WEB end intelligent bidding document structured processing system based on hybrid AI analysis engine

Abstract

The invention belongs to the technical field of intelligent document processing, and provides a WEB-end intelligent bidding document structured processing system based on a hybrid AI analysis engine, comprising: a multi-modal document analysis module extracts key information of a bidding document by using a fuzzy starvation game algorithm, and dynamically sorts core terms according to semantic association; the bidding document blind box analysis module safely disassembles the encrypted bidding document through a block chain technology to generate a structured review matrix; the cloud collaborative review module integrates the multi-dimensional data board, simulates bid evaluation by using a Monte Carlo algorithm, and generates a risk thermodynamic diagram; the risk early warning module identifies dispute points, provides compliance suggestions in combination with a knowledge base, and synchronizes the compliance suggestions to all review terminals; according to the method, automatic and structured processing of the WEB end bidding document is realized by fusing the mixed AI analysis engine, so that the efficiency and accuracy of bidding document review are improved, manual intervention is reduced, the review process is accelerated, and the risk of misjudgment and omission is reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent document processing, and more specifically, is a WEB-side intelligent bidding document structured processing system based on a hybrid AI parsing engine. Background Art

[0002] Bid processing is a core component of the current digital transformation of the bidding industry. However, the structured processing of intelligent bids on the web primarily relies on manual data entry and traditional information extraction methods. These methods often require manual analysis of bid content, which is susceptible to subjective judgment and leads to a high rate of misjudgment.

[0003] Furthermore, while some systems use rule engines or simple keyword matching techniques to parse bid documents, they are unable to process complex unstructured data (such as text in images or scanned documents). Their understanding of the semantics of clauses is weak, making it easy to miss or misjudge key information, resulting in a high rate of misjudgment. Furthermore, existing technologies lack data integrity verification mechanisms, exposing bid documents to the risk of tampering during transmission and processing, making it difficult to ensure the fairness of the review results.

[0004] Furthermore, in traditional bid evaluation processes, risk assessments are often based on expert experience and lack quantitative analysis methods, making it difficult to effectively identify potential risks such as bid rigging, collusion, and fraudulent bidding. For example, the determination of abnormal bid prices and identical technical solutions lacks scientific model support and is highly subjective. When disputes arise during evaluations, existing systems are unable to quickly link industry policies and historical precedents, requiring experts to manually search regulatory documents. This is not only time-consuming and labor-intensive, but also makes it difficult to ensure the compliance and consistency of recommendations, resulting in a low level of standardization in evaluations. Furthermore, existing technologies are unable to integrate historical bidding data with real-time market parameters, unable to provide dynamic and intelligent support for evaluation decisions, and unable to adapt to the rapidly changing business needs of the bidding market. Summary of the Invention

[0005] In order to solve the above technical problems, the present invention provides a WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine to solve the problems raised in the background technology.

[0006] The web-based intelligent bidding document structured processing system based on a hybrid AI parsing engine includes:

[0007] A multimodal document parsing module, which extracts key information from tender documents using a fuzzy Hunger Games search algorithm and dynamically prioritizes core clauses based on semantic relevance weights.

[0008] The blind box parsing module for bidding documents is used to securely disassemble encrypted bidding documents through blockchain technology and generate a structured review matrix;

[0009] The cloud-based collaborative review module integrates a multi-dimensional review data dashboard based on the web platform, and generates a dynamic risk heat map by simulating the bid evaluation process through the Monte Carlo algorithm. The multi-dimensional review data dashboard integrates bid document analysis results, historical bid data, and real-time market parameters.

[0010] The risk warning module is used to identify controversial points marked by review experts, call the knowledge base to generate compliance recommendations, and synchronize them to all review terminals in real time.

[0011] Preferably, the multimodal document parsing module further includes:

[0012] Format Adaptive Unit, used to recognize unstructured data in bids of various formats;

[0013] A key information extraction unit uses a fuzzy Hunger Game search algorithm to optimize the detection of rejected bid items. The fuzzy Hunger Game search algorithm determines the information extraction priority by calculating the cosine similarity between the semantic vector of the terms and the preset bid template;

[0014] The dynamic ranking unit establishes a multi-dimensional ranking matrix based on the weights of technical scoring items and the urgency of qualification requirements.

