Intelligent bidding document review system integrating quantitative scoring and image recognition

Abstract

The invention relates to the technical field of data analysis, in particular to a bidding document intelligent review system fusing quantitative scoring and image recognition, comprising: a data acquisition unit used for acquiring hierarchical information corresponding to target qualification information corresponding to each bidding document; the search analysis unit is used for determining a region division mode according to a judgment condition and determining a region type according to association level similarity and a link difference coefficient; the information acquisition unit is used for determining a hierarchical state according to the link reference value and the link radiation coefficient and determining a qualification search mode according to the hierarchical state so as to acquire public qualification information; the optimization unit is used for determining an optimization mode according to the analysis condition; the evaluation unit is used for determining the file type of the bidding file to be evaluated according to the associated threshold value and the similar reference value and determining a file scoring mode according to the file type; according to the invention, the accuracy degree of bidding document evaluation can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and particularly to an intelligent bid document review system that integrates quantitative scoring and image recognition. Background Art

[0003] Chinese Patent Publication No. CN114492368A discloses an AI bid automatic scoring method, including: establishing a feature extraction model for natural language processing; uploading a tender document to obtain the tender details information of the tender document, and using the feature extraction model of natural language processing to extract the specific tender requirements in the tender details information; performing serial numbering on the specific tender requirements extracted by the feature extraction model of natural language processing; establishing a sub-database of the basic situations of multiple companies, and marking each sub-data in the database according to the alphabetical order of the characters; establishing a scoring model; setting screening conditions to screen out relevant companies that meet the conditions, and matching the sub-database of the basic situations of relevant companies with the tender specific requirement data marked with specified labels, and calling the scoring model to score the tender document. It can be seen that the above technical solution has the following problems: only scoring based on the document content, without considering the web page information related to the bid document content, resulting in a deviation in the accuracy of document evaluation. Summary of the Invention

[0004] Therefore, the present invention provides an intelligent bid document review system that integrates quantitative scoring and image recognition to overcome the problem in the prior art that the web page information related to the bid document content is not considered, resulting in a deviation in the accuracy of document evaluation.

[0005] To achieve the above object, the present invention provides an intelligent bid document review system that integrates quantitative scoring and image recognition, including: A data acquisition unit for obtaining the hierarchical information of the target qualification information corresponding to each bid document; A search and analysis unit connected to the data acquisition unit for determining the area division method according to the judgment condition, and determining the area type according to the correlation level similarity and the link difference coefficient. The area division method is uniform division according to the comprehensive evaluation value or associated division according to the level threshold; An information acquisition unit connected to the search and analysis unit for determining the level status according to the link reference value and the link radiation coefficient, and determining the qualification search method according to the level status to obtain the public qualification information. The qualification search method is to determine the optimization method according to the proportion of the first-class area and the first-class area distribution coefficient or to determine the search priority coefficient of the link according to the relevance of the link information; An optimization unit, which is connected to the information acquisition unit, is used to determine an optimization method according to analysis conditions. The optimization method is to determine a search area according to link similarity and regional correlation, or to determine a regional search duration according to a regional threshold. An evaluation unit, which is respectively connected to the data acquisition unit, the information acquisition unit and the optimization unit, is used to record the bid documents corresponding to the target qualification information with a qualification difference degree less than a preset qualification difference degree from the public qualification information as bid documents to be evaluated, determine the document type of the bid documents to be evaluated according to a correlation threshold and a similarity reference value, and determine a document evaluation coefficient determination method according to the document type. The document evaluation coefficient determination method is to determine the document evaluation coefficient according to an index fluctuation value and a tender classification value or according to the tender classification value.

[0006] Further, the judgment condition for the search analysis unit to respond is that the hierarchical quantity reference value is less than a preset link difference degree or the link reference value is less than a preset link reference value, and the area division method is to make a uniform division according to a comprehensive evaluation value. The judgment condition for the search analysis unit to respond is that the hierarchical quantity reference value is less than a preset hierarchical quantity reference value and the link reference value is greater than or equal to a preset link reference value, and the area division method is to make a correlation division according to a hierarchical threshold.

[0007] Further, the correlation division by the search analysis unit according to the hierarchical threshold includes: Performing correlation analysis on each link in the target level in a preset order. When performing correlation analysis on a single link, mark the link as the target link, and mark the links outside the target link that have not been recorded in the sub-region as reference links; Cumulatively add the hierarchical threshold corresponding to the target link to the hierarchical thresholds corresponding to each link in the preset order one by one until the total value of the hierarchical thresholds is greater than the preset total value of the hierarchical thresholds. Mark the reference links before the preset order of the hierarchical threshold corresponding to the last cumulative hierarchical threshold and the target link as associated links, and mark the smallest rectangular area that can contain the associated links as a sub-region; And continue to perform correlation analysis on the links that have not been recorded in the sub-region until all links are recorded in the sub-region; The preset order is the order of the associated distances corresponding to each link from small to large.

