Artificial intelligence-based method and system for fair review of bidding documents

Abstract

This invention relates to the field of artificial intelligence review technology, and discloses a method and system for fair review of bidding documents based on artificial intelligence. The system includes a document acquisition unit, a scoring model unit, and a fair review optimization unit. The document acquisition unit acquires bidding document information and corresponding scoring rules input by the user. The bidding document information includes document type information and scoring weight information, and the scoring rules include commercial scoring rules and technical scoring rules. The scoring model unit determines the scoring procedure based on the document type information. The fair review optimization unit determines the scoring optimization procedure based on the document type information, scoring rules information, and scoring procedure, and completes the fair review through this procedure. This system can solve the problems of scoring bias, procedural confusion, and low efficiency in traditional manual review. Through multi-unit collaboration, it achieves automation and standardization of review, maintains a fair competitive environment for bidding, and meets the industry's demand for efficient and fair review.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence review technology, specifically to a method and system for fair review of bidding documents based on artificial intelligence. Background Technology

[0002] In current bidding and tendering activities, document review is a crucial step in ensuring the standardized progress of the entire process, and its fairness directly affects the credibility of the bidding and tendering results and the rationality of market resource allocation. As the number of bidding and tendering projects continues to increase and the scope of the fields involved becomes wider, the complexity of bidding and tendering documents is also constantly rising, and the traditional model of relying on manual document review is gradually revealing many problems.

[0003] In traditional review processes, reviewers need to meticulously check and score a large number of bidding documents one by one according to the scoring criteria for different projects. Because different reviewers may have varying interpretations of the scoring criteria, and because individual subjective judgment is easily influenced by factors such as experience and emotions, the same batch of documents may receive different scores from different reviewers, making it difficult to ensure consistency in the review process. Furthermore, bidding documents contain information from multiple dimensions, including business and technical aspects. Different types of documents correspond to different scoring focuses and procedures. Manual review makes it difficult to accurately grasp the differences in scoring standards for various types of documents, easily leading to improper allocation of scoring weights and confusion in scoring procedures, further affecting the fairness of the review.

[0004] Manual review is inefficient. Faced with a massive volume of bidding documents, it often consumes significant time and manpower, potentially extending the review cycle and impacting the overall progress of bidding projects. Furthermore, fatigue errors due to prolonged work can further reduce review quality. While some sectors have attempted to introduce simple IT tools to assist in the review process, these tools mostly only store documents and extract basic information. They cannot automatically determine scoring procedures based on document type, nor can they dynamically optimize scoring according to detailed scoring rules. Therefore, they fail to fundamentally address the issues of fairness and efficiency in the review process and cannot meet the current demands for efficient and fair review in bidding activities. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for fair review of bidding documents based on artificial intelligence, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an artificial intelligence-based fair review system for bidding documents, the system comprising:

[0007] Document acquisition unit, scoring model unit, and fair review optimization unit;

[0008] The document acquisition unit is used to acquire the bidding document information input by the user and to acquire the scoring rules information corresponding to the bidding document; wherein, the bidding document information includes document type information and scoring weight information, and the scoring rules information includes commercial scoring rules and technical scoring rules.

[0009] The scoring model unit is used to determine the scoring procedure based on file type information;

[0010] The fair review optimization unit is used to determine the scoring optimization procedure based on document type information, scoring details information, and scoring procedure, and to complete the fair review through the scoring optimization procedure.

[0011] Preferably, the system further includes a review duration acquisition unit and an expert feedback acquisition unit; the review duration acquisition unit is used to acquire the review duration corresponding to the scoring details information, and the review duration is the time span for collecting the scoring details information; when the document type information is material bidding information, equipment bidding information, or subcontracting bidding information, and when the review duration is less than the preset review duration, the expert feedback acquisition unit is used to acquire the business score and technical score from expert feedback; through the first-level optimization model unit, a first-level optimization procedure for scoring is determined based on the business score, technical score, and scoring procedure; when the review duration is greater than or equal to the preset review duration, through the fair review optimization unit, a second-level optimization procedure for scoring is determined based on the document type information, scoring details information, and the first-level optimization procedure for scoring, and a fair review is completed through the second-level optimization procedure for scoring.

[0012] Preferably, the system further includes a parameter acquisition unit; the parameter acquisition unit is used to acquire the adjustment rate of the first-level scoring optimization program within the business scoring range, and to acquire the key scoring time period of the first-level scoring optimization program in the business scoring and technical scoring; wherein, the key scoring time period is the time period with the highest scoring weight, and different file types correspond to key scoring time periods of different durations; through the fair review optimization unit, based on the file type information, scoring rules information, and the key scoring time period of the first-level scoring optimization program, the key scoring optimization time period in the second-level scoring optimization program is determined, limiting the adjustment rate adjustment range of the adjustment rate to be less than the preset adjustment rate adjustment range, and limiting the duration adjustment range of other scoring time periods of the first-level scoring optimization program to be less than the preset duration adjustment range.

