Artificial intelligence-based bid document review assistance method and system

Abstract

The application relates to the technical field of information technology of construction engineering, and particularly discloses a bidding document review auxiliary method and system based on artificial intelligence, which comprises bidding document preprocessing, technical specification similarity analysis, commercial specification consistency analysis and analysis result collection. The scheme establishes an objective quantitative evaluation system based on the combination of TF-IDF statistical characteristics and pre-training word vector semantic characteristics, provides accurate numerical similarity scores, enhances the robustness of technical specification text content detection through deep semantic understanding, realizes integrated data analysis through a unified numerical extraction framework, can find cross-table type correlation abnormalities, adopts adaptive extraction technology based on numerical pattern recognition, constructs a multi-dimensional abnormal pattern recognition system, effectively detects bid-rigging behavior among bidders, introduces feature contribution degree analysis and abnormal pattern recognition mechanism, and greatly improves the reliability of analysis results.

Description

Technical Field

[0001] This invention relates to the field of information technology in construction engineering, specifically to a method and system for assisting in the review of tender documents based on artificial intelligence. Background Technology

[0002] Bidding and tendering for construction projects is a core link in the allocation of resources in the construction market. However, bid rigging and collusion severely undermine the fairness and competitiveness of the bidding process. Traditional tender documents rely on manual review, which is highly subjective, inconsistent in standards, inefficient, and lacks in-depth semantic analysis capabilities, making it difficult to provide systematic similarity data support. Traditional commercial bid analysis is fragmented, typically requiring separate checks of bills of quantities, material and equipment tables, and unit prices, resulting in a disjointed analysis process. Anomaly detection methods lack multi-dimensional pattern recognition capabilities, and risk assessment lacks quantitative support. Summary of the Invention

[0003] To address the aforementioned issues and overcome the shortcomings of existing technologies, this invention provides an AI-based method and system for assisting in the review of tender documents. Addressing the problems of traditional manual review methods, which are characterized by strong subjectivity, inconsistent standards, low efficiency, lack of deep semantic analysis capabilities, and difficulty in providing systematic similarity data support, this solution establishes an objective quantitative evaluation system based on a combination of TF-IDF statistical features and pre-trained word vector semantic features. This system provides accurate numerical similarity scores and enhances the robustness of technical bid content detection by integrating deep semantic understanding. Furthermore, addressing the fragmented analysis dimensions of traditional commercial bids, which typically require separate checks of itemized lists, material and equipment tables, and comprehensive unit prices, resulting in a disjointed analysis process, a lack of multi-dimensional pattern recognition capabilities in anomaly detection, and a lack of quantitative support for risk assessment, this solution achieves integrated analysis of all data through a unified numerical extraction framework. This framework can detect cross-table type correlation anomalies and employs adaptive extraction technology based on numerical pattern recognition to construct a multi-dimensional anomaly pattern recognition system. This effectively detects bid-rigging behavior among bidders and introduces feature contribution analysis and anomaly pattern recognition mechanisms, significantly improving the credibility of the analysis results.

[0004] The technical solution adopted in this invention is as follows: The artificial intelligence-based bid document review assistance method provided by this invention includes the following steps:

[0005] Step S1: Tender document preprocessing. Collect the tender documents from each bidder and convert them into a unified format. Preprocess the tender documents, including document parsing, core data extraction, data cleaning and standardization.

[0006] Step S2: Technical bid similarity analysis. Calculate the similarity of the technical bid text content in the tender documents, establish a quantitative evaluation system based on a combination of TF-IDF statistical features and pre-trained word vector semantic features, and generate a technical bid similarity analysis report.

[0007] Step S3: Commercial bid consistency analysis. A unified data analysis is performed on the commercial bid price data in the bid documents. A machine learning model is used to conduct a risk assessment of the bidders and generate a commercial bid consistency analysis report.

[0008] Step S4: Summarize the analysis results. Statistically summarize the analysis results of each dimension of the technical and commercial bids, calculate the overall consistency index of the bid documents, provide an interactive manual review interface for review experts, integrate automated analysis results with manual review opinions, generate a structured review report and support export.

