Burgling behavior identification method and electronic equipment

Abstract

The invention relates to the technical field of data analysis, and discloses a bidding behavior recognition method and electronic equipment, and the method comprises the steps: obtaining a plurality of bidding files of a target project; for each bidding file, extracting a first feature vector of the bidding file in multiple dimensions, and fusing the first feature vectors in multiple dimensions to obtain a second feature vector corresponding to the bidding file; wherein the multiple dimensions comprise at least two of text features, image features and file attribute features; and based on the second feature vector, identifying the bidding files with the bidding behavior in the plurality of bidding files to obtain a bidding behavior identification result. According to the invention, the problem of low accuracy of bid surrounding and bid stringing behavior recognition can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to a surrounding bid behavior identification method and an electronic device. BACKGROUND

[0002] In the bidding and tendering activities, surrounding bid is an unfair competition behavior, which violates the principle of fair competition and damages the interests of the tenderer. At present, the surrounding bid behavior identification mainly depends on a single text or image in the tender document, and the accuracy of the surrounding bid behavior identification is low. SUMMARY

[0003] The present application provides a surrounding bid behavior identification method and an electronic device to solve the problem of low accuracy of surrounding bid behavior identification.

[0004] In a first aspect, the present application provides a waybill data analysis method, which comprises: obtaining a plurality of tender documents of a target project; for each of the tender documents, extracting a first feature vector of the tender document in multiple dimensions, and fusing the first feature vectors in the multiple dimensions to obtain a second feature vector corresponding to the tender document; wherein the multiple dimensions include at least two of text features, image features and file attribute features; based on the second feature vector, identifying the tender documents with surrounding bid behavior in the plurality of tender documents to obtain a surrounding bid behavior identification result.

[0005] In a second aspect, the present application provides a surrounding bid behavior identification device, which comprises: a file acquisition module for obtaining a plurality of tender documents of a target project; a feature extraction module for, for each of the tender documents, extracting a first feature vector of the tender document in multiple dimensions, and fusing the first feature vectors in the multiple dimensions to obtain a second feature vector corresponding to the tender document; wherein the multiple dimensions include at least two of text features, image features and file attribute features; a data analysis module for, based on the second feature vector, identifying the tender documents with surrounding bid behavior in the plurality of tender documents to obtain a surrounding bid behavior identification result.

[0006] In a third aspect, the present application provides an electronic device, which comprises a memory and a processor, the memory and the processor are communicatively connected with each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the surrounding bid behavior identification method of the first aspect or any of the corresponding embodiments thereof.

[0007] The method for identifying surrounding bid stringing behavior provided by the embodiment of the application extracts the first feature vector of the bid document in at least two dimensions, fuses the first feature vectors of multiple dimensions, and obtains the second feature vector corresponding to the bid document, so as to provide more abundant information for the identification of surrounding bid stringing behavior of the bid document and provide accurate data basis for the identification of surrounding bid stringing behavior. Then, the second feature vector can be used to accurately identify the bid document with surrounding bid stringing behavior in multiple bid documents, obtain the identification result of surrounding bid stringing behavior, and effectively improve the accuracy of the identification of surrounding bid stringing behavior. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the specific embodiments of the application or the technical solutions in the prior art, the drawings needed to be used in the description of the specific embodiments or the prior art will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0009] Figure 1 Fig. 1 is a first flowchart of the method for identifying surrounding bid stringing behavior according to the embodiment of the application; Figure 2 Fig. 2 is a first page display diagram of the method for identifying surrounding bid stringing behavior according to the embodiment of the application; Figure 3 Fig. 3 is a second page display diagram of the method for identifying surrounding bid stringing behavior according to the embodiment of the application; Figure 4 Fig. 4 is a second flowchart of the method for identifying surrounding bid stringing behavior according to the embodiment of the application; Figure 5 Fig. 5 is a third flowchart of the method for identifying surrounding bid stringing behavior according to the embodiment of the application; Figure 6 Fig. 6 is a structural block diagram of the device for identifying surrounding bid stringing behavior according to the embodiment of the application; Figure 7 Fig. 7 is a hardware structure diagram of the electronic device according to the embodiment of the application. DETAILED DESCRIPTION

[0010] In order to make the purpose, technical solutions and advantages of the embodiments of the application more clear, the technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are some embodiments of the application, but not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0011] It can be understood that before using the technical solutions disclosed in the embodiments of the present application, the type of personal information involved in the present application, the use range, the use scene and the like should be informed to the user and the authorization of the user should be obtained according to relevant laws and regulations.

[0012] The terms "first", "second", "third", etc. are used only for descriptive purposes and should not be construed as implying or suggesting relative importance or an implied number of the technical features indicated. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0013] Currently, the following methods are mainly used to check the bid documents for duplication: 1. Text similarity comparison: based on Term Frequency-Inverse Document Frequency (TF-IDF), cosine similarity, Longest Common Subsequence (LCS), the text similarity between two bid documents is compared to identify whether there is a surrounding bid behavior between the two bid documents. However, this method cannot handle unstructured content such as pictures and tables, and ignores the text semantics and industry characteristics, resulting in low accuracy of surrounding bid behavior identification.