[0013] Preferably, the system further includes an enterprise qualification acquisition unit and a constraint margin identification unit; the enterprise qualification acquisition unit is used to acquire the user's enterprise qualification information, which includes registered capital, industry experience and historical performance information; through the constraint margin identification unit, based on the scoring rules information and the enterprise qualification information, the preset adjustment rate adjustment range and the preset duration adjustment range are determined.

[0014] Preferably, the system further includes a review database unit; when the file type information is sensitive project bidding information, and when the review time is less than the preset review time, the review database unit obtains the scoring base optimization program corresponding to the sensitive project bidding information, and the scoring base optimization program is used to meet the basic review requirements of the sensitive project corresponding to the sensitive project bidding information; when the review time is greater than or equal to the preset review time, the fair review optimization unit determines the scoring secondary optimization program based on the sensitive project bidding information, scoring details information, and scoring base optimization program, and completes the fair review through the scoring secondary optimization program.

[0015] Preferably, the system further includes a parameter weight acquisition unit; the parameter weight acquisition unit is used to acquire the associated and unassociated parameters of the scoring secondary optimization procedure, and to acquire the adjustment weights corresponding to the associated and unassociated parameters respectively; wherein, the associated parameters include business weights, technical weights, and key scoring optimization time periods, and the unassociated parameters include control parameters other than the associated parameters in the scoring secondary optimization procedure; through the fair review optimization unit, based on sensitive project bidding information, scoring rules information, scoring basic optimization procedure, and the adjustment weights corresponding to the associated and unassociated parameters respectively, the system determines the associated parameter adjustment values ​​corresponding to the associated parameters and the unassociated parameter adjustment values ​​corresponding to the unassociated parameters in the scoring secondary optimization procedure.

[0016] Preferably, the system further includes a sensitivity level acquisition unit; the sensitivity level acquisition unit is used to acquire the project sensitivity level corresponding to the bidding information of sensitive projects; wherein, the project sensitivity level includes the sensitivity levels corresponding to high-sensitivity projects, medium-sensitivity projects and low-sensitivity projects respectively; and based on the project sensitivity level, the adjustment weights corresponding to the associated parameters and non-associated parameters are determined respectively.

[0017] Preferably, the system further includes a scoring and evaluation unit; when the document type information is ordinary project bidding information, the scoring and evaluation unit determines the score reduction score according to the scoring details; when the score reduction score is greater than or equal to the preset score reduction score, and when the review time is less than the preset review time, the review database unit obtains the scoring base optimization procedure corresponding to the preset score reduction score; when the review time is greater than or equal to the preset review time, the fair review optimization unit determines the scoring secondary optimization procedure according to the document type information, scoring details, and scoring base optimization procedure, and completes the fair review through the scoring secondary optimization procedure.

[0018] Preferably, the system further includes a degradation level determination unit; the degradation level determination unit is used to determine the degradation level corresponding to the degradation score; and to obtain different rating base optimization programs corresponding to the degradation level by reviewing the database unit, wherein the parameter thresholds of the different rating base optimization programs are different.

[0019] Preferably, the present invention also includes an artificial intelligence-based method for fair review of bidding documents, the method comprising all the modules and method flow of the artificial intelligence-based bidding document fair review system described above.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] By leveraging the synergy of the document acquisition unit, the scoring model unit, and the fair review optimization unit, various problems in the traditional bidding document review process are effectively improved. The document acquisition unit can comprehensively acquire the bidding document information input by the user and the corresponding scoring details. The bidding document information includes document type information and scoring weight information, while the scoring details include commercial scoring details and technical scoring details. This design allows the system to grasp complete review criteria at the initial stage of the review, avoiding review deviations caused by missing information and ensuring that the review process always revolves around clear standards.

[0022] The scoring model unit determines the scoring procedure based on document type information, enabling it to develop suitable scoring processes for different types of bidding documents. This solves the problem of scoring procedure confusion caused by document type discrepancies in traditional manual review. Different types of bidding documents differ significantly in terms of business requirements and technical specifications, and the corresponding scoring logic and steps should also differ. The system automatically matches the scoring procedure using artificial intelligence technology, accurately distinguishing the scoring focus of various documents without human intervention. This reduces the impact of human factors on the selection of scoring procedures, making the review process for different types of documents more targeted and standardized.