[0009] Furthermore, in step S2, the technical similarity analysis specifically includes the following steps:

[0010] Step S21: Text segmentation. Extract the chapter structure of the technical bid text content in each bidder's bid documents, use regular expressions to segment the technical bid text content into logical paragraphs, and further divide it into sentences separated by periods, question marks, and exclamation marks as analysis units.

[0011] Step S22: Sentence segmentation. Segment each sentence into words and create a word sequence for the technical bid text content in each bidder's bid document.

[0012] Step S23: Text vectorization. A multi-level text vectorization method based on the TF-IDF weighted bag-of-words model and pre-trained word vectors is used to convert the text content into vector form. The specific steps are as follows:

[0013] Step S231: Vocabulary Construction. Summarize the word sequences from the technical bid texts of all bidders to construct a vocabulary, with the following dimensions: The format is as follows: ;

[0014] In the formula, Represents a vocabulary list. Indicates the size of the vocabulary. Different words;

[0015] Step S232: Document TF-IDF Vector Generation. Treat the technical bid text of each bidder as a document. For each word in the document and vocabulary, calculate the TF-IDF weight value of that word in the document. Arrange the TF-IDF weight values ​​of all words in vocabulary order to form the document's TF-IDF vector, with dimension [missing value]. The formula used is as follows: ; ;

[0016] In the formula, Representing words, Represents a document. Words In the document TF-IDF weight values ​​in Words In the document word frequency in Words Inverse document frequency, Document TF-IDF vector;

[0017] Step S233: Load the word vector model, using a pre-trained word vector model to map each word to... 3D word vectors;

[0018] Step S234: Document semantic vector generation. Average the word vectors of all words in each sentence of the document to obtain the sentence vector. Then average the vectors of all sentences in the document to obtain the document semantic vector, with dimension [missing information]. The formula used is as follows: ;

[0019] In the formula, Document semantic vectors, Document A collection of sentences, Document The total number of sentences in the text Document One of the sentences, Sentence A collection of words, Sentence The total number of words in Sentence One of the words, Represents word vectors;

[0020] Step S235: Document vector generation. The final document vector is generated through weighted concatenation, with dimensions [dimension number missing]. The formula used is as follows: ;

[0021] In the formula, Represents a document vector. The weight coefficients represent the semantic vectors of a document. This represents a vector concatenation operation;

[0022] Step S24: Document-level similarity calculation. Calculate the cosine similarity between the content of any two bidders' technical bid documents. The closer the value is to 1, the closer the two documents are in the vector space, meaning the more similar the technical bid documents are. Set a document-level similarity threshold; technical bid documents exceeding this threshold will be marked as similar pairs. The formula used is as follows: ;

[0023] In the formula, Indicates document-level similarity. , This refers to the content of the technical bids from any two bidders. , Indicates the bidder , Document vectors;

[0024] Step S25: Generate a technical bid similarity analysis report, which includes: a document-level similarity matrix between all bidders, a list of bidder pairs marked as similar and their similarity scores.

[0025] Furthermore, in step S3, the commercial bid consistency analysis specifically includes the following steps:

[0026] Step S31: Numerical data extraction. The commercial tender document is scanned using a numerical pattern recognition algorithm to identify and extract all numerical data. Each numerical data and its metadata together constitute a data item. All data items are normalized and converted into a unified unit of measurement to construct a set of numerical data items.

[0027] Step S32: Data Feature Analysis. Perform feature analysis on the set of numerical data items and automatically generate analysis indicators. The specific steps are as follows:

[0028] Step S321: Quantify the data distribution characteristics, calculate the global mean, standard deviation and coefficient of variation of the numerical data, and calculate the within-group statistical characteristics of the bidders in groups;

[0029] Step S322: Numerical data classification. The K-means clustering algorithm is used to perform unsupervised classification of numerical data, identify and label data categories;

[0030] Step S323: Adaptive generation of analysis indicators, assign similarity calculation indicators to each data category, assign weights to each data item based on the coefficient of variation, with data items having higher weights for smaller coefficients of variation, and construct a set of analysis indicators;