[0008] Further, the search analysis unit determines the area type according to the associated hierarchical similarity and the link difference coefficient. The area types include: A type of area where the associated hierarchical similarity is less than a preset associated hierarchical similarity or the link difference coefficient is greater than or equal to a preset link difference coefficient; A type of area where the associated hierarchical similarity is greater than or equal to a preset associated hierarchical similarity and the link difference coefficient is less than a preset link difference coefficient.

[0009] Furthermore, the associated level similarity is determined according to the link information degree; If the link information degree is greater than or equal to the preset link information degree, the associated level similarity is determined according to the number of identical keywords, where the associated level similarity has a positive correlation with the number of identical keywords; If the link information degree is less than the preset link information degree, the associated level similarity is determined according to the number of associated keywords, where the associated level similarity has a positive correlation with the number of associated keywords.

[0010] Furthermore, the information acquisition unit determines the level status according to the link reference value and the link radiation coefficient, and the level status includes: The first level status where the link reference value is greater than or equal to the preset link reference value or the link radiation coefficient is greater than or equal to the preset link radiation coefficient; The second level status where the link reference value is less than the preset link reference value and the link radiation coefficient is less than the preset link radiation coefficient.

[0011] Furthermore, the information acquisition unit responds to the level status to determine the qualification search method; When the information acquisition unit responds to the first level status, the qualification search method is to determine the optimization method according to the proportion of the first type of area and the distribution coefficient of the first type of area; When the information acquisition unit responds to the second level status, the qualification search method is to determine the search priority coefficient of the link according to the link information relevance; The search priority coefficient has a positive correlation with the link information relevance.

[0012] Furthermore, the optimization unit responds to the analysis condition to determine the optimization method; The analysis condition that the optimization unit responds to is that the proportion of the first type of area is greater than or equal to the preset proportion of the first type of area or the distribution coefficient of the first type of area is greater than or equal to the preset distribution coefficient of the first type of area, and the optimization method is to determine the search area according to the link similarity and the regional association degree; The analysis condition that the optimization unit responds to is that the proportion of the first type of area is less than the preset proportion of the first type of area and the distribution coefficient of the first type of area is less than the preset distribution coefficient of the first type of area, and the optimization method is to determine the regional search duration of each sub - area according to the regional threshold; The search area is the first type of area where the link similarity is less than the preset link similarity and the regional association degree is less than the preset regional association degree; The regional search duration of a single sub - area has a positive correlation with the regional threshold.

[0013] Furthermore, the evaluation unit determines the document type of the tender documents to be evaluated according to the association threshold and the similarity reference value, and the document type includes: A type of document where the association threshold is greater than or equal to the preset association threshold or the similarity reference value is greater than or equal to the preset similarity reference value; A type of document where the association threshold is less than the preset association threshold and the similarity reference value is less than the preset similarity reference value.

[0014] Furthermore, the evaluation unit determines the method for determining the document evaluation coefficient according to the document type; For type - one documents, the method for determining the document evaluation coefficient is to determine the document evaluation coefficient according to the index fluctuation value and the tender classification value; For type - two documents, the method for determining the document evaluation coefficient is to determine the document evaluation coefficient according to the tender classification value.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows. In the technical solution of the present invention, the area division method is determined according to the judgment conditions. The judgment conditions effectively reflect the link difference degree of the target level corresponding to the target qualification information, the number of links and the link distribution status in the sub - levels corresponding to the target level. Furthermore, different area division methods are adaptively selected according to the judgment conditions, making the selection of the area division method more in line with the actual application scenario, avoiding the problem of poor relevance of the search results of the public qualification information caused by unreasonable area division, being beneficial to improving the accuracy of the target qualification information review, and further improving the accuracy of the tender document evaluation.

[0016] Furthermore, in the present invention, the link information degree effectively reflects the link information situation of the associated level corresponding to the target level. Then, the similarity of the associated level is determined according to the link information degree, making the confirmation method of the similarity of the associated level more in line with the actual application scenario. The similarity of the associated level and the link difference coefficient effectively reflect the similarity degree of the links of the target level and its associated level. Then, the area type is determined according to the similarity of the associated level and the link difference coefficient, making the determination of the area type more in line with the actual application scenario, avoiding the problem of poor relevance of the search results of the public qualification information caused by inaccurate determination of the area type, and further improving the accuracy of the tender document evaluation.