[0014] 2. Image similarity detection: using Perceptual Hashing Algorithm (pHash), Scale-Invariant Feature Transform (SIFT), the image similarity between two bid documents is compared to identify whether there is a surrounding bid behavior between the two bid documents. However, this method cannot identify the text embedded in the picture, and is sensitive to format adjustment (such as scaling, color change), resulting in low accuracy of surrounding bid behavior identification.

[0015] 3. Using a general duplicate detection tool to compare the duplication between two bid documents to identify whether there is a surrounding bid behavior between the two bid documents. However, this method only supports basic text comparison, and the comparison dimension is single, resulting in low accuracy of surrounding bid behavior identification.

[0016] As described above, the related art currently has the following defects: 1. Single dimension, missing depth information. Mainly reflected in independent analysis of text or image, lack of text-image correlation, without fusion of multi-modal features, and the text (such as seal and handwritten remarks) in the picture and the file attributes (author, revision record) are not included in the analysis, the information is not comprehensive, resulting in missing of surrounding behavior identification.

[0017] 2. High false positive rate. This is mainly reflected in the fact that pure text similarity analysis cannot distinguish between "reasonable similarity" and "malicious duplication". For example, all bidders must respond to the technical standards and specification clauses of the same bidding document, which will inevitably be highly similar, but this is within the reasonable range. Existing methods cannot effectively filter these public contents, resulting in a large number of false positives.

[0018] 3. Lack of industry characteristics. Traditional duplicate detection technology can only perform general duplicate detection, resulting in inaccurate detection.

[0019] For the convenience of the following description, the terms involved in the following are explained as follows: 1. Surrounding behavior: refers to the behavior risk in which multiple bidders collude with each other, exclude other competitors or harm the interests of the tenderer by agreeing on prices and cooperating in advance, including surrounding and stringing.

[0020] 2. Feature fusion: refers to the use of different types (modalities) of data sources (such as text, images, structured attributes, etc.) for joint analysis and decision-making to improve recognition accuracy.

[0021] 3. OCR: Optical Character Recognition, a technology that converts text in images into editable and searchable text data.

[0022] 3. Industry information: refers to technical content in the bidding documents of a specific industry (such as housing construction) that has high recognition and key significance, such as construction technology, main material brand model, mechanical equipment configuration, key personnel information, construction organization design node map, etc.

[0023] 4. File attributes: refers to the metadata information carried by the electronic document itself, including but not limited to: creator, last modifier, creation / modification time, software version used, total editing time, etc.

[0024] 5. First filtering method: a data preprocessing method that filters data according to pre-set rules to exclude irrelevant or low-value interference information.

[0025] 6. Second filtering method: specifically refers to a mechanism that selectively extracts and compares file content based on the inherent characteristics of the bidding document directory structure and technical standard chapters, aiming to focus on the core technical part that is prone to plagiarism and collusion.

[0026] According to the embodiment of the present application, a method for identifying surrounding bid behavior is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in a different order.

[0027] In the present embodiment, a method for identifying surrounding bid behavior is provided, which can be used in electronic devices such as servers, computers, etc. Figure 1 The flowchart of the method for identifying surrounding bid behavior according to the embodiment of the present application is shown in FIG. 1, which includes the following steps: Figure 1 Step S101: Obtain a plurality of bid files of a target project.

[0028] Optionally, the plurality of bid files can be obtained by manually uploading to the system by the user, or can be obtained by accessing an electronic bidding platform, which is not limited herein.

[0029] Optionally, the format of the plurality of bid files can be PDF format, or can be WORD format, which is not limited herein.

[0030] It should be noted that the plurality of bid files in the above step S101 can be all bid files under the same target project, or can be part of the bid files under the same target project, which is not limited herein.

[0031] Step S102: For each bid file, extract a first feature vector of the bid file in multiple dimensions, and fuse the first feature vectors in multiple dimensions to obtain a second feature vector corresponding to the bid file; wherein the multiple dimensions include at least two of text features, image features and file attribute features.

[0032] Specifically, the first feature vector of each bid file in multiple dimensions under the same project is extracted, and the second feature vector corresponding to the bid file is obtained by feature fusion of the first feature vectors in multiple dimensions.

[0033] Step S103: Based on the second feature vector, identify the bid file with surrounding bid behavior in the plurality of bid files to obtain a surrounding bid behavior identification result.

[0034] Specifically, the second feature vector is processed, and based on the processing result of the second feature vector, the bid file with surrounding bid behavior in the plurality of bid files is identified to obtain a surrounding bid behavior identification result.

[0035] ​The surrounding bid behavior identification method provided by the embodiment extracts the first feature vectors of the bid documents in at least two dimensions, fuses the first feature vectors in multiple dimensions, obtains the second feature vectors corresponding to the bid documents, and thus provides more abundant information for surrounding bid behavior identification of the bid documents and provides accurate data basis for surrounding bid behavior identification. Then, the second feature vectors can be used to accurately identify the bid documents with surrounding bid behavior in the multiple bid documents, obtain the surrounding bid behavior identification result, and effectively improve the accuracy of surrounding bid behavior identification.

[0036] In some optional embodiments, the step S103 of identifying the bid documents with surrounding bid behavior in the multiple bid documents based on the second feature vectors and obtaining the surrounding bid behavior identification result includes the following steps. Step a1, identifying the bid documents with surrounding bid behavior in the multiple bid documents based on the similarity of the second feature vectors between the bid documents and obtaining the first identification result.