[0031] Step S33: Multi-dimensional analysis of numerical consistency. Based on the analysis indicators, analyze the numerical consistency among the bidders. The specific steps are as follows:

[0032] Step S331: Numerical Vector Construction. Based on the metadata of the data items, the data items of all bidders are automatically aligned to ensure that the same position corresponds to the same quotation content. A numerical vector is constructed for each bidder, in the following form: ;

[0033] In the formula, Represents a numerical vector. Indicates the total number of data items. Representing different data items;

[0034] Step S332: Multi-dimensional similarity calculation, calculate the similarity index between bidder pairs, including cosine similarity, weighted similarity, and quantile similarity;

[0035] Step S333: Numerical pattern anomaly detection, calculate anomaly pattern features, including the proportion of completely consistent items, linear correlation coefficient, and uniformity of proportional changes;

[0036] Step S34: Machine learning risk assessment, the specific steps are as follows:

[0037] Step S341: Feature engineering. Based on the multi-dimensional similarity calculation results and numerical pattern anomaly detection results, construct a multi-dimensional feature vector for each bidder pair and introduce historical labeled data to train a risk assessment model.

[0038] Step S342: Risk probability calculation, calculate the risk probability of bid rigging for each bidder pair, and identify the main risk contribution characteristics;

[0039] Step S343: Risk classification, automatically classifying bidders based on risk probability;

[0040] Step S35: Generate a commercial bid consistency analysis report, which includes: the probability and classification results of bid rigging risk, the results of multi-dimensional similarity calculation, and the results of numerical pattern anomaly detection.

[0041] The present invention provides an artificial intelligence-based bid document review assistance system, which includes a bid document preprocessing module, a technical bid similarity analysis module, a commercial bid consistency analysis module, and an analysis result summary module.

[0042] The bid document preprocessing module collects the bid documents from each bidder and converts them into a unified format. It then preprocesses the bid documents and sends the preprocessed bid documents to the technical bid similarity analysis module and the commercial bid consistency analysis module.

[0043] The technical bid similarity analysis module calculates the similarity of the technical bid text content in the bid documents, establishes a quantitative evaluation system based on the combination of TF-IDF statistical features and pre-trained word vector semantic features, generates a technical bid similarity analysis report, and sends the technical bid similarity analysis report to the analysis result summary module.

[0044] The commercial bid consistency analysis module performs unified data analysis on the commercial bid price data in the bid documents, uses machine learning models to conduct risk assessment on bidders, generates a commercial bid consistency analysis report, and sends the commercial bid consistency analysis report to the analysis results summary module.

[0045] The analysis results summary module statistically summarizes the analysis results of various dimensions of the technical and commercial bids, calculates the overall consistency index of the bid documents, provides an interactive manual review interface for review experts, integrates automated analysis results with manual review opinions, generates a structured review report, and supports export.

[0046] The beneficial effects achieved by the present invention using the above solution are as follows:

[0047] (1) In view of the problems that traditional tender documents adopt manual review method, which is subjective, inconsistent in standards, inefficient and lacks deep semantic analysis capability, making it difficult to provide systematic similarity data support, this solution establishes an objective quantitative evaluation system based on the combination of TF-IDF statistical features and pre-trained word vector semantic features, provides accurate numerical similarity scores, and enhances the robustness of technical tender text content detection by integrating deep semantic understanding.

[0048] (2) In response to the fragmented dimensions of traditional commercial bid analysis, which usually requires separate checks of tables such as the list of sub-items, labor, materials and machinery, and comprehensive unit price, the analysis process is fragmented, the anomaly detection methods lack multi-dimensional pattern recognition capabilities, and the risk assessment lacks quantitative support, this solution achieves integrated analysis of all data through a unified numerical extraction framework, which can discover cross-table type correlation anomalies. It adopts an adaptive extraction technology based on numerical pattern recognition to construct a multi-dimensional anomaly pattern recognition system, effectively detects bid-rigging behavior among bidders, and introduces feature contribution analysis and anomaly pattern recognition mechanism to significantly improve the credibility of the analysis results. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the AI-based bid document review assistance method proposed in this invention.

[0050] Figure 2 This is a schematic diagram of the AI-based bid document review assistance system proposed in this invention.