[0015] Preferably, the fuzzy Hunger Games search algorithm includes:

[0016] In the initialization phase, n candidate parsing paths are randomly generated, each containing a different order of clause extraction;

[0017] During the iteration phase, the following formula is used to calculate the path survival probability and eliminate inefficient paths:

[0018]

[0019] Where P is the path survival probability, ranging from 0 to 1. A larger P value indicates a higher survival probability, i.e., a more efficient path. α is a parameter that controls the steepness of the function. The semantic relevance is the semantic matching degree between the extracted clause and the preset bid template obtained by cosine similarity.

[0020] In the output stage, the optimal solution with information extraction completeness ≥ 95% in the surviving path is retained.

[0021] Preferably, the bidding document blind box parsing module further includes:

[0022] The blockchain encryption unit uses the SHA-256 hash algorithm to record the bid disassembly process;

[0023] Structured output unit, generating a structured review matrix including a qualification review form and a technical solution comparison chart;

[0024] The authenticity verification unit verifies the integrity of the document by comparing the digital fingerprint of the bidding document with the blockchain evidence data.

[0025] Preferably, the method for generating the structured review matrix includes:

[0026] S1. Mapping the tender document chapters to the preset XML Schema nodes;

[0027] S2. Locate the physical location of the technical clause in the document using XPath;

[0028] S3. Automatically generate a review navigation directory with hyperlinks based on the node weight values.

[0029] Preferably, the cloud collaborative review module further includes:

[0030] Data aggregation unit, used to integrate supplier performance evaluation indicators and real-time market parameters from historical bidding data;

[0031] The risk simulation unit uses a Monte Carlo algorithm to calculate the probability of bid rigging and collusion. The input parameters of the Monte Carlo algorithm include bid price dispersion and technical solution similarity coefficient;

[0032] Visualization unit generates multi-level dynamic risk heat maps that support drill-down analysis.

[0033] Preferably, the method for generating the dynamic risk heat map includes:

[0034] Establish a risk assessment model that includes a price sensitivity factor and a technical deviation factor. The price sensitivity factor is the degree to which the bid price deviates from the market average, and the technical deviation factor is the proportion of the proposal that does not meet the requirements of the bidding documents.

[0035] Through Markov Chain Monte Carlo sampling, a large number of bid evaluation scenarios are simulated to calculate the risk probability distribution of each bidder;

[0036] The bidder's risk level is intuitively displayed with a color gradient, and drill-down is supported to view specific risk factor details.

[0037] Preferably, the risk warning module further includes:

[0038] Annotation recognition unit, which uses NLP technology to extract controversial keywords from expert-annotated texts;

[0039] Knowledge graph query unit, which links mandatory provisions and historical precedents in the industry knowledge base;

[0040] The suggestion generation unit uses the Seq2Seq model to generate compliant correction suggestions.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] 1. This invention reduces manual input and format conversion time through automated parsing and structured processing, shortening the overall review cycle. Its dynamic sorting and visualization tools help experts focus on key terms, effectively improving decision-making efficiency.

[0043] 2. This invention ensures data integrity through multimodal analysis and blockchain evidence storage, effectively reducing the misjudgment rate, and provides quantitative risk assessment and compliance support through Monte Carlo algorithms and knowledge graphs, reducing human omissions.

[0044] 3. The present invention integrates historical data and real-time market parameters to provide dynamic risk prediction and assist scientific decision-making; it also automatically generates correction suggestions to improve review consistency and standardization. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a block diagram of the overall processing system of the present invention;

[0046] Figure 2 This is a flowchart of the multimodal document parsing process of the present invention;

[0047] Figure 3 This is a flowchart of the blind box parsing process of the bidding document of the present invention;

[0048] Figure 4 This is a flowchart of the cloud-based collaborative review process of the present invention;

[0049] Figure 5 This is a risk warning flow chart of the present invention. DETAILED DESCRIPTION

[0050] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0051] As attached Figure 1 As shown:

[0052] Embodiment: The present invention provides a WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine, comprising:

[0053] The multimodal document parsing module is used to extract key information from tender documents using a fuzzy Hunger Games search algorithm and dynamically prioritize core clauses based on semantic relevance weights. The multimodal document parsing module also includes:

[0054] The format adaptation unit is used to identify unstructured data in bid documents of various formats. By identifying unstructured data in multiple document formats, it is compatible with the bid document formats of different tenderers and improves the parsing and generalization capabilities.