[0017] Furthermore, in the present invention, the level status is determined according to the link reference value and the link radiation coefficient. The link reference value and the link radiation coefficient effectively reflect the number of links and the link distribution status in the target level. Then, different qualification search methods are adaptively selected according to the level status, making the selection of the qualification search method more in line with the actual application scenario, avoiding the problem of poor relevance of the search results of the public qualification information caused by too large a search range, and improving the accuracy of the evaluation effect.

[0018] Furthermore, in the present invention, the analysis conditions effectively reflect the distribution of a certain type of area. Then, different optimization methods are adaptively selected according to the analysis conditions, making the selection of the optimization method more in line with the actual application scenario, avoiding the problem of poor search efficiency caused by too large a search range and difficult search for public qualification information, helping to improve the relevance of search results, and further improving the accuracy of bid document evaluation.

[0019] Furthermore, in the present invention, the document type of the bid document to be evaluated is determined according to the correlation threshold and the similarity reference value. The similarity between bid documents is effectively reflected by the correlation threshold and the similarity reference value. Then, different methods for determining the document evaluation coefficient are adaptively selected according to the document type, making the selection of the method for determining the document evaluation coefficient more in line with the actual application scenario, avoiding the problem of poor evaluation accuracy, and further improving the accuracy and fairness of the tender. By manually reviewing the bid invitation documents to be evaluated with a document evaluation coefficient less than the preset document evaluation coefficient, not only the cost of manual review is reduced, but also the bid invitation documents to be evaluated that may have risks can be screened out, thereby improving the accuracy of the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is the unit connection diagram of the intelligent bid document review system that integrates quantitative scoring and image recognition of the present invention; Figure 2 It is the flowchart for determining the area type according to the correlation level similarity and the link difference coefficient of the present invention; Figure 3 It is the flowchart for determining the qualification search method according to the hierarchical status of the present invention; Figure 4 It is the flowchart for determining the document type of the bid document to be evaluated according to the correlation threshold and the similarity reference value of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] In order to make the objectives and advantages of the present invention clearer, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0022] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0023] It should be noted that in the description of the present invention, the terms indicating directions or positional relationships such as "upper", "lower", "left", "right", "inner", "outer", etc. are based on the directions or positional relationships shown in the drawings. This is only for convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation to the present invention.

[0024] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection", and "coupling" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two elements. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0025] Please refer to Figures 1 to 4 As shown, the present invention provides an intelligent bid document review system that integrates quantitative scoring and image recognition, including: A data acquisition unit for obtaining the hierarchical information of the target qualification information corresponding to each bid document; A search and analysis unit connected to the data acquisition unit, for determining the region division method according to the judgment conditions, and determining the region type according to the correlation level similarity and the link difference coefficient. The region division method is to perform uniform division according to the comprehensive evaluation value or perform associated division according to the level threshold; An information acquisition unit connected to the search and analysis unit, for determining the level status according to the link reference value and the link radiation coefficient, and determining the qualification search method according to the level status to obtain public qualification information. The qualification search method is to determine the optimization method according to the proportion of the first-class region and the first-class region distribution coefficient or determine the search priority coefficient of the link according to the relevance of the link information; An optimization unit connected to the information acquisition unit, for determining the optimization method according to the analysis conditions. The optimization method is to determine the search region according to the link similarity and the region correlation, or determine the region search duration according to the region threshold; An evaluation unit connected to the data acquisition unit, the information acquisition unit, and the optimization unit respectively, for recording the bid documents corresponding to the target qualification information with a qualification difference degree less than the preset qualification difference degree from the public qualification information as the bid documents to be evaluated, determining the document type of the bid documents to be evaluated according to the association threshold and the similarity reference value, and determining the document evaluation coefficient determination method according to the document type. The document evaluation coefficient determination method is to determine the document evaluation coefficient according to the index fluctuation value and the tender classification value or according to the tender classification value.

[0026] The application scenario of the present invention is the scoring of tender documents. The hierarchical information includes, but is not limited to, the link position, the number of links, and the link name in the target level. The target level is the web page obtained by searching with the enterprise name in the target qualification information. The target level contains several links, and each link corresponds to a sub-level. The present invention extracts the target qualification information in the tender document through image recognition technology. The image recognition technology can collect OCR technology, and specific limitations are not made. The target qualification information includes, but is not limited to, the enterprise name, the business license number, and the financial report data. This is easily understood by those skilled in the art and will not be elaborated herein; In the present invention, several historical records are correspondingly set. Any one of the historical records records at least one reference value of the number of levels, the total threshold value of the levels, the similarity of associated levels, the link radiation coefficient, and the distribution coefficient of the first-class area, etc. in the historical process of tender document scoring. And each historical record corresponds to a qualified mark, which records whether the accuracy of the tender document scoring meets the user's requirements. The qualified mark can be manually recorded.