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0052] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0053] Example 1, see Figure 1 The present invention provides an artificial intelligence-based bid document review assistance method, which includes the following steps:

[0054] Step S1: Tender document preprocessing. Collect the tender documents from each bidder and convert them into a unified format. Preprocess the tender documents, including document parsing, core data extraction, data cleaning and standardization.

[0055] Step S2: Technical bid similarity analysis. Calculate the similarity of the technical bid text content in the tender documents, establish a quantitative evaluation system based on a combination of TF-IDF statistical features and pre-trained word vector semantic features, and generate a technical bid similarity analysis report.

[0056] Step S3: Commercial bid consistency analysis. A unified data analysis is performed on the commercial bid price data in the bid documents. A machine learning model is used to conduct a risk assessment of the bidders and generate a commercial bid consistency analysis report.

[0057] Step S4: Summarize the analysis results. Statistically summarize the analysis results of each dimension of the technical and commercial bids, calculate the overall consistency index of the bid documents, provide an interactive manual review interface for review experts, integrate automated analysis results with manual review opinions, generate a structured review report and support export.

[0058] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S2, the technical mark similarity analysis specifically includes the following steps:

[0059] Step S21: Text segmentation. Extract the chapter structure of the technical bid text content in each bidder's bid documents, use regular expressions to segment the technical bid text content into logical paragraphs, and further divide it into sentences separated by periods, question marks, and exclamation marks as analysis units.

[0060] Step S22: Sentence segmentation. Use the jieba segmentation tool in conjunction with a custom dictionary in the construction engineering field to segment each sentence and establish a word sequence for the technical bid text content in each bidder's bid document;

[0061] Step S23: Text vectorization. A multi-level text vectorization method based on the TF-IDF weighted bag-of-words model and pre-trained word vectors is used to convert the text content into vector form. The specific steps are as follows:

[0062] Step S231: Vocabulary Construction. Summarize the word sequences from the technical bid texts of all bidders to construct a vocabulary, with the following dimensions: The format is as follows: ;

[0063] In the formula, Represents a vocabulary list. Indicates the size of the vocabulary. Different words;

[0064] Step S232: Document TF-IDF Vector Generation. Treat the technical bid text of each bidder as a document. For each word in the document and vocabulary, calculate the TF-IDF weight value of that word in the document. Arrange the TF-IDF weight values ​​of all words in vocabulary order to form the document's TF-IDF vector, with dimension [missing value]. The formula used is as follows: ; ;

[0065] In the formula, Representing words, Represents a document. Words In the document TF-IDF weight values ​​in Words In the document word frequency in Words Inverse document frequency, Document TF-IDF vector;

[0066] Step S233: Load the word vector model, using a pre-trained word vector model to map each word to... 3D word vectors;

[0067] Step S234: Document semantic vector generation. Average the word vectors of all words in each sentence of the document to obtain the sentence vector. Then average all sentence vectors in the document to obtain the document semantic vector, with dimension [dimension missing]. The formula used is as follows: ;

[0068] In the formula, Document semantic vectors, Document A collection of sentences, Document The total number of sentences in the text Document One of the sentences, Sentence A collection of words, Sentence The total number of words in Sentence One of the words, Represents word vectors;

[0069] Step S235: Document vector generation. The final document vector is generated through weighted concatenation, with dimensions [dimension number missing]. The formula used is as follows: ;

[0070] In the formula, Represents a document vector. The weight coefficients represent the semantic vectors of a document. This represents a vector concatenation operation;

[0071] Step S24: Document-level similarity calculation. Calculate the cosine similarity between the content of any two bidders' technical bid documents. The closer the value is to 1, the closer the two documents are in the vector space, meaning the more similar the technical bid documents are. Set the document-level similarity threshold to 0.75. Technical bid documents exceeding this threshold will be marked as similar pairs. The formula used is as follows: ;

[0072] In the formula, Indicates document-level similarity. , This refers to the content of the technical bids from any two bidders. , Indicates the bidder , Document vectors;

[0073] Step S25: Generate a technical bid similarity analysis report, which includes: a document-level similarity matrix between all bidders, a list of bidder pairs marked as similar and their similarity scores.