[0055] The key information extraction unit uses a fuzzy Hunger Game search algorithm to optimize the detection of rejected bid items. The fuzzy Hunger Game search algorithm determines the information extraction priority by calculating the cosine similarity between the semantic vector of the terms and the preset bid template; the fuzzy Hunger Game search algorithm is used to optimize the detection of rejected bid items. The matching degree between the terms and the preset template is calculated through cosine similarity, and high-risk terms are extracted first to reduce the missed detection rate.

[0056] The dynamic ranking unit establishes a multi-dimensional ranking matrix based on the weights of technical scoring items and the urgency of qualification requirements; based on the multi-dimensional matrix of technical scoring weights, qualification urgency, etc., it prioritizes the core clauses to assist review experts in quickly identifying the key points and shortening the review time.

[0057] Specifically, the fuzzy Hunger Games search algorithm includes:

[0058] In the initialization phase, n candidate parsing paths are randomly generated, each containing a different order of clause extraction;

[0059] During the iteration phase, the following formula is used to calculate the path survival probability and eliminate inefficient paths:

[0060]

[0061] Where P is the path survival probability, ranging from 0 to 1. A larger P value indicates a higher survival probability, i.e., a more efficient path. α is a parameter that controls the steepness of the function. The semantic relevance is the semantic matching degree between the extracted clause and the preset bid template obtained by cosine similarity.

[0062] In the output stage, the optimal solution with information extraction completeness ≥ 95% in the surviving path is retained.

[0063] Specifically, the method for the dynamic sorting unit to establish a multi-dimensional sorting matrix includes:

[0064] Establish the evaluation dimension set D = {d1, d2, ..., d n}, where d1 is the weight of the technical scoring item, d2 is the urgency of the qualification requirements, and d3 is the timeliness of the clause; normalize each dimension to obtain

[0065] Use the following formula to construct the priority decision function:

[0066]

[0067] Among them, α+β+γ=1, and α>β>γ, which is used to reflect the dominance of the technical scoring items;

[0068] Thus, the sorting matrix M is generated:

[0069]

[0070] Among them, the row vector represents the mth bidding document, and the column vector represents the nth evaluation dimension.

[0071] The bid document blind box parsing module is used to securely disassemble encrypted bid documents using blockchain technology and generate a structured review matrix. The bid document blind box parsing module also includes:

[0072] The blockchain encryption unit uses the SHA-256 hash algorithm to record the bid disassembly process, ensuring traceability, preventing tampering, and improving the fairness of the review;

[0073] The structured output unit generates a structured review matrix containing a qualification review form and a technical solution comparison chart. This interactive review matrix converts unstructured data into a visual table, effectively improving review efficiency.

[0074] The authenticity verification unit verifies the integrity of the bidding document by comparing the digital fingerprint of the bidding document with the blockchain evidence data; by comparing the digital fingerprint with the blockchain evidence, the integrity of the document is verified in real time, false bidding is eliminated, and the error rate is reduced.

[0075] Specifically, the method for generating a structured review matrix includes:

[0076] S1. Mapping the tender document chapters to the preset XML Schema nodes;

[0077] S2. Locate the physical location of the technical clause in the document using XPath; that is, locate the document node corresponding to the technical clause using XPath and, in conjunction with the document rendering engine, obtain the physical coordinates of the node in the original document, where the physical coordinates include the page number, paragraph number, and a two-dimensional offset relative to the upper left corner of the page;

[0078] S3. Automatically generate a review navigation directory with hyperlinks based on the node weight values.

[0079] The cloud-based collaborative review module integrates a multi-dimensional review data dashboard based on the web platform, and uses the Monte Carlo algorithm to simulate the bid evaluation process to generate a dynamic risk heat map. The multi-dimensional review data dashboard integrates bid document analysis results, historical bid data, and real-time market parameters. The cloud-based collaborative review module also includes:

[0080] The data aggregation unit is used to integrate supplier performance evaluation indicators and real-time market parameters in historical bidding data; by integrating historical performance data and market parameters, it provides a more comprehensive evaluation basis.