[0023] The Fair Review Optimization Unit combines document type information, scoring details, and scoring procedures to determine the optimized scoring process and complete the fair review, further enhancing the fairness and reasonableness of the review. This unit can comprehensively consider the scoring focus corresponding to the document type, the specific requirements of the business and technical scoring details, and the established scoring procedures to dynamically optimize the scoring process, avoiding imbalances in scoring weight allocation and inconsistent application of scoring standards. For example, when dealing with bidding documents with high technical requirements, the system can appropriately strengthen the scoring weight of technical indicators in the optimized scoring process based on the specific clauses of the technical scoring details, while strictly adhering to the corresponding scoring procedures to ensure more accurate scoring in the technical dimensions. For documents with prominent business attributes, the system can focus on optimizing the scoring logic based on the business scoring details to ensure the reasonableness of the business portion of the scoring.

[0024] The entire system leverages artificial intelligence (AI) technology to automate review and optimization, significantly improving review efficiency. Compared to traditional manual review, which requires a significant amount of time to meticulously verify documents and determine scoring procedures, the system can quickly process massive amounts of bidding documents, automatically extracting information, matching procedures, and optimizing scores. This significantly shortens the review cycle and reduces labor costs. Simultaneously, the application of AI technology avoids subjective biases and fatigue errors that may occur in manual review, making the review results more objective and consistent. Different batches and types of bidding documents can be reviewed under unified standards and processes, effectively maintaining a fair competitive environment for bidding activities, enhancing the credibility of bidding results, helping to regulate market order, and promoting the bidding industry towards a more efficient and equitable direction. Attached Figure Description

[0025] Figure 1 This is a timing diagram of the artificial intelligence-based fair review system for bidding documents described in this invention.

[0026] Figure 2 A flowchart optimized for key scoring time periods. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] Please see Figure 1 The present invention provides a fair review system for bidding documents based on artificial intelligence. The system includes: a document acquisition unit, a scoring model unit, and a fair review optimization unit.

[0029] The document acquisition unit acquires user-input bidding document information and corresponding scoring rules. The bidding document information includes document type and scoring weight information. The scoring rules include commercial and technical scoring rules. The document type information may cover bidding types such as materials, equipment, subcontracting, or general projects. The scoring weight information defines the proportion of commercial and technical aspects in the overall score. The scoring rules further refine the specific scoring rules and standards for commercial and technical aspects. The scoring model unit determines the applicable scoring procedure based on the document type information. The scoring procedure is an AI-based algorithm module that can parse the document type and call the corresponding scoring logic. For example, for materials bidding, the scoring procedure may focus on supplier qualifications and delivery cycle, while for technical bidding, it may focus more on innovation and the feasibility of the technical solution. The fair review optimization unit generates a scoring optimization procedure based on the document type information, scoring rules, and the determined scoring procedure. This optimization procedure ensures the fairness of the review process by dynamically adjusting scoring weights and rule parameters, such as using machine learning models to detect potential biases or optimize scoring thresholds. Ultimately, it completes the automated fair review of the bidding documents and outputs the review results.

[0030] Example 1: During the operation of the review duration acquisition unit, the system first automatically calculates the time span from the generation of the scoring details information to the current moment through the timestamp comparison function. This time span is defined as the review duration. Its calculation logic is based on the metadata extraction function of the file management system. The system scans the creation time, last modification time, and version iteration records of the scoring details document, and obtains the time interval value accurate to the minute through the time difference algorithm. The preset review duration is not a fixed value, but a threshold parameter dynamically matched in the background configuration library according to the file type information. For example, the preset duration for material bidding is 72 hours, for equipment bidding it is 120 hours, and for subcontracting bidding it is 96 hours. These thresholds are derived from the big data analysis results of historical project review cycles. When the system detects that the file type belongs to material, equipment, or subcontracting bidding, and the actual review duration is less than the corresponding preset value, the expert feedback acquisition unit immediately activates the expert review interface module. This module obtains professional review data through two methods: one is to connect to the enterprise's internal expert database system, automatically push the documents to be reviewed and receive structured scoring forms; the other is to access the historical review database to match expert scoring records of similar projects. The acquired business and technical scores need to be processed by the data cleaning module. The system will remove outliers that exceed the scoring criteria and trigger a secondary verification process for data with a dispersion of more than 15%.

[0031] After receiving expert scoring data, the first-level optimization model unit activates the multi-source scoring fusion engine, which comprises three core processing layers: The first layer performs a difference analysis between the scoring program and expert scores, employing a dynamic weight allocation algorithm to calculate the confidence weight of each scoring item. For example, when the deviation between the expert score and the AI ​​score is less than 5%, it is assigned 80% weight; when the deviation is 5%-10%, the weight is reduced to 50%. The second layer constructs a scoring mapping matrix, establishing a correlation between the 32 scoring points of the business scoring rules and the 47 scoring points of the technical scoring rules, eliminating differences in scoring dimensions through matrix transformation. The third layer implements rule reconstruction, automatically generating supplementary scoring rules and embedding them into the original scoring program for special cases marked as "not covered by the scoring rules" in expert feedback. The first-level optimization program generated after these three layers of processing retains 60% of the basic logic of the original scoring program, integrates 30% of the expert scoring elements, and adds 10% of rule extension content, forming a primary optimization model with dynamic adaptability.