[0027] Qualification difference degree = (the number of keywords in the target qualification information - the number of the same keywords in the public qualification information and the target qualification information) / the number of keywords in the target qualification information; The value of the preset qualification difference degree can be determined according to the actual application scenario. The larger the value of the preset qualification difference degree, the greater the user's need for no qualification fraud in the document to be scored. A value of the preset qualification difference degree is provided, and the preset qualification difference degree is 30%.

[0028] Specifically, the search and analysis unit responds to the judgment condition to determine the area division method; The judgment condition to which the search and analysis unit responds is that the reference value of the number of levels is less than the preset link difference degree or the link reference value is less than the preset link reference value, and the area division method is uniform division according to the comprehensive evaluation value; The judgment condition to which the search and analysis unit responds is that the reference value of the number of levels is less than the preset reference value of the number of levels and the link reference value is greater than or equal to the preset link reference value, and the area division method is associated division according to the level threshold.

[0029] Among them, in the uniform division according to the comprehensive evaluation value, the target area is divided into rectangular areas with a preset number and equal sizes. The preset number is in a positive correlation with the comprehensive evaluation value; The judgment condition includes a first judgment condition and a second judgment condition. The first judgment condition is that the reference value of the number of levels is less than the preset link difference degree or the link reference value is less than the preset link reference value. The second judgment condition is that the reference value of the number of levels is less than the preset reference value of the number of levels and the link reference value is greater than or equal to the preset link reference value; Divide the target area into rectangular areas with a preset quantity that meet the preset conditions, where each rectangular area is the sub-area obtained by the division; Build a tree structure diagram based on the target level. Conduct node analysis for the target level, extract all sub-levels corresponding to the target level, create several new sub-nodes for each sub-level, connect each sub-node to the root node, conduct node analysis for each sub-level, and connect each sub-level to the corresponding sub-node until the node analysis stops when the level threshold of the sub-level is 0. Finally, build a tree structure diagram with the target level as the root node and each sub-level as the sub-node. This is content that is easy for those skilled in the art to understand and will not be elaborated here; The reference value of the level quantity is the quantity of all sub-nodes in the tree structure diagram built based on the target level, and the reference value of the link is the total quantity of links in the target level; Comprehensive evaluation value = reference value of the level quantity + reference value of the link, and the level threshold is the quantity of links in a single sub-level; Regarding the values of the preset reference value of the level quantity and the preset reference value of the link, the user can determine them according to the actual application scenario. The larger the values of the preset reference value of the level quantity and the preset reference value of the link, the greater the user's need for uniform division according to the comprehensive evaluation value. Provide a value for the preset reference value of the level quantity and the preset reference value of the link. The preset reference value of the link is 15. According to the historical records of uniform division based on the comprehensive evaluation value, record the average value of the reference values of the level quantity corresponding to the historical records that can meet the user's needs as the preset reference value of the level quantity.

[0030] Specifically, the search and analysis unit conducts associated division according to the level threshold, including: Conduct associated analysis for each link in the target level in the preset order. When conducting associated analysis for a single link, record this link as the target link, and record the links outside the target link that have not been recorded in the sub-area as reference links; Accumulate the level threshold corresponding to the target link and the level thresholds corresponding to each link in the preset order one by one until the total value of the level thresholds is greater than the preset total value of the level thresholds. Record the reference links before the preset order of the reference link corresponding to the last accumulated level threshold and the target link as associated links, and record the smallest rectangular area that can contain the associated links as a sub-area; And continue to conduct associated analysis for the links that have not been recorded in the sub-area until all links are recorded in the sub-area; The preset order is the order of the associated distances corresponding to each link from small to large.

[0031] Among them, the links after the preset order of the target link are the links before the order in which the target link is located after sorting all links in ascending order according to the association order; the reference link corresponding to the last accumulated hierarchical value is denoted as the end reference link, and the reference links before the preset order of the end reference link are the reference links before the order in which the end reference link is located; The total hierarchical threshold value is the sum of the hierarchical thresholds corresponding to the accumulated links; the association distance is the shortest distance from the center point of the link to the reference point. The center point of a single link is the intersection of the diagonals of the smallest rectangle that can contain the link, and the reference point is the vertex located at the upper left corner of the target layer; The value of the preset total hierarchical threshold can be determined by the user according to the actual application scenario. The greater the user's demand for the richness of the information contained in the sub-region, the greater the value of the preset total hierarchical threshold. A value of the preset total hierarchical threshold is provided. The historical records of the association division according to the hierarchical threshold are detected, and the average value of the total hierarchical threshold values corresponding to the historical records that can meet the user's needs is denoted as the preset total hierarchical threshold.