[0081] The risk simulation unit uses the Monte Carlo algorithm to calculate the probability of bid rigging and collusion. The input parameters of the Monte Carlo algorithm include the dispersion of bid prices and the coefficient of similarity of technical solutions.

[0082] Specifically, the steps of executing the Monte Carlo algorithm by the risk simulation unit include:

[0083] The bid price dispersion A is set as the ratio of the bid standard deviation to the mean, and is expressed using the following formula:

[0084]

[0085] Among them, B is the quotation, σ B is the quotation standard deviation, μ B is the average of the quotations;

[0086] The similarity coefficient S of the technical solution is calculated through the text similarity matrix:

[0087]

[0088] Where n is the number of bids participating in the comparison, t i ,t j are the technical proposal texts of the i-th and j-th bids, respectively, v t is the BERT embedding vector of the technical solution, and Sim(·) is the cosine similarity. Using the Monte Carlo algorithm, we input bid dispersion and the coefficient of technical solution similarity to simulate the probability of bid rigging and collusion, identifying potential risks (such as abnormally low bids) in advance, effectively improving risk identification accuracy.

[0089] The visualization unit generates a multi-level dynamic risk heat map that supports drill-down analysis. By generating a multi-level risk heat map and supporting drill-down analysis, it helps experts quickly identify high-risk bidders.

[0090] Specifically, the method for generating a dynamic risk heat map includes:

[0091] Establish a risk assessment model that includes price sensitivity factors and technical deviation factors. The price sensitivity factor is the degree to which the bid price deviates from the market average, and the technical deviation factor is the proportion of the proposal that does not meet the requirements of the bidding documents.

[0092] Through Markov Chain Monte Carlo sampling, a large number of bid evaluation scenarios are simulated to calculate the risk probability distribution of each bidder;

[0093] The bidder's risk level is intuitively displayed with a color gradient, and drill-down is supported to view specific risk factor details.

[0094] The risk warning module is used to identify controversial points marked by review experts, call the knowledge base to generate compliance recommendations, and synchronize them to all review terminals in real time. The risk warning module also includes:

[0095] The annotation recognition unit uses NLP technology to extract dispute keywords from expert-annotated texts and automatically classify the dispute types, improving processing efficiency.

[0096] The knowledge graph query unit links industry knowledge bases and historical precedents to quickly locate the corresponding provisions of the disputed points, reducing the time for manual searches by experts;

[0097] The suggestion generation unit uses the Seq2Seq model to generate compliant correction suggestions, improve the standardized response rate, and reduce the subjectivity of the review.

[0098] From the above, we can see that by combining the fuzzy Hunger Game search algorithm with semantic relevance analysis, we can achieve intelligent extraction and dynamic sorting of key information, breaking through the rigid matching limitations of traditional rule engines; and combining blockchain evidence with encrypted bid document disassembly to ensure the transparency and traceability of the review process, solving the pain points of data security and operation audit in the existing system; at the same time, through the Monte Carlo algorithm, the bid evaluation process is quantitatively simulated to generate dynamic risk visualization results, replacing traditional empirical risk assessment, and improving the scientificity and intuitiveness of risk prediction. Based on the NLP and Seq2Seq models, a closed loop for review dispute resolution is constructed, which automatically associates the knowledge base with precedents to generate suggestions, realizing the full process of "discovering problems-locating basis-outputting solutions" intelligence, which is different from the traditional manual retrieval mode.

[0099] It is important to note that the construction and arrangement of the present application shown in a plurality of different exemplary embodiments are merely illustrative. Although only a few embodiments are described in detail in this disclosure, it will be readily understood by those who consult this disclosure that many modifications are possible without departing substantially from the novel teachings and advantages of the subject matter described in this application. Other replacements, modifications, changes, and omissions may be made in the design, operating conditions, and arrangement of the exemplary embodiments without departing from the scope of the present invention. Therefore, the present invention is not limited to specific embodiments, but extends to a variety of modifications still falling within the scope of the appended claims.