[0037] Specifically, the similarity of the second feature vectors between any two bid documents is obtained, and the bid documents with surrounding bid behavior are identified according to the comparison relationship between the similarity of the second feature vectors between any two bid documents and a preset similarity threshold, and the first identification result is obtained.

[0038] Step a2, performing cluster analysis on the second feature vectors of the multiple bid documents, and identifying the bid documents with surrounding bid behavior in the multiple bid documents according to the cluster result and obtaining the second identification result.

[0039] Specifically, if the cluster result represents that most bid documents are dispersed, and the second feature vectors of several bid documents are abnormally clustered in the feature space to form a tight cluster, it is determined that the bid documents corresponding to the second feature vectors in the cluster have surrounding bid behavior.

[0040] Step a3, performing comprehensive analysis based on the first identification result and the second identification result, and obtaining the surrounding bid behavior identification result.

[0041] Specifically, if the first identification result and the second identification result are consistent, the first identification result or the second identification result is determined as the surrounding bid behavior identification result. That is, if the first identification result represents that two bid documents have surrounding bid behavior, and the second identification result also represents that the two bid documents have surrounding bid behavior, it is determined that the two bid documents have surrounding bid behavior. If the first identification result represents that two bid documents do not have surrounding bid behavior, and the second identification result also represents that the two bid documents do not have surrounding bid behavior, it is determined that the two bid documents do not have surrounding bid behavior.

[0042] Further, if the first recognition result is inconsistent with the second recognition result, the first recognition result and the second recognition result are determined as the recognition result of the surrounding bid string behavior according to the priority between the first recognition result and the second recognition result.

[0043] Alternatively, if the first recognition result includes the first similarity between the plurality of bid files and the second recognition result includes the second similarity between the plurality of bid files, in the case that the first recognition result is inconsistent with the second recognition result, the weight corresponding to the first recognition result and the second recognition result is obtained; for any two bid files, the first similarity corresponding to the two bid files in the first recognition result and the second similarity corresponding to the two bid files in the second recognition result are weighted and fused by using the obtained weight, and whether the two bid files have the surrounding bid string behavior is judged according to the weighted and fused result, so as to obtain the surrounding bid string recognition result.

[0044] For example, for the weighted and fused result of the similarity of any two bid files, if the weighted and fused result is greater than a preset fusion threshold, it is determined that the two bid files have the surrounding bid string behavior.

[0045] The surrounding bid string behavior recognition method provided by the embodiment improves the credibility of the recognition result in the data aspect by comparing the similarity of the second feature vectors between any two bid files with the preset similarity threshold to obtain the first recognition result, further obtains the second recognition result by clustering analysis on the second feature vectors of the plurality of bid files, and can recognize the surrounding bid string behavior of the plurality of bid files, and finally obtains the surrounding bid string behavior recognition result by comprehensively analyzing the first recognition result and the second recognition result, and improves the accuracy of the surrounding bid string behavior recognition.

[0046] In some optional embodiments, the step a1 of identifying the bid file having the surrounding bid string behavior from the plurality of bid files based on the similarity of the second feature vectors between the bid files to obtain the first recognition result includes: Step a11, determining the similarity of the second feature vectors between any two bid files.

[0047] Optionally, the similarity of the second feature vectors is cosine similarity.

[0048] In addition to the cosine similarity, the similarity of the second feature vectors between any two bid files can also be obtained by using similarity calculation methods such as Term Frequency-Inverse Document Frequency (TF-IDF) and Longest Common Subsequence (LCS).

[0049] In practical applications, a matrix can be used to record the similarity of the second feature vector between any two bid documents, resulting in a similarity matrix. This allows for a better determination of whether there is collusion or bid-rigging between the two bid documents.

[0050] Step a12: If the similarity of the second feature vector between any two bid documents is greater than the preset similarity threshold, then it is determined that there is collusion between the two bid documents, so as to obtain the first identification result.

[0051] The first document is used to characterize high-risk pairs with collusion behavior.

[0052] Specifically, if the similarity of the second feature vector between any two bid documents is greater than a preset similarity threshold, it is determined that there is collusion between the two bid documents, and the two bid documents are marked as the first document pair.

[0053] The bid-rigging behavior identification method provided in this embodiment determines whether there is bid-rigging behavior between two bid documents by analyzing the comparison results of the similarity of the second feature vector between any two bid documents and a preset similarity threshold, thereby achieving accurate identification of high-risk bid document pairs.

[0054] Furthermore, the method for identifying bid-rigging behavior of the present invention also includes: generating a bid-rigging behavior analysis report based on the first identification result, the similarity of the second feature vectors of the bid documents with bid-rigging behavior in the first identification result, text similarity, image similarity, file attribute similarity, and industry information similarity.

[0055] As an application example, see the page display of the collusion behavior analysis report. Figure 2 . Figure 2 The risk level is used to characterize the collusion behavior of bidding entities, and the risk level ranges from [0, 100]. The collusion risk analysis results record the risk level identification results between any two bidding entities, and the analysis results details record the identification details between any two bid documents with collusion risk, including four parts: text similarity, image similarity, document attribute similarity, and key content similarity.

