A method and system for analyzing bidding texts based on a bidding and tendering system

Abstract

The present invention discloses a method and system for analyzing bidding texts based on a bidding and tendering system, which relates to the technical field of text analysis and improves the accuracy and comprehensiveness of the analysis of bidding documents. The present invention converts bidding documents and tendering documents into digital files, extracts corresponding characteristic target information and corresponding grade characteristic vectors from the digital files, establishes a semantic association network according to the characteristic target information and corresponding grade characteristic vectors of each digital file, and then connects each semantic association network to obtain a multiple semantic bidding information network and a multiple semantic tendering information network, and obtains a tendering fit value between each multiple semantic tendering information network and the multiple semantic bidding information network by improving the cosine-Jaccard composite similarity algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of text analysis, and particularly to a method and system for analyzing tender texts based on a tendering and bidding system. Background Art

[0002] In tendering and bidding activities, accurately evaluating the compatibility between tender documents and bidding documents is a key link to ensure fairness, justice, and efficient completion of projects. Traditional tendering and bidding evaluation methods mainly rely on manual review. Evaluators need to spend a lot of time and energy carefully studying and comparing the contents of bidding documents and tender documents. This is not only inefficient but also easily interfered by human factors, resulting in subjectivity and deviation in evaluation results.

[0003] With the development of digital technology, most existing tendering and bidding evaluation methods only stay at the level of simple comparison of document texts, making it difficult to deeply explore the internal relationships and semantic relationships between various information in the documents. It is easy to miss some important potential compatibility points or misevaluate some seemingly similar but actually mismatched contents, thus affecting the quality and effect of tendering and bidding. Therefore, a method and system for analyzing tender texts based on a tendering and bidding system are provided. Summary of the Invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a method and system for analyzing tender texts based on a tendering and bidding system.

[0005] In order to achieve the above purpose, the present invention provides the following technical solutions:

[0006] A method for analyzing tender texts based on a tendering and bidding system includes the following steps:

[0007] Step S1: Digitally scan the bidding document to generate a digital bidding document, and set a three-level dynamic capture window to capture the features of the digital bidding document, thereby generating characteristic business target information, characteristic target technical information, and corresponding grade characteristic vectors;

[0008] Step S2: Set grade text relationship nodes for each grade characteristic vector, and connect the grade text relationship nodes in sequence according to the position distribution of each characteristic target information in the digital bidding document, thereby obtaining a semantic association network corresponding to the characteristic business target information and the characteristic target technical information;

[0009] Step S3: Obtain the association values between the grade text relationship nodes in the two semantic association networks, and then connect the semantic association networks corresponding to the characteristic business target information and the characteristic target technical information to obtain a multiple semantic bidding information network;

[0010] Step S4: Generate multiple multi-semantic bid information networks corresponding to the bid documents in the process of Steps S1 to S3, and then obtain the tender matching values between each multi-semantic bid information network and the multi-semantic tender information network through the improved cosine-Jaccard composite similarity algorithm.

[0011] Further, the extraction process of the characteristic business target information and the characteristic target technical information includes:

[0012] Digitally scan the tender document, generate the corresponding tender digital file according to the digital scan result, set up a three-level dynamic capture window and multiple entity information extraction targets, traverse the tender digital file respectively through the three-level dynamic capture windows of each entity information extraction target, and extract the characteristic business target information and the characteristic target technical information from the tender digital file respectively according to the traversal result. Among them, each characteristic target information has three levels of information built in according to the extraction result of the three-level dynamic capture window, namely sentence-level information, paragraph-level information and document-level information;

[0013] Set up a convolutional neural network to extract features from each level of information in each characteristic target information, and generate the corresponding level feature vectors according to the feature extraction results.

[0014] Further, the three-level dynamic capture windows are the document-level dynamic capture window, the paragraph-level dynamic capture window and the sentence-level dynamic capture window in sequence, and the maximum number of captured words of the three-level dynamic capture windows is 1024, 256 and 64 respectively.

