document specifically includes:

[0012] Extracting multiple word tags from the bidding document to obtain bidding word features;

[0013] The bidding document is divided into a plurality of clause units, thereby obtaining bidding sentence features.

[0014] In some embodiments, determining the feature mutual information between the tender word feature and the tender sentence feature specifically includes:

[0015] Determining a weight matrix for feature fusion according to the bidding word features and the bidding sentence features;

[0016] The feature mutual information between the tender word feature and the tender sentence feature is determined based on the weight matrix.

[0017] In some embodiments, performing confidence fusion on the tender word features and the tender sentence features based on the feature mutual information to obtain the tender confidence features of the tender document specifically includes:

[0018] Merging the tender word features and the tender sentence features according to the feature mutual information to obtain candidate features of the tender document;

[0019] A bidding confidence feature is determined based on the candidate features.

[0020] In some embodiments, constructing a label reference tree for classifying all tender documents in the tender document data according to the tender key information in the tender document data specifically includes:

[0021] Get the label hierarchy of bidding information;

[0022] Extracting key bidding information from the bidding document data;

[0023] Extracting labels from the bidding key information to obtain a bidding label set;

[0024] The label hierarchical structure and the tender label set are merged to obtain a label reference tree for classifying all tender documents in the tender document data.

[0025] In some embodiments, performing label decoupling on the label reference tree according to the bidding confidence feature of the bidding document and all labels of the bidding document to obtain a label decoupling chain of the bidding document in the label reference tree specifically includes:

[0026] determining a label distribution domain of the bidding document according to the label reference tree and the bidding confidence feature;

[0027] A tag decoupling chain of all tags of the bidding document in the tag reference tree is determined based on the tag distribution domain.

[0028] In some embodiments, pushing matching bidding documents to the target company based on all matching credibility specifically includes:

[0029] Determine the matching bidding document set suitable for the target company to bid based on all matching confidence levels;

[0030] All the bidding documents in the matching bidding document set are pushed to the target company.

[0031] In a second aspect, the present application provides an artificial intelligence-based bidding information intelligent management system, comprising:

[0032] A collection module is used to collect all public bidding documents to obtain bidding document data, and to construct a label reference tree for classifying all bidding documents in the bidding document data based on the bidding key information in the bidding document data;

[0033] a processing module, configured to select a tender document from the collected tender documents, extract tender word features and tender sentence features from the selected tender document, further determine feature mutual information between the tender word features and the tender sentence features, and perform confidence fusion on the tender word features and the tender sentence features based on the feature mutual information to obtain a tender confidence feature of the tender document;

[0034] The processing module is further configured to perform label decoupling on the label reference tree according to the bidding confidence feature of the bidding document and all labels of the bidding document, to obtain a label decoupling chain of the bidding document in the label reference tree;

[0035] The processing module is further configured to obtain the target company's bidding information, perform label matching on the bidding information through the label decoupling chain, obtain a matching credibility between the bidding document and the bidding information, and repeat the above steps to continue obtaining matching credibility between the remaining bidding documents and the bidding information.

[0036] The execution module is used to push the matching bidding documents to the target company according to all matching credibility when matching the bids of the target company.

[0037] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned artificial intelligence-based intelligent management method for bidding information.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned artificial intelligence-based intelligent management method for bidding information.

[0039] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0040] The artificial intelligence-based bidding information intelligent management method and system provided in the present application first collects all public bidding documents to obtain bidding document data, and then constructs a label reference tree for classifying all bidding documents in the bidding document data based on bidding key information in the bidding document data; selects a bidding document from the collected bidding documents, extracts bidding word features and bidding sentence features from the selected bidding document, and then determines feature mutual information between the bidding word features and the bidding sentence features; performs confidence fusion on the bidding word features and the bidding sentence features based on the feature mutual information to obtain bidding confidence features of the bidding document; performs label decoupling on the label reference tree based on the bidding confidence features of the bidding document and all labels of the bidding document to obtain a label decoupling chain of the bidding document in the label reference tree; obtains bidding information of a target company, performs label matching on the bidding information using the label decoupling chain to obtain a matching confidence level between the bidding document and the bidding information, and repeats the above steps to continue to obtain matching confidence levels between the remaining bidding documents and the bidding information; when matching bids for the target company, pushes matching bidding documents to the target company based on all matching confidence levels.