[0056] In addition, the analysis details of the two bid documents can be visualized, see [link / reference]. Figure 3 Taking tender documents B and C as examples, the analysis details page for both tender documents B and C includes four parts: text, images, document attributes, and key content. Clicking on the key content section will display the key content of tender documents B and C in the target display area of ​​the analysis details page, and will also highlight similar content between the two tender documents within the key content section.

[0057] In some optional embodiments, the second feature vectors of the plurality of bid files are subjected to cluster analysis in step a2, and the bid files with the surrounding bid behavior are identified from the plurality of bid files according to the cluster results to obtain a second identification result, which comprises: Step a21, the second feature vectors of the plurality of bid files are subjected to cluster analysis to obtain a plurality of cluster clusters.

[0058] Optionally, the cluster analysis method can be a density-based spatial clustering algorithm (Density-Based Spatial Clustering of Applications with Noise, DBSCAN), or a spectral clustering or hierarchical clustering, which is not specifically limited here.

[0059] The cluster cluster is a small cluster formed by the second feature vectors corresponding to the plurality of bid files.

[0060] Step a22, based on the aggregation degree of the second feature vectors in each cluster cluster, a target cluster cluster is determined from the plurality of cluster clusters.

[0061] The aggregation degree is used to represent the aggregation degree of the second feature vectors in the cluster cluster, and is obtained by the above cluster analysis method.

[0062] Optionally, the method for determining the target cluster cluster is as follows: 1. According to a preset aggregation threshold, the target cluster cluster is determined: if the aggregation degree of the second feature vectors in the cluster cluster is greater than the preset aggregation threshold, it indicates that the cluster cluster is abnormally aggregated, and the cluster cluster corresponding to the aggregation degree is taken as the target cluster cluster.

[0063] 2. The target cluster cluster is determined according to the sorting of the aggregation degree: the aggregation degrees of the second feature vectors of all cluster clusters are sorted, and the top few (such as the first) cluster clusters are taken as the target cluster cluster.

[0064] 3. The target cluster cluster is determined according to the proportion of the cluster cluster: according to a preset proportion, the plurality of cluster clusters are taken as the target cluster cluster, for example, the first ten percent of all cluster clusters are taken as the target cluster cluster.

[0065] Step a23, based on the second feature vectors corresponding to the bid files in the target cluster cluster, a second identification result is obtained.

[0066] Specifically, the bid files corresponding to the second feature vectors in the target cluster cluster are determined to have the surrounding bid behavior, and a second identification result is obtained.

[0067] The surrounding string behavior identification method provided in the embodiment performs clustering analysis on the second feature vectors to obtain a plurality of clustering clusters, and filters out a target clustering cluster from the plurality of clustering clusters according to the clustering degree of the second feature vectors in each clustering cluster to obtain a second identification result. The surrounding string risk of a plurality of bid documents is identified through structured clustering analysis.

[0068] In some optional embodiments, before the step of extracting the first feature vectors of the bid documents in multiple dimensions in the above step S102, the method comprises: performing content filtering on the bid documents based on a target filtering mode; wherein the target filtering mode comprises at least one of a first filtering mode and a second filtering mode; the first filtering mode is used to perform content filtering on the bid documents based on the target file content of the bidding document of the target project; and the second filtering mode is used to perform content filtering on the bid documents based on the directory information of the bid documents.

[0069] In some optional embodiments, the above content filtering on the bid documents based on the target filtering mode comprises: Step b1, if the target filtering mode comprises the first filtering mode, filtering the file content in the bid documents that matches the target file content in the bidding document.

[0070] Optionally, the target file content can include technical specifications and general terms in the bid documents, and can also include other public content, which is not limited herein.

[0071] In actual application, the target file content can be determined according to the historical bidding documents and historical bid documents of the target project, or determined according to the file content of the bidding document to be contained in the bid document, which is not limited herein.

[0072] Specifically, the target file content in the bidding document is taken as a basic template, and through text comparison, the highly matched part in the file content of all bid documents is identified and filtered out, and the highly matched part in the bid document is excluded.

[0073] Step b2, if the target filtering mode comprises the second filtering mode, determining the target hierarchical unit to be filtered in the hierarchical unit under the directory information of the bid document based on a preset unit feature, and filtering the file content corresponding to the target hierarchical unit in the bid document; wherein the preset unit feature comprises at least one of the feature of the hierarchical unit to be retained and the feature of the hierarchical unit to be filtered.

[0074] The hierarchical unit is a chapter in the directory of the bid document. The feature of the hierarchical unit to be retained represents a key chapter, and the feature of the hierarchical unit to be filtered represents a non-key chapter.

[0075] Optionally, the features of the level units to be retained include features of sections to be retained in the bid document. For example, at least one of "construction technical scheme", "project management organization", "labor plan", and "main material equipment table". The features of the level units to be filtered include features of sections to be filtered in the bid document. For example, at least one of "corporate profile", "list of qualification certificates", and "letter of commitment".

[0076] Illustratively, the directory information of the bid document is extracted and parsed, the key sections of the bid document are automatically located according to the preset unit features, and only the key sections in the bid document are retained. Or the non-key sections of the bid document are automatically located according to the preset unit features, and the non-key sections in the bid document are filtered.