[0074] By performing the aforementioned operations, this solution addresses the problems of traditional manual review methods for tender documents, which are characterized by strong subjectivity, inconsistent standards, low efficiency, lack of deep semantic analysis capabilities, and difficulty in providing systematic similarity data support. It establishes an objective quantitative evaluation system based on a combination of TF-IDF statistical features and pre-trained word vector semantic features, providing accurate numerical similarity scores. By integrating deep semantic understanding, it enhances the robustness of technical tender text content detection.

[0075] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S3, the commercial bid consistency analysis specifically includes the following steps:

[0076] Step S31: Numerical data extraction. The commercial tender document is scanned using a numerical pattern recognition algorithm to identify and extract all numerical data. Each numerical data and its metadata together constitute a data item. All data items are normalized and converted into a unified unit of measurement to construct a set of numerical data items.

[0077] Step S32: Data Feature Analysis. Perform feature analysis on the set of numerical data items and automatically generate analysis indicators. The specific steps are as follows:

[0078] Step S321: Quantify the data distribution characteristics, calculate the global mean, standard deviation and coefficient of variation of the numerical data, and calculate the within-group statistical characteristics of the bidders in groups;

[0079] Step S322: Numerical data classification. The K-means clustering algorithm is used to perform unsupervised classification of numerical data, identify and label data categories;

[0080] Step S323: Adaptive generation of analysis indicators, assign similarity calculation indicators to each data category, assign weights to each data item based on the coefficient of variation, with data items having higher weights for smaller coefficients of variation, and construct a set of analysis indicators;

[0081] Step S33: Multi-dimensional analysis of numerical consistency. Based on the analysis indicators, analyze the numerical consistency among the bidders. The specific steps are as follows:

[0082] Step S331: Numerical Vector Construction. Based on the metadata of the data items, the data items of all bidders are automatically aligned to ensure that the same position corresponds to the same quotation content. A numerical vector is constructed for each bidder, in the following form: ;

[0083] In the formula, Represents a numerical vector. Indicates the total number of data items. Representing different data items;

[0084] Step S332: Multi-dimensional similarity calculation, calculate the similarity index between bidder pairs, including cosine similarity, weighted similarity, and quantile similarity;

[0085] Step S333: Numerical pattern anomaly detection, calculate anomaly pattern features, including the proportion of completely consistent items, linear correlation coefficient, and uniformity of proportional changes;

[0086] Step S34: Machine learning risk assessment, the specific steps are as follows:

[0087] Step S341: Feature engineering. Based on the multi-dimensional similarity calculation results and numerical pattern anomaly detection results, construct a multi-dimensional feature vector for each bidder pair and introduce historical labeled data to train a risk assessment model.

[0088] Step S342: Risk probability calculation, calculate the risk probability of bid rigging for each bidder pair, and identify the main risk contribution characteristics;

[0089] Step S343: Risk classification, automatically classifying bidders based on risk probability: a bid-rigging risk probability below 0.3 is low risk, a bid-rigging risk probability between 0.3 and 0.7 is medium risk, and a bid-rigging risk probability above 0.7 is high risk;

[0090] Step S35: Generate a commercial bid consistency analysis report, which includes: the probability and classification results of bid rigging risk, the results of multi-dimensional similarity calculation, and the results of numerical pattern anomaly detection.

[0091] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S333, the abnormal mode features are specifically as follows: Percentage of items with exact same data: The bidder's numerical patterns that are completely identical on multiple data items; Linear correlation coefficient: There is a linear relationship between the numerical vectors of the bidders; Consistency of geometrical changes: The numerical vectors of the bidders exhibit a geometrical relationship.

[0092] By performing the aforementioned operations, this solution addresses the fragmented nature of traditional commercial bid analysis, which typically requires separate checks of tables such as itemized lists, material and equipment lists, and comprehensive unit prices. This fragmented analysis process, coupled with the lack of multi-dimensional pattern recognition capabilities in anomaly detection and the absence of quantitative support for risk assessment, resolves the issues. This solution achieves integrated analysis of all data through a unified numerical extraction framework, enabling the discovery of cross-table type correlation anomalies. It employs adaptive extraction technology based on numerical pattern recognition to construct a multi-dimensional anomaly pattern recognition system, effectively detecting bid-rigging behavior among bidders. Furthermore, the introduction of feature contribution analysis and anomaly pattern recognition mechanisms significantly enhances the reliability of the analysis results.