[0041] It can be seen that in this application, the bidding information is labeled matched through the label decoupling chain to obtain the matching credibility between the bidding document and the bidding information. When the target company is matched with the bid, the matching bidding document is pushed to the target company according to the matching credibility. Among them, the bidding confidence feature is determined by feature fusion to obtain the text features captured from different scales in the bidding document. The above steps comprehensively consider the various feature information extracted from the bidding document, which helps to improve the understanding of the bidding document and the matching accuracy in the subsequent label decoupling and matching process; then, the label reference tree construction process can obtain the inclusion in the classification process of the bidding document through the bidding key information. The above steps can better identify the relationship and hierarchy between labels and the label set of corresponding keywords in the bidding industry, ensure the accuracy and effectiveness of the decoupling process, so that the position of the bidding document in the label decoupling chain is clearer, which is helpful for subsequent label matching; the label decoupling chain obtained by label decoupling can determine the label path describing the mutual dependency relationship between multiple labels in the bidding document. The above steps can refine and decompose the labels in the label reference tree, thereby identifying more specific label associations, and thus improving the credibility and accuracy of the matching results; in summary, this application can realize the hierarchical label classification of bidding documents, thereby improving the accuracy of the target company's bidding information push. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is an exemplary flow chart of an artificial intelligence-based intelligent management method for bidding information according to some embodiments of the present application;

[0043] Figure 2 is an exemplary flow chart of determining a tag reference tree according to some embodiments of the present application;

[0044] Figure 3 is an exemplary flow chart of determining a tag decoupling chain according to some embodiments of the present application;

[0045] Figure 4 is a schematic diagram of exemplary hardware and / or software of an artificial intelligence-based bidding information intelligent management system according to some embodiments of the present application;

[0046] Figure 5 It is a structural diagram of a computer device for implementing an artificial intelligence-based intelligent management method for bidding information according to some embodiments of the present application. DETAILED DESCRIPTION

[0047] The embodiment of the present application provides an artificial intelligence-based intelligent management method and system for bidding information, the core of which is to collect all public bidding documents, obtain bidding document data, and then build a label reference tree based on the bidding document data; select a bidding document, extract bidding word features and bidding sentence features, and then obtain the bidding confidence features of the bidding document; obtain a label decoupling chain of the bidding document based on the bidding confidence features; obtain the bidding information of the target company, obtain the matching credibility between the bidding document and the bidding information through the label decoupling chain, repeat the above steps to continue to obtain the matching credibility between the remaining bidding documents and the bidding information; when matching the bids of the target company, push the matching bidding documents to the target company based on all matching credibility, which can realize hierarchical label classification of the bidding documents, thereby improving the accuracy of pushing the bidding information of the target company.

[0048] In order to better understand the above technical solution, the following will be combined with the accompanying drawings and specific implementation methods to describe the above technical solution in detail. Figure 1 , which is an exemplary flow chart of an artificial intelligence-based bidding information intelligent management method according to some embodiments of the present application. The artificial intelligence-based bidding information intelligent management method 100 mainly includes the following steps:

[0049] In step 101, all public bidding documents are collected to obtain bidding document data, and a label reference tree is constructed for classifying all bidding documents in the bidding document data based on bidding key information in the bidding document data.

[0050] It should be noted that in this application, the bidding document in this application is a Hypertext Markup Language (HTML) document of the bidding notice published on the bidding website, which details the relevant information of the corresponding bidding project; the bidding document data refers to the collection of all bidding documents after data cleaning. In specific implementation, all public bidding documents are collected to obtain bidding document data, which can be achieved in the following way, namely: first, crawler technology can be used to collect HTML documents of all bidding notices published on each bidding website at a preset collection time interval (the default is 12 hours), and the most recently collected, for example, 200 HTML documents are used as all bidding documents; then all collected bidding documents can be cleaned, and the collection of all bidding documents after data cleaning is used as bidding document data, where data cleaning includes deduplication and deletion of expired documents, which is not specifically limited here.

[0051] In some embodiments, reference Figure 2This figure is an exemplary flow chart of determining a label reference tree according to some embodiments of the present application. In the present application, the label reference tree for classifying all the tender documents in the tender document data based on the tender key information in the tender document data can be implemented by the following steps:

[0052] In step 1011, a label hierarchy of the bidding information is obtained;

[0053] In step 1012, key bidding information is extracted from the bidding document data;

[0054] In step 1013, the bidding key information is subjected to label extraction to obtain a bidding label set;

[0055] In step 1014, the label hierarchical structure and the tender label set are merged to obtain a label reference tree for classifying all tender documents in the tender document data.