[0015] Further, the establishment process of the semantic association network includes:

[0016] Set up level text relationship nodes for each level feature vector, store each level feature vector in the corresponding level text relationship node, and connect each level text relationship node in sequence according to the corresponding relationship and arrangement order of each level feature vector in the tender digital file to obtain the semantic association network;

[0017] In the connection process of each level text relationship node, take the level text relationship node corresponding to the document-level information as the first central node, the level text relationship node corresponding to the paragraph-level information as the second central node, and the level text relationship node corresponding to the sentence-level information as the third central node, and the three central nodes are connected in sequence according to the corresponding information levels.

[0018] Further, the establishment process of the multi-semantic tender information network includes:

[0019] Starting from the hierarchical text relationship nodes of paragraph-level information, match the hierarchical text relationship nodes of the same level in the semantic association networks corresponding to any two different feature target information. At the same time, retrieve the hierarchical text relationship nodes of all statement-level information within the hierarchical text relationship nodes of the two paragraph-level information respectively;

[0020] Furthermore, obtain the association value between the hierarchical text relationship nodes of the two paragraph-level information. Set the statement association value threshold and the paragraph association threshold. If there are two hierarchical text relationship nodes of statement-level information with association values greater than or equal to the statement association value threshold in the semantic association networks of different feature target information, then connect the two hierarchical text relationship nodes of statement-level information; otherwise, do nothing;

[0021] If there are two hierarchical text relationship nodes of paragraph-level information with association values greater than or equal to the paragraph association value threshold in the semantic association networks of different feature target information, then connect the two hierarchical text relationship nodes of paragraph-level information; otherwise, do nothing;

[0022] In the process of connecting the paragraph-level information and the hierarchical text relationship nodes of the statement level in different semantic association networks, obtain the association value between the hierarchical text relationship nodes of the document-level information in different semantic association networks. Then connect the hierarchical text relationship nodes of the document-level information in different semantic association networks to obtain the multiple semantic tender information network corresponding to the tender digital file.

[0023] Furthermore, the calculation formula of the association value is:

[0024] ;

[0025] ;

[0026] where S(a, b) represents the association iteration value between the hierarchical text relationship nodes a and b of the paragraph-level information, α is a decay factor greater than 0 and less than 1, num a and num b respectively represent the total number of hierarchical text relationship nodes of statement-level information contained in the hierarchical text relationship nodes a and b of the paragraph-level information, Γ i (a), Γ j (b), θ, and S(Γ i (a), Γ j (b)) respectively represent the hierarchical feature vectors, the included angle between the i-th and j-th hierarchical text relationship nodes of statement-level information in the hierarchical text relationship nodes a and b of the paragraph-level information, and their association value.

[0027] Further, the process of obtaining the tender compliance value between each multi-semantic tender information network and the multi-semantic bidding information network includes:

[0028] Perform the same processes of steps S1 to S3 on the tender documents of each tendering company, and then generate a multi-semantic tender information network corresponding to each tender document;

[0029] First, set different tender-bid association thresholds for each level of information, use the improved cosine-Jaccard composite similarity algorithm to iterate the tender compliance value between the paragraph-level hierarchical text relationship nodes in the tender document and the bidding document, and then input the iteration result into the tender compliance value calculation formula for secondary iteration, so as to obtain the tender compliance value between each tender document and the bidding document.

[0030] Further, the formula for obtaining the tender compliance value includes:

[0031] ;

[0032] ;

[0033] where α is a decay factor greater than 0 and less than 1, represents the tender compliance value between the paragraph-level information, document-level information, and the entire document between tender document A and bidding document B. N1 and N2 respectively represent the total number of hierarchical text relationship nodes associated with the hierarchical text relationship nodes of the lower-level information on the left side of the equation. , γ, and respectively represent the qth and pth hierarchical text relationship nodes, the included angle, and the association value between them in tender document A and bidding document B. J represents the ratio of the number of hierarchical text relationship nodes with an association value greater than the tender-bid association threshold to the number of hierarchical text relationship nodes with an association value less than or equal to the tender-bid association threshold in the lower-level information on the left side of the equation.