[0077] The surrounding string identification method provided by the embodiment filters the file content in the bid document that matches the target file content in the bid document through a first filtering manner, and filters the key sections or non-key sections of the bid document through a second filtering manner. The content with high standardization degree and strong similarity is filtered out, the high-bid-file surrounding area is screened, and the computing resources are concentrated to analyze the key sections.

[0078] In some optional embodiments, the first feature vector of the bid document in multiple dimensions in the above step S102 includes: In step c1, if the multiple dimensions include text features, a first text of the bid document is obtained, the first text is subjected to word segmentation processing, and a second text obtained by the word segmentation processing is subjected to feature extraction to obtain the first feature vector of the bid document corresponding to the text features.

[0079] The first text of the bid document is the text after content filtering of the bid document.

[0080] In actual application, for the file layer of the bid document, all optional texts and their position information in the bid document are extracted to obtain the first file and its position information.

[0081] Specifically, the text after content filtering of the bid document (i.e. the first text) is subjected to preprocessing such as word segmentation and stop word removal, and a second text obtained by the word segmentation processing is subjected to feature extraction to convert the first text into a high-dimensional vector representation to obtain the first feature vector of the bid document corresponding to the text features.

[0082] Optionally, the first feature vector can be obtained by using a bag-of-words model (Term Frequency-Inverse Document Frequency, TF-IDF) or a deep learning model, which is not limited here.

[0083] Optionally, the deep learning model can be a Sentence Bidirectional Encoder Representations from Transformers (Sentence-BERT) model or a Transformer model, which is not limited herein.

[0084] In step c2, if the multiple dimensions include image features, the target image in the bid document is obtained, third text in the target image is extracted, and feature extraction is performed on the third text to obtain a first feature vector corresponding to the image features of the bid document.

[0085] In actual applications, for the image layer of the bid document, each embedded image and page snapshot in the bid document is intercepted to obtain a target image in the bid document.

[0086] Specifically, the text content in the target image is recognized using an Optical Character Recognition (OCR) engine, the text content in the target image is extracted as third text, and the third text is converted into an OCR feature vector to obtain a first feature vector corresponding to the image features of the bid document.

[0087] Optionally, the OCR engine can be a Tesseract OCR engine or a Paddle OCR engine, which is not limited herein.

[0088] It should be noted that after the third text of the target image is extracted, the third text needs to be fused with the first text.

[0089] In step c3, if the multiple dimensions include file attribute features, a target file attribute of the bid document is obtained, and feature extraction is performed on the target file attribute to obtain a first feature vector corresponding to the file attribute features of the bid document.

[0090] The target file attribute is metadata obtained after the bid document is decomposed.

[0091] Optionally, the target file attribute can include at least one of an author attribute, a company attribute, a last saver attribute, a bid document modification attribute, and a bid document editing attribute, or can be other types of attribute information, which is not limited herein.

[0092] Specifically, target file attributes of the bid file are extracted, the target file attribute values are numerized, and a first feature vector corresponding to the file attribute features of the bid file is obtained. For example, the target file attributes include author attributes, company attributes, and last saver attributes. If the author attributes, company attributes, and last saver attributes of two bid files are completely the same, the first feature vector of the file attribute features is set to 100%. If only one kind of attribute in the target file attributes is the same, the first feature vector of the file attribute features is set to 33.3%.

[0093] The surrounding string identification method provided in the embodiment realizes the fusion of text, image, and attribute, describes the characteristics of the bid file from multiple dimensions, provides data support for the subsequent second feature vector, and improves the accuracy of surrounding string identification.

[0094] In some optional embodiments, the multiple dimensions in the step S102 include industry features, and the first feature vectors of the bid file in the multiple dimensions are extracted, including: Step c4, obtaining a target industry to which the target project belongs; Optionally, the target industry to which the target project belongs can be configured by a user, or obtained by performing semantic analysis on a project title of the target project, which is not limited herein.

[0095] Step c5, extracting industry information in the bid file based on an industry knowledge base corresponding to the target project, to obtain target industry information; wherein the industry knowledge base is used to define information related to the target industry; The industry knowledge base corresponding to the target project is obtained by a background according to expert experience and business accumulation.

[0096] Optionally, the information related to the target industry can be defined by a named entity related to the target industry, or defined by an information extraction template of the target industry, which is not limited herein.

[0097] Specifically, the industry information in the second text and the third text is extracted according to the industry knowledge base corresponding to the target project, to obtain the target industry information.

[0098] In actual application, taking the housing construction industry as an example, the constructed industry knowledge base includes: Project information: at least one of “company / organization”, “place name / address”, “project”, and “award”.

[0099] Key personnel information: at least one of the name, identity, phone number, unified social credit code, and certificate number of “project manager” and “technical director”.

[0100] Main material equipment table: at least one of "reinforcing bar brand", "cement grade", "elevator model".

[0101] Construction machinery table: at least one of "tower crane model", "number of excavators".

[0102] Step c6, feature extraction is performed on the target industry information to obtain a first feature vector corresponding to the industry characteristics of the bid document.

[0103] Specifically, the target industry information in the second text and the third text is extracted using a large model or a preset keyword template, and the target industry information is converted into structured data, and then through encoding (such as hash encoding or vectorization), a first feature vector corresponding to the industry characteristics of the bid document is obtained.