[0056] It should be noted that in this application, the label reference tree is a tree-structured label set, which includes multiple classification labels in the bidding industry and multiple keywords extracted from the bidding key information; the label hierarchy is a nested structure of multiple labels in the bidding document, and the label is used to identify and describe different aspects of the bidding information in each bidding project. In other embodiments, the label hierarchy may also include other labels in the bidding document, which is not limited here; the common label hierarchy in bidding information may include: project information (which includes project name, project description, project location, project scale), bidding unit information (which includes unit name, contact information, address, contact person), bidding content (which includes bidding scope, bidding type, bidding document collection method, bidding deadline, opening time), and the bidding information can be better organized and presented through this label hierarchy; the bidding label set is a set composed of multiple keywords that best represent the bidding key information, and there is only one of each keyword in the set.

[0057] In specific implementation, first, obtaining the label hierarchy of the bidding information can be implemented in the following manner, namely: obtaining a bidding document with bookmarks, extracting the bookmark hierarchy of the bidding document, and using the bookmark hierarchy as the label hierarchy of the bidding information; secondly, extracting bidding key information from the bidding document data; then, performing label extraction on the bidding key information, and then obtaining a bidding label set can be implemented in the following manner, namely: first presetting a word frequency threshold, then determining the word frequency of each keyword in the bidding key information, and then taking a set of all keywords whose word frequency is greater than or equal to the word frequency threshold as a bidding label set, wherein, for each keyword, the ratio of the number of times the keyword appears in the bidding key information to the total number of keywords in the bidding key information can be used as the word frequency of the keyword. In addition, the word frequency threshold can be preset according to the specific situation. For example, in this application, the word frequency threshold is set to 5; finally, the label hierarchy and the bidding label are combined. The tag set is merged to obtain a tag reference tree for classifying all the bidding documents in the bidding document data. This can be achieved in the following manner: first, the tag hierarchical structure is converted into a tree structure to obtain a tree-like tag hierarchical structure; a keyword is selected from the bidding tag set; a tag matching the keyword is determined in the tree-like tag hierarchical structure; and the keyword is placed in the next layer of the tag as a subtag of the tag; if there is no tag matching the tag type, the keyword is not processed; the above steps are repeated to continue adding the remaining keywords in the bidding tag set to the tree-like tag hierarchical structure; and finally, the new tree-like tag hierarchical structure is used as the tag reference tree. Whether a keyword and a tag match can be judged according to the tag classification standard of the bidding industry. For example, Nanjing (keyword) matches the address (tag) under the bidding unit information (tag), but does not match the contact person (tag) under the bidding unit information (tag).

[0058] Preferably, in the above embodiment, extracting the bidding key information from the bidding document data can be achieved by adopting the following steps:

[0059] Segmenting all bidding documents in the bidding document data to obtain key bidding data;

[0060] The key bidding information is determined according to the key bidding data.

[0061] It should be noted that in this application, the key tender information is a set composed of keywords of multiple tender information; the key tender data is a set composed of tender information of all tender documents in the tender document data. Among them, the tender information includes: tender project name, tender unit name, tender unit contact information, administrative region to which the tender unit belongs, tender project category, bid deadline, opening date, opening place, bid bond, project budget amount, project description, and project requirements.

[0062] When specifically implemented, the key tender data can be obtained by splitting all the tender documents in the tender document data in the following way: first, use an HTML parsing tool to split each tender document in the tender document data, and extract multiple tender information in the tender document, and then use the set composed of all the extracted tender information as the key tender data. Among them, there are multiple implementation methods for splitting the tender document using an HTML parsing tool, which can be selected according to specific circumstances. For example, in this application, the position of each tender information in the tender document is first determined, and then the tender document is split according to the position of each tender information. In addition, the HTML parsing tool can be selected according to specific circumstances. For example, the HTML parsing tool used in this application is the BeautifulSoup library function in Python; the key tender information can be determined according to the key tender data in the following way: first, select a tender information in the key tender data, use a Chinese word segmentation tool to divide the tender information into multiple individual words, and then remove the high-frequency words without practical meaning in the tender information, such as "de", "le", "he", and other function words without specific meanings, and use the remaining words as the keywords of the tender information. If there is only one individual word in the tender information, directly use the individual word as the keyword of the tender information. Finally, repeat the above steps to continue dividing the remaining tender information in the key tender data, and use the set composed of the keywords of all the obtained tender information as the key tender information. Among them, the Chinese word segmentation tool can be selected according to specific circumstances. For example, the jieba library function in Python is used in this application.