[0093] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the consistency analysis of materials, labor, and machinery specifically involves: statistically summarizing the analysis results of each dimension of the technical bid and commercial bid, calculating the overall consistency index of the bid documents using a weighted average method, classifying risk levels according to preset thresholds, displaying the analysis results in a visual form, marking high-risk anomalies and the bidders involved, providing an interactive manual verification interface for review experts, supporting the review, annotation, and supplementary opinions on the anomalies marked by the system, integrating automated analysis results with manual review opinions, generating a structured review report, and supporting export.

[0094] Example 6, see Figure 2 Based on the above embodiments, the present invention provides an artificial intelligence-based bid document review assistance system, which includes a bid document preprocessing module, a technical bid similarity analysis module, a commercial bid consistency analysis module, and an analysis result summary module.

[0095] The bid document preprocessing module collects the bid documents from each bidder and converts them into a unified format. It then preprocesses the bid documents and sends the preprocessed bid documents to the technical bid similarity analysis module and the commercial bid consistency analysis module.

[0096] The technical bid similarity analysis module calculates the similarity of the technical bid text content in the bid documents, establishes a quantitative evaluation system based on the combination of TF-IDF statistical features and pre-trained word vector semantic features, generates a technical bid similarity analysis report, and sends the technical bid similarity analysis report to the analysis result summary module.

[0097] The commercial bid consistency analysis module performs unified data analysis on the commercial bid price data in the bid documents, uses machine learning models to conduct risk assessment on bidders, generates a commercial bid consistency analysis report, and sends the commercial bid consistency analysis report to the analysis results summary module.

[0098] The analysis results summary module statistically summarizes the analysis results of various dimensions of the technical and commercial bids, calculates the overall consistency index of the bid documents, provides an interactive manual review interface for review experts, integrates automated analysis results with manual review opinions, generates a structured review report, and supports export.

[0099] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0100] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0101] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. An artificial intelligence-based method for assisting in the review of tender documents, characterized by: The method includes the following steps: Step S1: Tender document preprocessing. Collect the tender documents from each bidder and convert them into a uniform format. Preprocess the tender documents. Step S2: Technical bid similarity analysis. Calculate the similarity of the technical bid text content in the tender documents, establish a quantitative evaluation system based on a combination of TF-IDF statistical features and pre-trained word vector semantic features, and generate a technical bid similarity analysis report. Step S3: Commercial bid consistency analysis. A unified data analysis is performed on the commercial bid price data in the bid documents. A machine learning model is used to conduct a risk assessment of the bidders and generate a commercial bid consistency analysis report. Step S4: Summarize the analysis results. Statistically summarize the analysis results of each dimension of the technical and commercial bids, calculate the overall consistency index of the bid documents, provide an interactive manual review interface for review experts, integrate automated analysis results with manual review opinions, generate a structured review report and support export.

2. The artificial intelligence-based bid document review assistance method according to claim 1, characterized in that: In step S2, the technical signature similarity analysis specifically includes the following steps: Step S21: Text segmentation. Extract the chapter structure of the technical bid text content in each bidder's bid documents, use regular expressions to segment the technical bid text content into logical paragraphs, and further divide it into sentences as analysis units. Step S22: Sentence segmentation. Segment each sentence into words and create a word sequence for the technical bid text content in each bidder's bid document. Step S23: Text vectorization representation. A multi-level text vectorization method based on the TF-IDF weighted bag-of-words model and pre-trained word vectors is adopted to transform the text content into vector form. Step S24: Document-level similarity calculation. Calculate the cosine similarity of the technical bid text content of any two bidders. The closer the value is to 1, the closer the two documents are in the vector space, that is, the more similar the technical bid text content is. Set a document-level similarity threshold. Technical bid text content that exceeds this threshold will be marked as similar bid pairs. Step S25: Generate a technical similarity analysis report.