[0034] A tendering and bidding text analysis system based on a tendering and bidding system includes a document feature extraction module, a feature information fusion module, and a tender compliance detection module;

[0035] The document feature extraction module is used to convert the bidding document and the tender document into digital documents, and extract the corresponding feature target information and the corresponding hierarchical feature vectors from the digital documents;

[0036] The feature information fusion module is used to establish a semantic association network for the feature target information and the corresponding hierarchical feature vectors of each digital document, and then connect each semantic association network to obtain a multi-semantic bidding information network and a multi-semantic tender information network;

[0037] The tender compliance detection module is used to obtain the tender compliance value between each multi-semantic tender information network and the multi-semantic tender invitation information network by improving the cosine-Jaccard composite similarity algorithm.

[0038] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0039] 1. By constructing a semantic association network corresponding to the characteristic business objective information and the characteristic target technical information, as well as a multi-semantic tender invitation information network and a multi-semantic tender information network, the present invention realizes in-depth excavation of the internal semantic relationship between various pieces of information in the document, and at the same time effectively avoids the limitations of traditional simple comparison methods, more comprehensively and accurately reflects the compliance degree between the tender document and the tender invitation document, and discovers potential matching points and differences.

[0040] 2. By adopting an improved cosine-Jaccard composite similarity algorithm to calculate the tender compliance value, the algorithm is based on an objective mathematical model, reduces the interference of human factors on the evaluation results, makes the tendering and bidding evaluation results more objective and fair, and improves the credibility and quality of the tendering and bidding activities. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention.

[0042] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] In order to make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation manners obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0044] As Figure 1 shown, a tendering and bidding text analysis method based on a tendering and bidding system includes the following steps:

[0045] Step S1: Digitally scan the tender invitation document to generate a digitized tender invitation document, and set a three-level dynamic capture window to capture the features of the digitized tender invitation document, so as to generate characteristic business objective information, characteristic target technical information, and corresponding hierarchical feature vectors;

[0046] Step S2: Set up rank text relationship nodes for each rank feature vector, and connect the rank text relationship nodes in sequence according to the position distribution of each feature target information in the tender digital file, so as to obtain a semantic association network corresponding to the feature business target information and the feature target technical information;

[0047] Step S3: Obtain the association values between the rank text relationship nodes in the two semantic association networks, and then connect the semantic association networks corresponding to the feature business target information and the feature target technical information to obtain a multi-semantic tender information network;

[0048] Step S4: Generate multi-semantic bid information networks corresponding to multiple bid documents by using the processes of Step S1 to S3, and then obtain the tender matching values between each multi-semantic bid information network and the multi-semantic tender information network through an improved cosine-Jaccard composite similarity algorithm.

[0049] The above Step S1 is implemented through the following process:

[0050] Digitally scan the tender document, generate a corresponding tender digital file according to the digital scan result, and set up a three-level dynamic capture window, where the three-level dynamic capture window is successively the document-level dynamic capture window, the paragraph-level dynamic capture window, and the sentence-level dynamic capture window. The maximum number of captured words in the three-level dynamic capture window is 1024, 256, and 64 respectively;

[0051] Set multiple entity information extraction targets, where the entity information extraction targets include business target information and technical target information. The business target information includes project basic information, bidder qualification requirements, tender price requirements, and contract terms;

[0052] The technical target information includes technical specifications and project implementation plans;

[0053] Set up a three-level dynamic capture window for each entity information extraction target. It should be noted that for the three-level dynamic capture window of each entity information extraction target, different feature extraction paragraphs or feature extraction words are built in. For example, the feature extraction sentences or feature extraction words corresponding to the technical target information include technical parameters, technical scheme requirements, and technical service requirements, etc.;

[0054] Then, traverse the tender digital file through the three-level dynamic capture windows of each entity information extraction target, and extract the feature business target information and the feature target technical information from the tender digital file respectively according to the traversal results. Each feature target information has three levels of information built in according to the extraction results of the three-level dynamic capture window, which are sentence-level information, paragraph-level information, and document-level information in sequence;

[0055] Match the hierarchical information at the same level in each feature target information, obtain the similarity between any two hierarchical information at the same level, and set a similarity threshold.