[0063] In step 102, select a tender document from the collected tender documents, extract the tender word features and tender sentence features of the selected tender document, and then determine the feature mutual information between the tender word features and the tender sentence features. Based on the feature mutual information, perform confidence fusion on the tender word features and the tender sentence features to obtain the tender confidence feature of the tender document.

[0064] In some embodiments, the extraction of the tender word features and the tender sentence features of the selected tender document can be implemented in the following steps:

[0065] Extracting multiple word tags from the bidding document to obtain bidding word features;

[0066] The bidding document is divided into a plurality of clause units, thereby obtaining bidding sentence features.

[0067] It should be noted that in this application, the bidding word feature refers to the word vector extracted from the bidding document; the bidding sentence feature refers to the sentence vector extracted from the bidding document, which captures the semantic information of all sentences in the bidding document.

[0068] In a specific implementation, multiple word tags are extracted from the tender document, and the tender word features can be obtained by the following method: first, the tender document is converted into a text format, and special characters and stop words in the document are removed, and then the processed tender document is input into a word segmenter (for example, a WordPiece word segmenter). The word segmenter will divide the text into multiple subword units, and then each subword unit is used as a tag. Each tag can be arranged in the order of its position in the tender document to obtain a sequence as a tag index table, and then the attention mask and sentence type identifier of each tag in the tag index table are determined, and all the obtained attention masks and sentence type identifiers are respectively composed of corresponding sequences according to the order of each tag in the tag index table. Then, the three sequences of the tag index table, the attention mask sequence and the sentence type identifier sequence are passed as input to the tensor function in PyTorch, outputting a matrix, and the output matrix is ​​input into a bidirectional encoder (BERT, Bidirectional Encoder Representations From Transformers) model, the output of the BERT model is obtained. The output result usually includes three parts: the hidden state of the last layer, the output of the pooling, and the hidden states of all layers. The vector output by the hidden state of the last layer can be used as the bidding word feature of the bidding document.

[0069] It should be noted that when processing bidding documents, the lengths of different bidding documents may be different. In order to make all bidding documents have the same length in a batch, it is usually necessary to fill the shorter bidding documents. The attention mask represents the marking information of the information type, which includes actual information and meaningless information, where the attention mask corresponding to the actual information is 1, the attention mask corresponding to the meaningless information is 0, and the attention mask corresponding to the meaningless information is 0; the sentence type identifier is a parameter used to distinguish the sentence type. The serial number of each sentence in the bidding document can be used as the sentence type identifier. For example, the first sentence from top to bottom in the bidding document corresponds to the marked sentence type identifier 1, and the second sentence corresponds to the marked sentence type identifier 2.

[0070] In specific implementation, the bidding document is split into multiple clause units, and then the bidding sentence features are obtained in the following way, namely: first convert the bidding document into text format, and use the existing natural language processing library to split the text into multiple sentences, arrange all sentences in the order in the text as a sentence label sequence, initialize a trained Doc2Vec model, use the sentence label sequence as the input parameter of the Doc2Vec model, use the Doc2Vec model to generate a vector for each sentence in the sentence label sequence, and use the set of all vectors as the bidding sentence features of the bidding document.

[0071] In some embodiments, determining the feature mutual information between the tender word feature and the tender sentence feature may be achieved by using the following steps:

[0072] Determining a weight matrix for feature fusion according to the bidding word features and the bidding sentence features;

[0073] The feature mutual information between the tender word feature and the tender sentence feature is determined based on the weight matrix.

[0074] It should be noted that in this application, the weight matrix is ​​a two-dimensional matrix composed of the weights of the tender word features and the tender sentence features when they are fused; the feature mutual information is a parameter that represents the degree of correlation between the tender word features and the tender sentence features. The larger the feature mutual information, the greater the correlation between the corresponding tender word features and the tender sentence features. At this time, it means that the extraction effect of the tender word features and the tender sentence features is better, that is, the greater the confidence of the candidate features obtained in the subsequent feature fusion step, and the value range of the feature mutual information is between 0 and 1.