3. The artificial intelligence-based bid document review assistance method according to claim 2, characterized in that: In step S23, the text vectorization representation specifically includes the following steps: Step S231: Vocabulary Construction. Summarize the word sequences from the technical bid texts of all bidders to construct a vocabulary, with the following dimensions: ; Step S232: Document TF-IDF Vector Generation. Treat the technical bid text of each bidder as a document. For each word in the document and vocabulary, calculate the TF-IDF weight value of that word in the document. Arrange the TF-IDF weight values ​​of all words in vocabulary order to form the document's TF-IDF vector, with dimension [missing value]. ; Step S233: Load the word vector model, using a pre-trained word vector model to map each word to... 3D word vectors; Step S234: Document semantic vector generation. Average the word vectors of all words in each sentence of the document to obtain the sentence vector. Then average the vectors of all sentences in the document to obtain the document semantic vector, with dimension [missing information]. ; Step S235: Document vector generation. The final document vector is generated through weighted concatenation, with dimensions [dimension number missing]. .

4. The artificial intelligence-based bid document review assistance method according to claim 1, characterized in that: In step S3, the commercial bid consistency analysis specifically includes the following steps: Step S31: Numerical data extraction. The commercial tender document is scanned using a numerical pattern recognition algorithm to identify and extract all numerical data. Each numerical data and its metadata together constitute a data item. All data items are normalized and converted into a unified unit of measurement to construct a set of numerical data items. Step S32: Data Feature Analysis. Perform feature analysis on the set of numerical data items and automatically generate analysis indicators. The specific steps are as follows: Step S321: Quantify the data distribution characteristics, calculate the global mean, standard deviation and coefficient of variation of the numerical data, and calculate the within-group statistical characteristics of the bidders in groups; Step S322: Numerical data classification. The K-means clustering algorithm is used to perform unsupervised classification of numerical data, identify and label data categories; Step S323: Adaptive generation of analysis indicators, assigning similarity calculation indicators to each data category, assigning weights to each data item based on the coefficient of variation, and constructing a set of analysis indicators; Step S33: Multi-dimensional analysis of numerical consistency. Based on the analysis indicators, construct numerical vectors for each bidder, calculate multi-dimensional similarity, and detect numerical pattern anomalies. Step S34: Machine learning risk assessment, the specific steps are as follows: Step S341: Feature engineering. Based on the multi-dimensional similarity calculation results and numerical pattern anomaly detection results, construct a multi-dimensional feature vector for each bidder pair and introduce historical labeled data to train a risk assessment model. Step S342: Risk probability calculation, calculate the risk probability of bid rigging for each bidder pair, and identify the main risk contribution characteristics; Step S343: Risk classification, automatically classifying bidders based on risk probability; Step S35: Generate a commercial bid consistency analysis report.

5. An AI-based bid document review assistance system, used to implement the AI-based bid document review assistance method as described in any one of claims 1-4, characterized in that: It includes a tender document preprocessing module, a technical bid similarity analysis module, a commercial bid consistency analysis module, and an analysis results summary module; The bid document preprocessing module collects the bid documents from each bidder and converts them into a unified format. It then preprocesses the bid documents and sends the preprocessed bid documents to the technical bid similarity analysis module and the commercial bid consistency analysis module. The technical bid similarity analysis module calculates the similarity of the technical bid text content in the bid documents, establishes a quantitative evaluation system based on the combination of TF-IDF statistical features and pre-trained word vector semantic features, generates a technical bid similarity analysis report, and sends the technical bid similarity analysis report to the analysis result summary module. The commercial bid consistency analysis module performs unified data analysis on the commercial bid price data in the bid documents, uses machine learning models to conduct risk assessment on bidders, generates a commercial bid consistency analysis report, and sends the commercial bid consistency analysis report to the analysis results summary module. The analysis results summary module statistically summarizes the analysis results of various dimensions of the technical and commercial bids, calculates the overall consistency index of the bid documents, provides an interactive manual review interface for review experts, integrates automated analysis results with manual review opinions, generates a structured review report, and supports export.

Images