[0032] Specifically, the search and analysis unit determines the region type according to the association level similarity and the link difference coefficient. The region types include: A type of region where the association level similarity is less than the preset association level similarity or the link difference coefficient is greater than or equal to the preset link difference coefficient; A type of region where the association level similarity is greater than or equal to the preset association level similarity and the link difference coefficient is less than the preset link difference coefficient.

[0033] Among them, the confirmation method of the association level is as follows: the sub-level corresponding to the link in a single sub-region is denoted as the first sub-level. Association analysis is performed on each first sub-level. When performing association analysis on a single first sub-level, the first sub-level is denoted as the target first sub-level, and the association level corresponding to the target first sub-level is analyzed. The sub-levels corresponding to the nodes adjacent to the left of the node corresponding to the target first sub-level in the tree structure diagram and the sub-levels corresponding to the nodes adjacent to the right of the node corresponding to the target first sub-page are denoted as the association levels, and the association analysis is continued for each association level until the hierarchical threshold of the association level is 0 and the association analysis stops; The sum of the number of links in the matching level corresponding to a single first sub-level is denoted as the link total. The maximum value among the link totals corresponding to each first sub-level is denoted as b1, and the minimum value among the link totals corresponding to each first sub-level is denoted as b2. The link difference coefficient = (b1 - b2) / b1; The values of the preset correlation level similarity and the preset link difference coefficient can be determined by the user according to the actual application scenario. The larger the value of the preset correlation level similarity and the smaller the value of the preset link difference coefficient, the greater the user's need to divide the sub-region into a certain type of region. Provide a set of values for the preset correlation level similarity and the preset link difference coefficient. When the preset link difference coefficient is 50%, record the historical records that can meet the user's needs and determine the sub-region as a certain type of region as reference records. Under the first confirmation condition, the preset correlation level similarity is the average value of the correlation level similarities corresponding to the historical records that determine the correlation level similarity according to the number of identical keywords in the reference records. Under the second confirmation condition, the preset correlation level similarity is the average value of the correlation level similarities corresponding to the historical records that determine the correlation level similarity according to the number of associated keywords in the reference records. The first confirmation condition is that the link information degree is greater than or equal to the preset link information degree, and the second confirmation condition is that the link information degree is less than the preset link information degree.

[0034] Specifically, the correlation level similarity is determined according to the link information degree. If the link information degree is greater than or equal to the preset link information degree, determine the correlation level similarity according to the number of identical keywords, where the correlation level similarity has a positive correlation with the number of identical keywords. If the link information degree is less than the preset link information degree, determine the correlation level similarity according to the number of associated keywords, where the correlation level similarity has a positive correlation with the number of associated keywords.

[0035] Among them, the link information degree is the number of characters in a single link. The value of the preset link information degree can be determined by the user according to the actual application scenario. The larger the value of the preset link information degree, the greater the user's need to determine the correlation level similarity according to the number of associated keywords. Provide a value for the preset link information degree, and the preset link information degree is 20. For a single first sub-level, record this first sub-level as the target sub-level. Record all the correlation levels obtained until the level threshold of the correlation level is 0 when performing correlation analysis on the target sub-level and the target sub-level as the analysis levels corresponding to the target sub-level. Record the links corresponding to each analysis level as the reference links corresponding to the target sub-level. Record the name of the reference link corresponding to the target sub-level as the reference text corresponding to the target sub-level. Record the characters that appear in the analysis levels corresponding to the target sub-level as the associated text corresponding to the target sub-level. The number of identical keywords is the number of keywords that exist in the reference texts corresponding to each first sub-level. The number of associated keywords is the number of keywords that exist in the associated texts corresponding to each first sub-level.

[0036] Specifically, the information acquisition unit determines the level status according to the link reference value and the link radiation coefficient. The level status includes: The first-level state where the link reference value is greater than or equal to the preset link reference value or the link radiation coefficient is greater than or equal to the preset link radiation coefficient; The second-level state where the link reference value is less than the preset link reference value and the link radiation coefficient is less than the preset link radiation coefficient.

[0037] Among them, the link radiation coefficient is the standard deviation of the associated distances corresponding to each link in the target level; the value of the preset link radiation coefficient can be determined by the user according to the actual application scenario. The larger the value of the preset link radiation coefficient, the greater the user's need to determine the search priority coefficient according to the link information relevance. Provide a value of the preset link radiation coefficient, detect the historical record of determining the search priority coefficient according to the link information relevance, and record the average value of the link radiation coefficients corresponding to the historical records that can meet the user's needs as the preset link radiation coefficient.

[0038] Specifically, the information acquisition unit responds to the level state to determine the qualification search method; When the information acquisition unit responds to the first-level state, the qualification search method is to determine the optimization method according to the proportion of the first-type area and the first-type area distribution coefficient; When the information acquisition unit responds to the second-level state, the qualification search method is to determine the search priority coefficient of the link according to the link information relevance; The search priority coefficient has a positive correlation with the link information relevance.