[0075] In specific implementation, the weight matrix for feature fusion is determined based on the tender word features and the tender sentence features, which can be implemented in the following manner, namely: the number of features to be fused is used as the input dimension of the weight matrix. In this application, there are two input dimensions, namely the tender word features and the tender sentence features, and the number of features obtained after fusion is used as the output dimension of the weight matrix. In this application, there is only one feature vector obtained after fusion, so the weight matrix is ​​a 2*1 matrix. Subsequently, the square root of the reciprocal of the ratio of the sum of the input dimension and the output dimension of the weight matrix to six is ​​taken to obtain a value A. Finally, two random numbers are generated in the interval [-A, A] according to a uniform distribution as the values ​​in the weight matrix. elements to obtain a weight matrix; determining the feature mutual information between the tender word feature and the tender sentence feature based on the weight matrix can be achieved in the following manner, namely: first, using a linear change through a fully connected layer in a neural network to transform the tender word feature and the tender sentence feature into the same dimension, and forming the tender word feature and the tender sentence feature into a two-dimensional matrix, and then taking the inverse of the product of the two-dimensional matrix and the weight matrix as the power of a natural constant, and taking the inverse of the value obtained by adding one as the feature mutual information between the tender word feature and the tender sentence feature, wherein, when the two-dimensional matrix is ​​multiplied by the weight matrix, the two-dimensional matrix is ​​used as a row vector, and the weight matrix is ​​used as a column vector.

[0076] In some embodiments, the confidence fusion of the tender word features and the tender sentence features based on the feature mutual information to obtain the tender confidence features of the tender document can be achieved by the following steps:

[0077] Merging the tender word features and the tender sentence features according to the feature mutual information to obtain candidate features of the tender document;

[0078] A bidding confidence feature is determined based on the candidate features.

[0079] It should be noted that in this application, the bidding confidence feature is the text feature of the bidding document captured from different scales. The bidding confidence feature can be used to better classify the bidding document and improve the classification performance; the candidate feature is an intermediate variable for determining the bidding confidence feature and has no practical significance. In some preferred embodiments, the bidding word feature and the bidding sentence feature are merged according to the feature mutual information to obtain the candidate feature of the bidding document. This can be achieved in the following manner: first, the difference between the feature mutual information and one is multiplied by the bidding sentence feature, and then the obtained vector is added to the product of the feature mutual information and the bidding word feature, and then the obtained vector is multiplied by the weight matrix in the above step, and then the obtained vector is used as the input of the hyperbolic tangent function, and the corresponding output vector is used as the candidate feature; determining the bidding confidence feature based on the candidate feature can be achieved in the following manner: first, the difference between the modulus of the candidate feature and one is multiplied by the bidding sentence feature, and the obtained vector is added to the product of the modulus of the candidate feature and the bidding word feature, and then the obtained vector is used as the bidding confidence feature.

[0080] In step 103, the tag reference tree is tag decoupled according to the bidding confidence feature of the bidding document and all tags of the bidding document to obtain a tag decoupling chain of the bidding document in the tag reference tree.

[0081] In some embodiments, reference Figure 3 This figure is an exemplary flow chart of determining a label decoupling chain according to some embodiments of the present application. In the present application, label decoupling is performed on the label reference tree based on the bidding confidence feature of the bidding document and all labels of the bidding document. Obtaining the label decoupling chain of the bidding document in the label reference tree specifically includes:

[0082] In step 1031, a label distribution domain of the bidding document is determined based on the label reference tree and the bidding confidence feature;

[0083] In step 1032, a tag decoupling chain of all tags of the bidding document in the tag reference tree is determined based on the tag distribution domain.

[0084] It should be noted that in this application, the label decoupling chain is a label path that describes the mutual dependency relationship between multiple class labels in the bidding document. Through the label decoupling chain, all labels corresponding to the document can be found in the label reference tree. The label contains the parent label and child label on the corresponding label path in the label reference tree; the label distribution domain is a probability matrix (i.e., a probability distribution matrix), and the element in the i-th row and j-th column of the matrix represents the probability that the i-th feature in the bidding confidence feature matches the j-th label in the label reference tree;

[0085] In specific implementation, the label distribution domain of the bidding document is determined according to the label reference tree and the bidding confidence feature. This can be achieved in the following manner: first, the new neural architecture Graphormer in the prior art is used to represent the label reference tree as a matrix, and the matrix is ​​used as the label feature matrix. Then, the bidding confidence feature is converted into a matrix form of the same size as the label feature matrix (that is, the values ​​in the bidding confidence feature are filled in the corresponding positions in the matrix, and other places without actual values ​​in the matrix are replaced by zeros), and the matrix is ​​used as the feature matrix of the bidding confidence feature. Subsequently, the similarity between each label in the label reference tree and each feature in the bidding confidence feature is obtained, and all the obtained similarities are filled in the matrix to obtain a similarity matrix. The element in the i-th row and j-th column of the similarity matrix represents the bidding position. The similarity between the i-th feature in the confidence feature and the j-th label in the label reference tree is calculated, and finally the sum of the values ​​obtained by raising all element values ​​in the similarity matrix as powers of natural constants is calculated, and then an element value is selected in the similarity matrix, and the element value is raised as the power of the natural constant, and the obtained value is divided by the sum value to obtain a quotient, and the obtained quotient is filled in a new matrix, and the position of the quotient in the new matrix is ​​the same as the position of the element value in the similarity matrix. Repeat the above steps, fill the quotients corresponding to the remaining element values ​​in the similarity matrix into the new matrix, and use the new matrix as the label distribution domain of the bidding document, wherein the similarity between each label and each feature can be calculated by calculating the dot product between the row vector corresponding to the feature in the feature matrix of the bidding confidence feature and the column vector corresponding to the label in the label feature matrix.