[0056] If the similarity between two hierarchical information at the same level is greater than or equal to the similarity threshold, randomly eliminate one of the hierarchical information; otherwise, do nothing.

[0057] Set a convolutional neural network to extract features from each hierarchical information in each feature target information, and generate corresponding hierarchical feature vectors according to the feature extraction results.

[0058] The step S2 is implemented through the following process:

[0059] Set a hierarchical text relationship node for each hierarchical feature vector, store each hierarchical feature vector in the corresponding hierarchical text relationship node, and connect each hierarchical text relationship node in sequence according to the corresponding relationship and arrangement order of each hierarchical feature vector in the tender digital file to obtain a semantic association network.

[0060] In the connection process of each hierarchical text relationship node, use the hierarchical text relationship node corresponding to the document-level information as the first central node, the hierarchical text relationship node corresponding to the paragraph-level information as the second central node, and the hierarchical text relationship node corresponding to the sentence-level information as the third central node. The three central nodes are connected in sequence according to the corresponding information levels.

[0061] For example, there is a document-level information composed of multiple paragraph-level information and non-paragraph-level information scattered. Therefore, the hierarchical feature vector corresponding to this document-level information is composed of the hierarchical feature vectors of multiple paragraph-level information in order. Corresponding to the hierarchical text relationship node, the hierarchical text relationship node corresponding to the hierarchical feature vector of the document-level information is connected to it in sequence by the hierarchical feature vectors of multiple paragraph-level information. The connection relationship between the hierarchical text relationship nodes of the paragraph-level information and the sentence-level information is the same.

[0062] The step S3 is implemented through the following process:

[0063] Starting from the hierarchical text relationship node of the paragraph-level information, match the hierarchical text relationship nodes at the same level in the semantic association networks corresponding to any two different feature target information, and simultaneously retrieve the hierarchical text relationship nodes of all sentence-level information within the hierarchical text relationship nodes of the two paragraph-level information respectively.

[0064] Furthermore, obtain the association value between the hierarchical text relationship nodes of the two paragraph-level information, where the calculation formula of the association value is:

[0065] ;

[0066] ;

[0067] Where S(a, b) represents the associated iteration value between the hierarchical text relationship nodes a and b of the paragraph-level information, α is a decay factor greater than 0 and less than 1, num a and num b respectively represent the total number of hierarchical text relationship nodes of the statement-level information contained within the hierarchical text relationship nodes a and b of the paragraph-level information, Γ i (a), Γ j (b), θ, and S(Γ i (a), Γ j (b)) respectively represent the hierarchical feature vectors of the hierarchical text relationship nodes of the i-th and j-th statement-level information within the hierarchical text relationship nodes a and b of the paragraph-level information, the included angle between them, and the associated value;

[0068] Set the statement association value threshold and the paragraph association threshold. If there are two hierarchical text relationship nodes of statement-level information with associated values greater than or equal to the statement association value threshold in the semantic association network of different feature target information, then connect the two hierarchical text relationship nodes of statement-level information, otherwise do nothing;

[0069] If there are two hierarchical text relationship nodes of paragraph-level information with associated values greater than or equal to the paragraph association value threshold in the semantic association network of different feature target information, then connect the two hierarchical text relationship nodes of paragraph-level information, otherwise do nothing;

[0070] In the process of connecting the hierarchical text relationship nodes of paragraph-level information and statement-level in different semantic association networks, obtain the associated values between the hierarchical text relationship nodes of document-level information in different semantic association networks, and then connect the hierarchical text relationship nodes of document-level information in different semantic association networks to obtain the multiple semantic tender information network corresponding to the tender digital file;

[0071] It should be noted that when there are new clauses added to the tender digital file subsequently, repeat the process of generating the semantic association network and connecting each semantic association network to generate the multiple semantic tender information network.