[0100] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment (i.e., those features that are not relevant to the best mode presently contemplated for carrying out the invention or those that are not relevant to implementing the invention) may not be described.

[0101] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A WEB-side intelligent bidding document structured processing system based on a hybrid AI parsing engine, characterized by: include: A multimodal document parsing module, which extracts key information from tender documents using a fuzzy Hunger Games search algorithm and dynamically prioritizes core clauses based on semantic relevance weights. The blind box parsing module for bidding documents is used to securely disassemble encrypted bidding documents through blockchain technology and generate a structured review matrix; The cloud-based collaborative review module integrates a multi-dimensional review data dashboard based on the WEB platform and generates a dynamic risk heat map by simulating the bid evaluation process through the Monte Carlo algorithm; The multi-dimensional review data dashboard integrates bid document analysis results, historical bid data and real-time market parameters; The risk warning module is used to identify controversial points marked by review experts, call the knowledge base to generate compliance recommendations, and synchronize them to all review terminals in real time.

2. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 1 is characterized in that: The multimodal document parsing module also includes: Format Adaptive Unit, used to recognize unstructured data in bids of various formats; A key information extraction unit uses a fuzzy Hunger Game search algorithm to optimize the detection of rejected bid items. The fuzzy Hunger Game search algorithm determines the information extraction priority by calculating the cosine similarity between the semantic vector of the terms and the preset bid template; The dynamic ranking unit establishes a multi-dimensional ranking matrix based on the weights of technical scoring items and the urgency of qualification requirements.

3. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 2 is characterized in that: The fuzzy Hunger Games search algorithm includes: In the initialization phase, n candidate parsing paths are randomly generated, each containing a different order of clause extraction; During the iteration phase, the following formula is used to calculate the path survival probability and eliminate inefficient paths: Where P is the path survival probability, ranging from 0 to 1. A larger P value indicates a higher survival probability, i.e., a more efficient path. α is a parameter that controls the steepness of the function. The semantic relevance is the semantic matching degree between the extracted clause and the preset bid template obtained by cosine similarity. In the output stage, the optimal solution with information extraction completeness ≥ 95% in the surviving path is retained.

4. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 1 is characterized in that: The bidding document blind box parsing module also includes: The blockchain encryption unit uses the SHA-256 hash algorithm to record the bid disassembly process; Structured output unit, generating a structured review matrix including a qualification review form and a technical solution comparison chart; The authenticity verification unit verifies the integrity of the document by comparing the digital fingerprint of the bidding document with the blockchain evidence data.

5. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 4 is characterized in that: The method for generating the structured review matrix includes: S1. Mapping the tender document chapters to the preset XML Schema nodes; S2. Locate the physical location of the technical clause in the document using XPath; S3. Automatically generate a review navigation directory with hyperlinks based on the node weight values.

6. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 1 is characterized in that: The cloud collaborative review module also includes: Data aggregation unit, used to integrate supplier performance evaluation indicators and real-time market parameters from historical bidding data; The risk simulation unit uses a Monte Carlo algorithm to calculate the probability of bid rigging and collusion. The input parameters of the Monte Carlo algorithm include bid price dispersion and technical solution similarity coefficient; Visualization unit generates multi-level dynamic risk heat maps that support drill-down analysis.

7. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 6 is characterized in that: The method for generating the dynamic risk heat map includes: Establish a risk assessment model that includes a price sensitivity factor and a technical deviation factor. The price sensitivity factor is the degree to which the bid price deviates from the market average, and the technical deviation factor is the proportion of the proposal that does not meet the requirements of the bidding documents. Through Markov Chain Monte Carlo sampling, a large number of bid evaluation scenarios are simulated to calculate the risk probability distribution of each bidder; The bidder's risk level is intuitively displayed with a color gradient, and drill-down is supported to view specific risk factor details.

8. The WEB-side intelligent bidding document structuring processing system based on a hybrid AI parsing engine as claimed in claim 1, characterized in that: The risk warning module also includes: Annotation recognition unit, which uses NLP technology to extract controversial keywords from expert-annotated texts; Knowledge graph query unit, which links mandatory provisions and historical precedents in the industry knowledge base; The suggestion generation unit uses the Seq2Seq model to generate compliant correction suggestions.

Images