[0086] In addition, in the specific implementation, the label decoupling chain of all labels of the bidding document in the label reference tree based on the label distribution domain can be implemented in the following manner, namely: first set a threshold, select all element values ​​in the label distribution domain that are greater than the threshold, and determine the label corresponding to each element value in the label distribution domain, then select a label, find the position of the label in the label reference tree, and use the path from the root of the tree to the position of the label as the label path, and repeat the above steps to continue to determine the label path of the remaining labels, and finally integrate all the label paths, that is, merge the overlapping paths to obtain a tree-like label path, and use the tree-like label path as the label decoupling chain of all labels of the bidding document in the label reference tree, wherein the value of the threshold can be set according to actual conditions. For example, in this application, the threshold is set to 0.5.

[0087] In step 104, the bidding information of the target company is obtained, and the bidding information is label matched through the label decoupling chain to obtain the matching credibility between the bidding document and the bidding information. The above steps are repeated to continue to obtain the matching credibility between the remaining bidding documents and the bidding information.

[0088] In specific implementation, the bid information of the target company can be obtained in the following manner, namely: when the target company uses the system, the basic information of the target company is collected through the customer registration process, and the basic information is used as the bid information of the target company.

[0089] It should be noted that the bidding information in this application is some basic information related to the target company that the target company will use when bidding. The basic information may include: company name, establishment time, registered capital, company address, main business scope, company's available working capital, qualification certificate and number, and related honors.

[0090] In some embodiments, label matching is performed on the bidding information through the label decoupling chain to obtain the matching credibility between the bidding document and the bidding information. This can be achieved by using the following steps:

[0091] Comparing the bidding information with the labels in the label decoupling chain to obtain a matching label set;

[0092] The matching credibility between the bidding document and the bidding information is determined through the matching tag set.

[0093] It should be noted that in this application, the matching credibility is a parameter that represents the degree of correlation between the bidding document and the bid information. The greater the matching credibility, the greater the correlation between the corresponding bidding document and the bid information, and the more suitable the corresponding bidding document is for the target company to bid. The value range of the matching credibility is between 0 and 1; the matching label set is a set of multiple labels whose label content is consistent with the bid information.

[0094] In specific implementation, first, the bidding information is compared with the labels in the label decoupling chain to obtain a matching label set, which can be implemented in the following manner, namely: judging whether the label content is consistent with the bidding information, and taking the set consisting of all labels in the label content that is consistent with the bidding information as the matching label set, wherein the conditions for whether the label content is consistent with the bidding information can be set according to actual conditions. For example, in this application, it is required that the company address of the target company is the same as the administrative area to which the bidding unit in the bidding document belongs, the main business scope of the target company is the same as the bidding project category in the bidding document or is included in the relationship, the company's available working capital of the target company is greater than the sum of the bid deposit and the project budget in the bidding document, and the target company's qualification certificate and related honors meet the project requirements in the bidding document; secondly, determining the matching credibility between the bidding document and the bidding information through the matching label set can be implemented in the following manner, namely: first, taking the inverse of the total number of labels in the matching label set as the power of a natural constant, and then taking the reciprocal of the sum of the obtained value and one as the matching credibility between the bidding document and the bidding information.

[0095] In step 105, when matching bids for a target company, matching bidding documents are pushed to the target company based on all matching credibility.

[0096] In some embodiments, when matching bids for a target company, pushing matching tender documents to the target company based on all matching credibility levels can be achieved by using the following steps:

[0097] Determine the matching bidding document set suitable for the target company to bid based on all matching confidence levels;

[0098] All the bidding documents in the matching bidding document set are pushed to the target company.

[0099] It should be noted that in this application, the matching tender document set is a collection of multiple tender documents matched by the target company; in specific implementation, first, all tender documents with matching credibility higher than the preset matching threshold can be combined into a collection, and this collection can be used as the matching tender document set suitable for the target company's bid.