[0072] The step S4 is implemented through the following process:

[0073] Execute the same process of steps S1 to S3 on the tender documents of each tendering company, and then generate the multiple semantic tender information networks corresponding to each tender document;

[0074] Set different tender-bid correlation thresholds for each level of information, and use the improved cosine-Jaccard composite similarity algorithm to sequentially obtain the tender-fit values between the multiple semantic tender information networks corresponding to each tender document and the multiple semantic tender information networks;

[0075] The formula for calculating the tender-fit value is:

[0076] ;

[0077] ;

[0078] where α is a decay factor greater than 0 and less than 1, represents the tender-fit value between the paragraph-level information, document-level information, and the entire document between tender document A and tender document B. N1 and N2 respectively represent the total number of level text relationship nodes associated with the level text relationship nodes of the lower-level information on the left side of the equation, 、γ and respectively represent the qth and pth level text relationship nodes, the included angle, and the association value between tender document A and tender document B. J represents the ratio of the number of level text relationship nodes in the lower-level information on the left side of the equation whose association value is greater than the tender-bid correlation threshold to the number of level text relationship nodes less than or equal to the tender-bid correlation threshold;

[0079] First, iterate the tender-fit value between the paragraph-level level text relationship nodes in the tender document and the tender document, and then input the iteration result into the tender-fit value calculation formula for secondary iteration to obtain the tender-fit value between each tender document and the tender document.

[0080] The present invention also discloses a tender-bid text analysis system based on a tender-bid system, including a document feature extraction module, a feature information fusion module, and a tender-fit detection module;

[0081] The document feature extraction module is used to convert the tender document and the tender document into digital documents, and extract the corresponding feature target information and the corresponding level feature vectors from the digital documents;

[0082] The feature information fusion module is used to establish a semantic association network for the feature target information and the corresponding level feature vectors of each digital document, and then connect each semantic association network to obtain a multiple semantic tender information network and a multiple semantic tender information network;

[0083] The tender-fit detection module is used to obtain the tender-fit value between each multiple semantic tender information network and the multiple semantic tender information network through the improved cosine-Jaccard composite similarity algorithm.

[0084] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for analyzing bidding texts based on a bidding and tendering system, characterized in that, It includes the following steps: Step S1: Digitally scan the tender document to generate a digital tender document, and set a three-level dynamic capture window to capture the features of the digital tender document, thereby generating characteristic business target information, characteristic target technical information, and corresponding level characteristic vectors; Step S2: Set level text relationship nodes for each level characteristic vector, and connect the level text relationship nodes in sequence according to the corresponding relationship and arrangement order of each level characteristic vector in the digital tender document, thereby obtaining a semantic association network corresponding to the characteristic business target information and the characteristic target technical information; Step S3: Obtain the association values between the level text relationship nodes in the two semantic association networks, and then connect the semantic association networks corresponding to the characteristic business target information and the characteristic target technical information to obtain a multi-semantic tender information network; Step S4: Generate multi-semantic bid information networks corresponding to multiple bid documents by using the processes of Steps S1 to S3, and then obtain the tender matching values between each multi-semantic bid information network and the multi-semantic tender information network through an improved cosine-Jaccard composite similarity algorithm.

2. The bidding text analysis method based on a bidding and tendering system according to claim 1, characterized in that, The extraction process of the characteristic business target information and the characteristic target technical information includes: Digitally scan the tender document, generate a digital tender document according to the digital scan result, set a three-level dynamic capture window and multiple entity information extraction targets, traverse the digital tender document through the three-level dynamic capture window of each entity information extraction target, and extract the characteristic business target information and the characteristic target technical information from the digital tender document respectively according to the traversal result. Among them, according to the extraction result of the three-level dynamic capture window, each characteristic target information has three levels of information built-in, which are sentence-level information, paragraph-level information, and document-level information in sequence; Set a convolutional neural network to extract features from each level of information in each characteristic target information, and generate corresponding level characteristic vectors according to the feature extraction results.