[0104] The surrounding bid behavior recognition method provided in the embodiment extracts industry information in a bid document according to an industry knowledge base corresponding to a target project to obtain target industry information, and obtains a first feature vector corresponding to the industry characteristics of the bid document through feature extraction. By establishing an industry knowledge base and comprehensively considering the dimensions of industry characteristics, the surrounding bid behavior recognition has strong industry pertinence, and the accuracy of the surrounding bid behavior recognition is improved.

[0105] In some optional embodiments, the fusion of the first feature vectors of the multiple dimensions in the step S102 to obtain a second feature vector corresponding to the bid document includes: Step d1, obtaining a weight corresponding to each dimension.

[0106] Specifically, the weight of the first feature vector of each dimension is set under the experience of experts. For example, the weight of the first feature vector corresponding to the industry characteristics of the bid document and the first feature vector corresponding to the image characteristics of the bid document is high.

[0107] Step d2, weighting and summing the first feature vectors of each dimension based on the weight to obtain a second feature vector corresponding to the bid document.

[0108] Specifically, the first feature vectors of each dimension are weighted and summed according to the weight to obtain a second feature vector corresponding to the bid document, as shown in the following formula:

[0109] wherein, is the second feature vector corresponding to the bid document, is the first feature vector corresponding to the text characteristics of the bid document, and a is the corresponding weight; is the first feature vector corresponding to the image characteristics of the bid document, and β is corresponding weight; is a first feature vector corresponding to the file attribute feature of the bid file, and γ is corresponding weight; is a first feature vector corresponding to the industry feature of the bid file, and ŋ is corresponding weight.

[0110] The surrounding bid behavior recognition method provided by the embodiment can dynamically adjust the corresponding weight according to the importance of the first feature vector of each dimension, so as to effectively recognize the key features of multiple dimensions. Furthermore, the first feature vectors of each dimension are weighted and summed to obtain a second feature vector corresponding to the bid file, thereby realizing effective association of multiple dimensions and improving the accuracy of surrounding bid behavior recognition.

[0111] As a specific application example, refer to Figure 4 . Obtain multiple bid files of the same target project, for each bid file, extract the first feature vectors of the bid file in multiple dimensions to obtain a first feature vector corresponding to the text feature of the bid file, a first feature vector corresponding to the image feature of the bid file, a first feature vector corresponding to the attribute feature of the bid file, and a first feature vector corresponding to the industry feature of the bid file. Furthermore, the first feature vectors of multiple dimensions are weighted and fused to obtain a second feature vector. The first recognition result is obtained by similarity calculation on the second feature vector, and the second recognition result is obtained by clustering analysis on the second feature vector. Finally, the surrounding bid behavior recognition result is obtained by comprehensively analyzing the first recognition result and the second recognition result.

[0112] As another specific application example, refer to Figure 5The image, text and file attribute of the bid document are acquired, similarity analysis is respectively performed on the image, text and file attribute of the bid document, the total similarity of the bid document is calculated based on weighted fusion, and a surrounding bid behavior recognition result is obtained. For the image of the bid document, the text in the certificate image is recognized by certificate recognition, the text in the non-certificate image is recognized by OCR, and the third text is obtained by integration. Then, the third text is filtered, a part of the third text continues to perform key information extraction to obtain key information in the image, and the image key information similarity is calculated. Another part of the third text is fused and de-duplicated with the filtered key text, key information is extracted, key information of the key text is obtained, and the key text key information similarity is calculated. For the text of the bid document, the normal text and the key text are obtained after filtering, and the normal text similarity is directly calculated. For the file attribute of the bid document, the information of the bid document author and the last time saver is acquired, and the similarity of the bid document author and the last time saver is calculated. Finally, the image key information similarity, the key text key information similarity, the normal text similarity and the author and the last time saver similarity are weighted and fused to obtain the total similarity calculated by weighting.

[0113] In the embodiment, a surrounding bid behavior recognition device is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and details are not repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware or a combination of software and hardware is also possible and is contemplated.

[0114] The embodiment provides a surrounding bid behavior recognition device, as shown in Figure 6 , comprising: A file acquisition module 601 is configured to acquire a plurality of bid documents of a target project.

[0115] A feature extraction module 602 is configured to extract, for each bid document, a first feature vector of the bid document in multiple dimensions, and fuse the first feature vectors in the multiple dimensions to obtain a second feature vector corresponding to the bid document, wherein the multiple dimensions include at least two of a text feature, an image feature and a file attribute feature.

[0116] A data analysis module 603 is configured to identify, based on the second feature vector, a bid document with a surrounding bid behavior from the plurality of bid documents, and obtain a surrounding bid behavior recognition result.

[0117] In some optional embodiments, the data analysis module 603 described above comprises: The first identification unit is configured to identify the bid documents with the surrounding bid behavior from the plurality of bid documents based on the similarity of the second feature vectors between the bid documents, and obtain a first identification result.

[0118] The second identification unit is configured to perform clustering analysis on the second feature vectors of the plurality of bid documents, and identify the bid documents with the surrounding bid behavior from the plurality of bid documents according to the clustering result, and obtain a second identification result.

[0119] The third identification unit is configured to perform comprehensive analysis based on the first identification result and the second identification result, and obtain a surrounding bid behavior identification result.