[0100] In addition, in another aspect of the present application, in some embodiments, the present application provides an artificial intelligence-based bidding information intelligent management system, referring to Figure 4 This figure is a schematic diagram of exemplary hardware and / or software of an artificial intelligence-based bidding information intelligent management system according to some embodiments of the present application. The artificial intelligence-based bidding information intelligent management system 400 includes: a collection module 401, a processing module 402, and an execution module 403, which are described as follows:

[0101] The acquisition module 401 in this application is mainly used to collect all public bidding documents to obtain bidding document data, and to construct a label reference tree for classifying all bidding documents in the bidding document data based on the bidding key information in the bidding document data;

[0102] Processing module 402, in this application, is primarily used to select a tender document from the collected tender documents, extract tender word features and tender sentence features from the selected tender document, further determine feature mutual information between the tender word features and the tender sentence features, and perform confidence fusion on the tender word features and the tender sentence features based on the feature mutual information to obtain a tender confidence feature of the tender document;

[0103] It should be noted that the processing module 402 in this application is further configured to perform label decoupling on the label reference tree according to the tender confidence feature of the tender document and all labels of the tender document, to obtain a label decoupling chain of the tender document in the label reference tree;

[0104] In addition, it should be noted that the processing module 402 in this application is also used to obtain the bid information of the target company, perform label matching on the bid information through the label decoupling chain, obtain the matching credibility between the bidding document and the bid information, and repeat the above steps to continue to obtain the matching credibility between the remaining bidding documents and the bid information;

[0105] Execution module 403, in this application, execution module 403 is mainly used to push matching bidding documents to the target company according to all matching credibility when matching the target company's bids.

[0106] In addition, the present application also provides a computer device, which includes a memory and a processor, the memory stores code, and the processor is configured to obtain the code and execute the above-mentioned artificial intelligence-based intelligent management method for bidding information.

[0107] In some embodiments, reference Figure 5 , which is a schematic diagram of the structure of a computer device for implementing an artificial intelligence-based bidding information intelligent management method according to some embodiments of the present application. The artificial intelligence-based bidding information intelligent management method in the above embodiment can be Figure 5 The computer device 500 shown in FIG. 5 is implemented as shown in FIG. 5 . The computer device 500 includes at least one processor 501 , a communication bus 502 , a memory 503 , and at least one communication interface 504 .

[0108] The processor 501 may be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0109] The communication bus 502 may be used to transmit information between the aforementioned components.

[0110] The memory 503 may be a read-only memory (ROM) or other type of static storage device that can store static information and instructions, a random access memory (RAM) or other type of dynamic storage device that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, an optical disc storage (including a compact disc, laser disc, optical disc, digital versatile disc, Blu-ray disc, etc.), a magnetic disk or other magnetic storage device, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 503 may exist independently and be connected to the processor 501 via the communication bus 502. The memory 503 may also be integrated with the processor 501.

[0111] Memory 503 is used to store program code for executing the solution of the present application, and is controlled by processor 501 for execution. Processor 501 is used to execute the program code stored in memory 503. The program code may include one or more software modules. The artificial intelligence-based bidding information intelligent management method in the above embodiment can be implemented by processor 501 and one or more software modules in the program code in memory 503.

[0112] The communication interface 504 uses any transceiver or other device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0113] In a specific implementation, as an example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. The processor herein may refer to one or more devices, circuits, and / or processing cores for processing data (e.g., computer program instructions).

[0114] The aforementioned computer device may be a general-purpose computer device or a dedicated computer device. In a specific implementation, the computer device may be a desktop computer, a portable computer, a network server, a personal digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. The embodiments of the present application do not limit the type of computer device.

[0115] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned artificial intelligence-based intelligent management method for bidding information.

[0116] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present application.

[0117] Obviously, those skilled in the art may make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if such changes and modifications fall within the scope of the claims of the present application and their equivalents, the present application is intended to include such changes and modifications.