3. The method for analyzing tendering and bidding texts based on a tendering and bidding system according to claim 2, wherein The three-level dynamic capture windows are the document-level dynamic capture window, the paragraph-level dynamic capture window, and the sentence-level dynamic capture window in sequence, and the maximum number of captured words of the three-level dynamic capture windows is 1024, 256, and 64 respectively.

4. The bidding text analysis method based on a bidding system according to claim 2, wherein, The establishment process of the semantic association network includes: Set level text relationship nodes for each level characteristic vector, store each level characteristic vector in the corresponding level text relationship node, and connect the level text relationship nodes in sequence according to the corresponding relationship and arrangement order of each level characteristic vector in the digital tender document to obtain a semantic association network.

5. The method for analyzing bidding documents based on a bidding system according to claim 4, wherein, The establishment process of the multi-semantic tender information network includes: Starting from the level text relationship nodes of the paragraph-level information, match the level text relationship nodes of the same level in the semantic association networks corresponding to any two different characteristic target information, and at the same time, respectively retrieve the level text relationship nodes of all sentence-level information in the level text relationship nodes of the two paragraph-level information; Obtain the associated values between the level text relationship nodes of each level information, and then connect the level text relationship nodes in different semantic association networks according to the associated values to obtain the multiple semantic tender information network corresponding to the tender digital file.

6. The method for analyzing tendering and bidding texts based on a tendering and bidding system according to claim 5, characterized in that, The process of obtaining the tender matching values between each multiple semantic tender information network and the multiple semantic tender information network includes: First, set different tender-bid association thresholds for each level of information, use the improved cosine-Jaccard composite similarity algorithm to iterate the tender matching values between the paragraph-level level text relationship nodes in the tender document and the bid document, and then input the iteration result into the tender matching value calculation formula for secondary iteration, so as to obtain the tender matching values between each tender document and the tender document.

7. A method for analyzing bidding texts based on a bidding and tendering system according to claim 6, characterized in that, The formula for obtaining the tender matching value includes: T(A q , B p ) = α|A q | × |B p | × cosγ; where α is an attenuation factor greater than 0 and less than 1, T(A, B) represents the tender compliance value between the paragraph-level information, document-level information, and the entire document between tender document A and bidding document B, N1 and N2 respectively represent the total number of hierarchical text relationship nodes associated with the hierarchical text relationship nodes of the lower-level information on the left side of the equation, A q , B p , γ, and T(A q , B p ) respectively represent the q-th and p-th hierarchical text relationship nodes, the included angle, and the association value between tender document A and bidding document B, and J represents the ratio of the number of hierarchical text relationship nodes with association values greater than the tender-bid association threshold and less than or equal to the tender-bid association threshold in the lower-level information on the left side of the equation.

8. A tendering and bidding text analysis system based on a tendering and bidding system, which is used to implement the tendering and bidding text analysis method based on a tendering and bidding system according to any one of claims 1 to 7, and is characterized in that, It includes a document feature extraction module, a feature information fusion module, and a tender matching detection module; The document feature extraction module is used to convert the tender document and the bid document into a tender digital file, and extract the corresponding feature target information and the corresponding level feature vector from the tender digital file; The feature information fusion module is used to establish a semantic association network according to the corresponding relationship and arrangement order of each level feature vector in the tender digital file, and then connect each semantic association network to obtain a multiple semantic tender information network and a multiple semantic tender information network; The tender matching detection module is used to obtain the tender matching values between each multiple semantic tender information network and the multiple semantic tender information network through the improved cosine-Jaccard composite similarity algorithm.

Images