[0120] In some optional embodiments, the second identification unit includes: The first analysis sub-unit is configured to perform clustering analysis on the second feature vectors of the plurality of bid documents, and obtain a plurality of clustering clusters.

[0121] The second analysis sub-unit is configured to determine a target clustering cluster from the plurality of clustering clusters based on the aggregation degree of the second feature vectors in each clustering cluster.

[0122] The third analysis sub-unit is configured to obtain the second identification result based on the bid documents corresponding to the second feature vectors in the target clustering cluster.

[0123] In some optional embodiments, the feature extraction module 602 further includes, before extracting the first feature vectors of the bid documents in the plurality of dimensions, a filtering unit configured to perform content filtering on the bid documents based on a target filtering manner; the target filtering manner includes at least one of a first filtering manner and a second filtering manner; the first filtering manner is configured to perform content filtering on the bid documents based on the target file content in the bid documents; and the second filtering manner is configured to perform content filtering on the bid documents based on directory information of the bid documents.

[0124] In some optional embodiments, the filtering unit includes: The first filtering sub-unit is configured to, if the target filtering manner includes the first filtering manner, filter the file content in the bid documents that matches the target file content in the bid documents.

[0125] The second filtering sub-unit is configured to, if the target filtering manner includes the second filtering manner, determine a target hierarchical unit to be filtered in the hierarchical units of the bid documents under the preset unit feature, and filter the file content corresponding to the target hierarchical unit in the bid documents; the preset unit feature includes at least one of a feature of a hierarchical unit to be retained and a feature of a hierarchical unit to be filtered.

[0126] ​In some optional embodiments, the feature extraction module 602 comprises: The first processing unit is configured to, if the multiple dimensions comprise text features, acquire a first text of the bid document, perform word segmentation processing on the first text, and perform feature extraction on a second text obtained by the word segmentation processing to obtain a first feature vector corresponding to the text features of the bid document.

[0127] The second processing unit is configured to, if the multiple dimensions comprise image features, acquire a target image in the bid document, extract a third text in the target image, and perform feature extraction on the third text to obtain a first feature vector corresponding to the image features of the bid document.

[0128] The third processing unit is configured to, if the multiple dimensions comprise file attribute features, acquire a target file attribute of the bid document, and perform feature extraction on the target file attribute to obtain a first feature vector corresponding to the file attribute features of the bid document.

[0129] In some optional embodiments, the feature extraction module 602 further comprises: The fourth processing unit is configured to acquire a target industry to which the target project belongs.

[0130] The fifth processing unit is configured to extract target industry information from the bid document based on an industry knowledge base corresponding to the target project, wherein the industry knowledge base is used to define information related to the target industry.

[0131] The sixth processing unit is configured to perform feature extraction on the target industry information to obtain a first feature vector corresponding to industry features of the bid document.

[0132] In some optional embodiments, the feature extraction module 602 further comprises: The weight acquisition unit is configured to acquire weights corresponding to the multiple dimensions.

[0133] The weighting unit is configured to perform weighted summation on the first feature vectors of the multiple dimensions based on the weights to obtain a second feature vector corresponding to the bid document.

[0134] The waybill data analysis apparatus provided by the embodiments of the present application can execute the waybill data analysis method provided by any of the embodiments of the present application, and has the corresponding function modules and beneficial effects of the execution method. The further function description of each module and unit is the same as that of the corresponding embodiments, and will not be repeated here.

[0135] Figure 7 A structural schematic diagram of an electronic device provided by the embodiments of the present application.

[0136] The following will be specifically described with reference to Figure 7which shows a structural diagram suitable for use to implement an electronic device in embodiments of the present application. The electronic device can include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 701, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage 708 into a random access memory (RAM) 703. Various programs and data required for operation of the electronic device are also stored in the RAM 703. The processor 701, the ROM 702, and the RAM 703 are connected to each other through a bus 704. An input / output (I / O) interface 707 is also connected to the bus 704.

[0137] Generally, the following devices can be connected to the I / O interface 707: input devices 707 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 707 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage 708 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 709. The communication device 709 can allow the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 7 The electronic device is shown with various devices, but it should be understood that all of the shown devices are not required, and more or fewer devices can alternatively be implemented.

[0138] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to embodiments of the present application. For example, embodiments of the present application include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication device 709, or installed from the storage 708, or installed from the ROM 702. When the computer program is executed by the processor 701, the above-described functions defined in the loop behavior identification method of embodiments of the present application are performed.

[0139] Figure 7 The electronic device shown is merely an example and should not impose any limitation on the functions and scope of use of embodiments of the present application.

[0140] The embodiments of the present application further provide a computer readable storage medium, and the method according to the embodiments of the present application can be implemented in hardware, firmware, or recorded in a storage medium, or stored in a remote storage medium or a non-transitory machine readable storage medium and downloaded to a local storage medium through network, so that the method described herein can be processed by such software on a storage medium using a general purpose computer, a special purpose processor, or programmable or special hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid state disk, etc. Further, the storage medium can also include a combination of the above-mentioned memories. It can be understood that the computer, the processor, the microprocessor controller, or the programmable hardware includes a storage component that can store or receive software or computer code, when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the above-mentioned embodiments of the method for identifying the surrounding string behavior are implemented.