[0039] Among them, the proportion of the first-type area = the number of first-type areas in the target level / the number of sub-areas in the target level; the first-type area distribution coefficient is the average value of the reference distances corresponding to each first-type point. For a single first-type point, record this first-type point as the target first-type point, record the other first-type points except the target first-type point as the reference first-type points, and record the average value of the shortest distances from the target first-type point to each reference first-type point as the reference distance; for a single link, record this link as the target link, record the name of the target link and the text of the sub-level corresponding to the target link as the link text corresponding to the target link, and the link information relevance is the number of the same keywords in the link text corresponding to a single link and the target qualification information.

[0040] Specifically, the optimization unit responds to the analysis condition to determine the optimization method; The analysis condition for which the optimization unit responds is that the proportion of the first-type area is greater than or equal to the preset proportion of the first-type area or the first-type area distribution coefficient is greater than or equal to the preset first-type area distribution coefficient, and the optimization method is to determine the search area according to the link similarity and the regional correlation degree; The analysis condition for optimizing the unit response is that the proportion of the first type of area is less than the preset proportion of the first type of area and the distribution coefficient of the first type of area is less than the preset distribution coefficient of the first type of area. The optimization method is to determine the area search duration of each sub-area according to the area threshold; The search area is the first type of area where the link similarity is less than the preset link similarity and the area correlation is less than the preset area correlation; The area search duration of a single sub-area is positively correlated with the area threshold.

[0041] Among them, for the values of the preset proportion of the first type of area and the preset distribution coefficient of the first type of area, the user can determine them according to the actual application scenario. The smaller the values of the preset proportion of the first type of area and the preset distribution coefficient of the first type of area, the greater the user's need to determine the search area according to the link similarity and the area correlation. Provide a set of values for the preset proportion of the first type of area and the preset distribution coefficient of the first type of area. The preset proportion of the first type of area is 60%. Detect the historical records of determining the area search duration according to the area threshold, and record the average value of the distribution coefficients of the first type of area corresponding to the historical records that can meet the user's needs as the preset distribution coefficient of the first type of area; The confirmation methods for link similarity and area correlation are as follows: For a single first type of area, denote this first type of area as the target first type of area, and denote the other first type of areas outside the target first type of area as reference areas. Denote the average value of the similarity coefficients between the target first type of area and each reference area as the link similarity. The similarity coefficient is the cosine value of the included angle between the text vector corresponding to the target first type of area and the text vector corresponding to a single reference area; The area correlation is the average value of the hierarchical similarities corresponding to each link in the target first type of area. For a single link in the target first type of area, denote this link as the target link, and denote the other links outside the target link as reference links. Denote the cosine value of the included angle between the text vector of the sub-hierarchy corresponding to the target link and the text vector of the sub-hierarchy corresponding to a single reference link as the hierarchical similarity corresponding to the target link; By using the TF-IDF technology to perform text conversion on the text of the first type of area and the text corresponding to each sub-hierarchy, text vectors can be obtained, which is easy for those skilled in the art to understand and will not be elaborated here; For the values of the preset link similarity and the preset area correlation, the user can determine them according to the actual application scenario. The greater the user's need to narrow the search range, the smaller the values of the preset link similarity and the preset area correlation. Provide a set of values for the preset link similarity and the preset area correlation. Detect the historical records of determining the search area according to the link similarity and the area correlation, and record the average value of the link similarities corresponding to the historical records that can meet the user's needs as the preset link similarity, and record the average value of the area correlations corresponding to the historical records that can meet the user's needs as the preset area correlation; The regional threshold = the link difference coefficient - the correlation level similarity, and the regional search duration is the time for searching a single sub-region.

[0042] Specifically, the evaluation unit determines the document type of the tender document to be evaluated according to the correlation threshold and the similarity reference value. The document types include: One type of document where the correlation threshold is greater than or equal to the preset correlation threshold or the similarity reference value is greater than or equal to the preset similarity reference value; Two types of documents where the correlation threshold is less than the preset correlation threshold and the similarity reference value is less than the preset similarity reference value.