Claims

1. An intelligent management method for bidding information based on artificial intelligence, characterized in that: The steps include: Collecting all public bidding documents to obtain bidding document data, and constructing a label reference tree for classifying all bidding documents in the bidding document data based on bidding key information in the bidding document data; A tender document is selected from the collected tender documents, tender word features and tender sentence features are extracted from the selected tender document, and feature mutual information between the tender word features and the tender sentence features is determined. Based on the feature mutual information, the tender word features and the tender sentence features are confidence-fused to obtain a tender confidence feature of the tender document. The feature mutual information is a parameter that represents the degree of correlation between the tender word features and the tender sentence features. The greater the feature mutual information, the greater the degree of correlation between the corresponding tender word features and the tender sentence features, that is, the greater the confidence of the candidate features obtained in the subsequent feature fusion step; performing label decoupling on the label reference tree according to the bidding confidence feature of the bidding document and all labels of the bidding document to obtain a label decoupling chain of the bidding document in the label reference tree; The tag decoupling of the tag reference tree is performed according to the bidding confidence feature of the bidding document and all tags of the bidding document to obtain the tag decoupling chain of the bidding document in the tag reference tree. Specifically, the tag decoupling includes: determining a label distribution domain of the bidding document according to the label reference tree and the bidding confidence feature; Determining a tag decoupling chain of all tags of the bidding document in the tag reference tree based on the tag distribution domain; The label decoupling chain is a label path that describes the mutual dependency between multiple class labels in the bidding document; Obtain the target company's bidding information, perform label matching on the bidding information through the label decoupling chain, obtain the matching credibility between the bidding document and the bidding information, repeat the above steps to continue to obtain the matching credibility between the remaining bidding documents and the bidding information; The tag matching of the bidding information through the tag decoupling chain to obtain the matching credibility between the bidding document and the bidding information specifically includes: Comparing the bidding information with the labels in the label decoupling chain to obtain a matching label set; Determining the matching credibility between the bidding document and the bidding information through the matching tag set; When matching bids to target companies, matching bidding documents are pushed to the target companies based on all matching credibility.

2. The method according to claim 1, wherein Extracting bidding word features and bidding sentence features from the selected bidding documents specifically includes: Extracting multiple word tags from the bidding document to obtain bidding word features; The bidding document is divided into a plurality of clause units, thereby obtaining bidding sentence features.

3. The method according to claim 1, wherein Determining the feature mutual information between the bidding word feature and the bidding sentence feature specifically includes: Determining a weight matrix for feature fusion according to the bidding word features and the bidding sentence features; The feature mutual information between the tender word feature and the tender sentence feature is determined based on the weight matrix.

4. The method according to claim 1, wherein The bidding word features and the bidding sentence features are confidently fused based on the feature mutual information to obtain the bidding confidence features of the bidding document, which specifically include: Merging the tender word features and the tender sentence features according to the feature mutual information to obtain candidate features of the tender document; A bidding confidence feature is determined based on the candidate features.

5. The method according to claim 1, wherein Constructing a label reference tree for classifying all the tender documents in the tender document data according to the tender key information in the tender document data specifically includes: Get the label hierarchy of bidding information; Extracting key bidding information from the bidding document data; Extracting labels from the bidding key information to obtain a bidding label set; The label hierarchical structure and the tender label set are merged to obtain a label reference tree for classifying all tender documents in the tender document data.

6. The method according to claim 1, wherein The matching bidding documents pushed to the target company based on all matching credibility include: Determine the matching bidding document set suitable for the target company to bid based on all matching confidence levels; All the bidding documents in the matching bidding document set are pushed to the target company.

7. An artificial intelligence-based bidding information intelligent management system, which uses the method according to any one of claims 1 to 6 to perform intelligent management of bidding information, characterized in that: The system includes: A collection module is used to collect all public bidding documents to obtain bidding document data, and to construct a label reference tree for classifying all bidding documents in the bidding document data based on the bidding key information in the bidding document data; a processing module, configured to select a tender document from the collected tender documents, extract tender word features and tender sentence features from the selected tender document, further determine feature mutual information between the tender word features and the tender sentence features, and perform confidence fusion on the tender word features and the tender sentence features based on the feature mutual information to obtain a tender confidence feature of the tender document; The processing module is further configured to perform label decoupling on the label reference tree according to the bidding confidence feature of the bidding document and all labels of the bidding document, to obtain a label decoupling chain of the bidding document in the label reference tree; The processing module is further configured to obtain the target company's bidding information, perform label matching on the bidding information through the label decoupling chain, obtain a matching credibility between the bidding document and the bidding information, and repeat the above steps to continue obtaining matching credibility between the remaining bidding documents and the bidding information. The execution module is used to push the matching bidding documents to the target company according to all matching credibility when matching the bids of the target company.

8. A computer device, characterized in that: The computer device includes a memory and a processor, the memory stores code, and the processor is configured to obtain the code and execute the artificial intelligence-based intelligent management method for bidding information as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the artificial intelligence-based intelligent management method for bidding information as described in any one of claims 1 to 6 is implemented.

Images