[0141] Part of the present application can be applied as a computer program product, for example, computer program instructions, when executed by a computer, the operation of the computer can call or provide the method and / or technical solutions according to the present application. Those skilled in the art should understand that the form of computer program instructions in computer readable medium includes but is not limited to source file, executable file, installation package file, etc. Correspondingly, the way of computer program instructions executed by computer includes but is not limited to: the computer directly executes the instructions, or the computer compiles the instructions and then executes the corresponding compiled program, or the computer reads and executes the instructions, or the computer reads and installs the instructions and then executes the corresponding installed program. Here, the computer readable medium can be any available computer readable storage medium or communication medium accessible to the computer.

[0142] Although the embodiments of the present application are described in conjunction with the accompanying drawings, various modifications and changes can be made by those skilled in the art without departing from the spirit and scope of the present application, and such modifications and changes fall within the scope defined by the appended claims.

Claims

1. A method for identifying a surrounding string behavior, characterized in that, The method comprises: obtaining a plurality of bid files of a target project; for each of the bid files, extracting a first feature vector of the bid file in multiple dimensions, and fusing the first feature vectors in the multiple dimensions to obtain a second feature vector corresponding to the bid file; wherein the multiple dimensions include at least two of text features, image features, and file attribute features; based on the second feature vector, identifying bid files with surrounding bidding behaviors from the plurality of bid files to obtain a surrounding bidding behavior identification result.

2. The method of claim 1, wherein, The method comprises: based on the similarity of the second feature vectors between the bid files, identifying bid files with surrounding bidding behaviors from the plurality of bid files to obtain a first identification result; performing cluster analysis on the second feature vectors of the plurality of bid files, and identifying bid files with surrounding bidding behaviors from the plurality of bid files according to the cluster result to obtain a second identification result; based on the first identification result and the second identification result, performing comprehensive analysis to obtain the surrounding bidding behavior identification result.

3. The method of claim 2, wherein, The method comprises: performing cluster analysis on the second feature vectors of the plurality of bid files to obtain a plurality of cluster clusters; based on the aggregation degree of the second feature vectors in each cluster cluster, determining a target cluster cluster from the plurality of cluster clusters; based on the bid file corresponding to the second feature vector in the target cluster cluster, obtaining the second identification result.

4. The method of claim 1, wherein, Before the step of extracting the first feature vector of the bid file in multiple dimensions, the method comprises: performing content filtering on the bid file based on a target filtering mode; wherein the target filtering mode includes at least one of a first filtering mode and a second filtering mode; the first filtering mode is used to perform content filtering on the bid file based on the bidding file of the target project; and the second filtering mode is used to perform content filtering on the bid file based on the directory information of the bid file.

5. The method of claim 4, wherein, The method comprises: if the target filtering mode includes the first filtering mode, filtering the file content in the bid file that matches the target file content in the bidding file; if the target filtering mode includes the second filtering mode, determining a target hierarchical unit to be filtered in the hierarchical unit under the directory information of the bid file based on a preset unit feature, and filtering the file content corresponding to the target hierarchical unit in the bid file; wherein the preset unit feature includes at least one of the features of the hierarchical unit to be retained and the features of the hierarchical unit to be filtered.

6. The method of claim 1, wherein, The method comprises: If the multiple dimensions include text features, a first text of the bid document is obtained, word segmentation processing is performed on the first text, and feature extraction is performed on second text obtained through the word segmentation processing to obtain a first feature vector corresponding to text features of the bid document; If the multiple dimensions include image features, a target image in the bid document is obtained, third text in the target image is extracted, and feature extraction is performed on the third text to obtain a first feature vector corresponding to image features of the bid document; If the multiple dimensions include file attribute features, a target file attribute of the bid document is obtained, and feature extraction is performed on the target file attribute to obtain a first feature vector corresponding to file attribute features of the bid document.

7. The method of claim 1, wherein, The multiple dimensions include industry features; the extracting the first feature vector of the bid document in multiple dimensions includes: obtaining a target industry to which the target project belongs; extracting target industry information from the bid document based on an industry knowledge base corresponding to the target project, wherein the industry knowledge base is used to define information related to the target industry; performing feature extraction on the target industry information to obtain a first feature vector corresponding to industry features of the bid document.

8. The method of claim 1, wherein, The fusing the first feature vectors of the multiple dimensions to obtain a second feature vector corresponding to the bid document includes: obtaining a weight corresponding to each of the dimensions; performing weighted summation on the first feature vectors of each of the dimensions based on the weight to obtain a second feature vector corresponding to the bid document.

9. A device for identifying a surrounding string action, characterized by The device includes: a file obtaining module configured to obtain multiple bid documents of a target project; a feature extraction module configured to, for each of the bid documents, extract a first feature vector of the bid document in multiple dimensions, and fuse the first feature vectors of the multiple dimensions to obtain a second feature vector corresponding to the bid document; wherein the multiple dimensions include at least two of text features, image features, and file attribute features; a data analysis module configured to, based on the second feature vector, identify a bid document with a surrounding bid behavior from the multiple bid documents to obtain a surrounding bid behavior identification result.

10. An electronic device, comprising: It includes: a memory and a processor, which are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the surrounding bid behavior identification method of any one of claims 1 to 8.

Images