[0043] Among them, the confirmation method of the correlation threshold and the similarity reference value is as follows: for a single tender document to be scored, mark this tender document to be scored as the target tender document to be scored, mark the other tender documents to be scored except the target tender document to be scored as the reference tender documents to be scored, and mark the maximum value among the correlation means of the target tender document and each reference tender document as the correlation threshold. The correlation mean is the average value of the correlation coefficients corresponding to each quotation index in the two tender documents to be scored; the quotation indexes include but are not limited to material cost, labor cost, management cost, and tax; For a single quotation index in any two tender documents to be scored, mark this quotation index as the target index. The calculation formula for the correlation coefficient w corresponding to the target index in the two tender documents to be scored is:

[0044] r is the number of data values in the target index of a single tender document to be scored; and are respectively the kth data value in the target index of the two tender documents to be scored, is the average value of each data value in the corresponding tender document to be scored, is the average value of each data value in the corresponding tender document to be scored, k = 1, 2, 3, ……, r; Mark the maximum value among the similarity quantities of the target tender document and each reference tender document as the similarity reference value; the similarity quantity is the number of the same keywords in the target tender document and a single reference tender document; For the values of the preset correlation threshold and the preset similarity reference value, the user can determine them according to the actual application scenario. The smaller the values of the preset correlation threshold and the preset similarity reference value, the greater the user's need to score according to the tender classification value. Provide a set of values for the preset correlation threshold and the preset similarity reference value. Detect the historical records of scoring according to the tender classification value, and mark the average value of the correlation thresholds corresponding to the historical records that can meet the user's needs as the preset correlation threshold, and mark the average value of the similarity reference values corresponding to the historical records that can meet the user's needs as the preset similarity reference value.

[0045] Specifically, the evaluation unit determines the method for determining the document evaluation coefficient according to the document type; For one type of document, the method for determining the document evaluation coefficient is to determine the document evaluation coefficient according to the index fluctuation value and the tender classification value; For the second type of document, the method for determining the document evaluation coefficient is to determine the document evaluation coefficient according to the tender classification value.

[0046] Among them, for one type of document, the document evaluation coefficient = tender classification value - index fluctuation value. The confirmation method of the index fluctuation value is that for a single one-type document, this one-type document is recorded as the target one-type document, and the document to be scored with the largest correlation threshold with the target one-type document is recorded as the reference document. The correlation coefficients corresponding to each quotation index of the target one-type document and the reference document are used, and the standard deviation of each correlation coefficient is recorded as the index fluctuation value; For the second type of document, the document evaluation coefficient has a positive correlation with the tender classification value; The tender classification value = service integrity + solution integrity + price reference value. In the present invention, a deep learning network is applied. The deep learning network includes, but is not limited to, a feedforward neural network, a convolutional neural network, and a recurrent neural network. Users can select according to actual needs, and then learn the to-be-evaluated tender document to obtain the service integrity, solution integrity, and price reference value corresponding to the to-be-evaluated tender document. This is content that is easily understood by those skilled in the art and will not be elaborated specifically; After obtaining the document evaluation coefficients corresponding to each to-be-evaluated tender document, the to-be-evaluated tender documents with document evaluation coefficients less than the preset document evaluation coefficient are subject to manual review. It can be understood that by subjecting the to-be-evaluated tender documents with document evaluation coefficients less than the preset document evaluation coefficient to manual review, not only the cost of manual review is reduced, but also the to-be-evaluated tender documents that may have risks can be screened out, thereby improving the accuracy of the evaluation results; For the value of the preset document evaluation coefficient, the user can determine it according to the actual application scenario. The greater the user's demand for improving the accuracy of patent examination, the greater the value of the preset document evaluation coefficient. Provide a value of the preset document evaluation coefficient, detect the historical records corresponding to the to-be-evaluated tender documents that have not undergone manual review, and record the average value of the document evaluation coefficients corresponding to the historical records that can meet the user's needs as the preset document evaluation coefficient.

[0047] So far, the technical solution of the present invention has been described in combination with the preferred embodiments shown in the drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0048] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention; for those skilled in the art, the present invention may have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent bid document review system that integrates quantitative scoring and image recognition, characterized in that, Including: A data acquisition unit for obtaining the hierarchical information of the target qualification information corresponding to each tender document; A search and analysis unit connected to the data acquisition unit for determining the regional division method according to the judgment condition and determining the regional type according to the correlation level similarity and the link difference coefficient. The regional division method is to make a uniform division according to the comprehensive evaluation value or to make a correlation division according to the level threshold; An information acquisition unit connected to the search and analysis unit for determining the level status according to the link reference value and the link radiation coefficient and determining the qualification search method according to the level status to obtain the public qualification information. The qualification search method is to determine the optimization method according to the proportion of the first-class area and the first-class area distribution coefficient or to determine the search priority coefficient of the link according to the relevance of the link information; An optimization unit connected to the information acquisition unit for determining the optimization method according to the analysis condition. The optimization method is to determine the search area according to the link similarity and the regional correlation, or to determine the regional search duration according to the regional threshold; An evaluation unit connected to the data acquisition unit, the information acquisition unit, and the optimization unit respectively for recording the tender document corresponding to the target qualification information with a qualification difference degree less than the preset qualification difference degree from the public qualification information as the tender document to be evaluated, determining the document type of the tender document to be evaluated according to the correlation threshold and the similarity reference value, and determining the document evaluation coefficient determination method according to the document type. The document evaluation coefficient determination method is to determine the document evaluation coefficient according to the index fluctuation value and the tender classification value or according to the tender classification value; 2. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 1, characterized in that The judgment condition for the search and analysis unit to respond is that the hierarchical quantity reference value is less than the preset link difference degree or the link reference value is less than the preset link reference value, and the regional division method is to make a uniform division according to the comprehensive evaluation value; The judgment condition for the search and analysis unit to respond is that the hierarchical quantity reference value is less than the preset hierarchical quantity reference value and the link reference value is greater than or equal to the preset link reference value, and the regional division method is to make a correlation division according to the level threshold; 3. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 2, wherein The search and analysis unit makes a correlation division according to the level threshold, including: Performing correlation analysis on each link in the target level in the preset order. When performing correlation analysis on a single link, the link is recorded as the target link, and the links outside the target link that are not recorded in the sub-region are recorded as reference links; Adding the level threshold corresponding to the target link to the level thresholds corresponding to each link in the preset order one by one until the total value of the level thresholds is greater than the preset total value of the level thresholds. Record the reference links before the preset order of the reference link corresponding to the last added level threshold and the target link as the associated links, and record the smallest rectangular area that can contain the associated links as a sub-region; And continue to perform correlation analysis on the links not recorded in the sub-region until all links are recorded in the sub-region; The preset order is the order of the associated distances corresponding to each link from small to large; 4. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 3, characterized in that, The search and analysis unit determines the regional type according to the correlation level similarity and the link difference coefficient. The regional types include: A type of area where the associated level similarity is less than the preset associated level similarity or the link difference coefficient is greater than or equal to the preset link difference coefficient; A type of area where the associated level similarity is greater than or equal to the preset associated level similarity and the link difference coefficient is less than the preset link difference coefficient.

5. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 4, characterized in that, The associated level similarity is determined according to the link information degree; If the link information degree is greater than or equal to the preset link information degree, the associated level similarity is determined according to the number of identical keywords, where the associated level similarity has a positive correlation with the number of identical keywords; If the link information degree is less than the preset link information degree, the associated level similarity is determined according to the number of associated keywords, where the associated level similarity has a positive correlation with the number of associated keywords.

6. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 4, wherein The information acquisition unit determines the level status according to the link reference value and the link radiation coefficient, and the level status includes: The first level status where the link reference value is greater than or equal to the preset link reference value or the link radiation coefficient is greater than or equal to the preset link radiation coefficient; The second level status where the link reference value is less than the preset link reference value and the link radiation coefficient is less than the preset link radiation coefficient.

7. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 6, characterized in that, The information acquisition unit responds to the level status to determine the qualification search method; When the information acquisition unit responds to the first level status, the qualification search method is to determine the optimization method according to the proportion of the first type of area and the distribution coefficient of the first type of area; When the information acquisition unit responds to the second level status, the qualification search method is to determine the search priority coefficient of the link according to the link information relevance; The search priority coefficient has a positive correlation with the link information relevance.

8. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 7, characterized in that, The optimization unit responds to the analysis condition to determine the optimization method; The analysis condition that the optimization unit responds to is that the proportion of the first type of area is greater than or equal to the preset proportion of the first type of area or the distribution coefficient of the first type of area is greater than or equal to the preset distribution coefficient of the first type of area, and the optimization method is to determine the search area according to the link similarity and the regional association degree; The analysis condition that the optimization unit responds to is that the proportion of the first type of area is less than the preset proportion of the first type of area and the distribution coefficient of the first type of area is less than the preset distribution coefficient of the first type of area, and the optimization method is to determine the regional search duration of each sub-area according to the regional threshold; The search area is a type of area where the link similarity is less than the preset link similarity and the regional association degree is less than the preset regional association degree; The regional search duration of a single sub-area has a positive correlation with the regional threshold.

9. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 8, characterized in that The evaluation unit determines the document type of the tender document to be evaluated according to the association threshold and the similarity reference value, and the document type includes: A type of document where the association threshold is greater than or equal to the preset association threshold or the similarity reference value is greater than or equal to the preset similarity reference value; A type of document where the association threshold is less than the preset association threshold and the similarity reference value is less than the preset similarity reference value.

10. The intelligent review system for tender documents integrating quantization scoring and image recognition according to claim 9, characterized in that, The evaluation unit determines the document evaluation coefficient determination method according to the document type; For a type of document, the document evaluation coefficient determination method is to determine the document evaluation coefficient according to the index fluctuation value and the tender classification value; For a type of document, the document evaluation coefficient determination method is to determine the document evaluation coefficient according to the tender classification value.

Images