[0032] When the review period reaches or exceeds a preset threshold, the fair review optimization unit initiates a secondary optimization process, which includes two core modules: a time decay compensation mechanism and a rule update mechanism. The time decay compensation module analyzes the specific number of days the review period exceeds the preset value and corrects time-sensitive parameters in the primary optimization scoring process using a daily decay coefficient of 0.5%. For example, the timeliness weight of the supplier delivery cycle scoring item decreases linearly with the number of overdue days. The rule update module accesses the latest regulatory database in real time, automatically comparing the updates of the 23 types of industry standards applicable to the project under review. When a revision is detected in the regulatory clauses upon which the scoring rules are based, the system automatically converts the new clauses into scoring rules and overwrites the old rules. During the execution of the secondary optimization process, the system employs a dual-channel verification mechanism: the main channel runs the optimized scoring process, while the secondary channel simultaneously runs the original scoring process as a reference benchmark. When the deviation between the two results exceeds 8%, a manual review process is automatically triggered. The final output of the secondary optimization scoring process generates an execution report containing 257 quality monitoring points, detailing the optimization path and parameter adjustment trajectory for each scoring item.

[0033] The system for processing expert feedback data has a strict quality control system. All expert scores must pass a qualification verification process. The system verifies the match between the expert's identity number and the validity period of their authorization certificate, and only accepts expert data with a matching degree of 85% or higher in the review field. Historical review data must meet three filtering conditions: the project similarity score exceeds 75 points, the review time is within the three-year validity period, and the review results have been audited and filed. A progressive fusion strategy is adopted in the data integration phase. The first round of fusion retains the original scoring procedure framework, the second round injects key correction items from the expert scores, and the third round supplements high-frequency feature values ​​from historical data. After each round of fusion, variance analysis is performed to ensure data stability.

[0034] The system employs a four-level response plan for handling anomalies: when expert scoring times out, the system automatically switches to historical data mode; when logical conflicts arise during scoring fusion, a rule priority arbitration procedure is initiated; when time decay compensation causes key parameters to exceed safety thresholds, a parameter boundary protection mechanism is triggered; and when the final review result continues to deviate from the reference benchmark, the system freezes the output and generates a red alert report. All intermediate data generated throughout the implementation process is stored in a blockchain-based evidence storage system, forming an immutable record chain containing timestamps, operator digital signatures, and data hash values.

[0035] Example 2: See Figure 2 When the parameter acquisition unit starts, it activates the multi-dimensional log analysis system. This system continuously tracks the parameter change records of the first-level optimization program within the business scoring range. It identifies the adjustment rate characteristics through time series analysis algorithms. The specific calculation of the adjustment rate is based on the difference in weight changes of the same scoring item within three consecutive review cycles. The system divides the change value by the time interval to obtain the hourly adjustment amount and marks it as the baseline rate. For the identification of key scoring time periods, the weight peak positioning technology is used. The system scans the time distribution heatmap of all scoring points in the scoring rules and automatically marks the time intervals where the weight ratio exceeds 15% of the total score and the continuous duration is greater than 20% of the total review time. The key duration configurations for different document types are stored in the type feature library. For example, the key time period for equipment bidding is the 45-minute review window corresponding to the technical solution demonstration session by default, while for subcontracting bidding, it focuses on the 60-minute core stage of the construction organization design defense. After receiving the key scoring time period data, the fair review optimization unit activates the time period optimization engine. This engine performs three levels of processing: The first level establishes the dynamic boundary of the key scoring optimization time period, matching a preset time period expansion coefficient based on document type information. For example, material procurement allows an expansion of 15% of the original time period, while engineering allows an expansion of 25%. The second level implements time period weight enhancement, applying a weighting factor of 1.2-1.5 times to the original scoring weight within the optimization time period. The weighting value is derived by back-calculating from the historical best review effect model. The third level constructs time period association rules, establishing a cross-time period correlation between the innovation indicators in the technical scoring details and the pricing strategy in the business scoring. A rate change monitoring device is introduced in the rate adjustment control stage. This component calculates the deviation between the current adjustment rate and the historical average in real time. When the rate change is detected to be close to the preset adjustment rate, a damping adjustment mechanism is automatically triggered, smoothing parameter fluctuations by inserting a 0.1-second data processing delay. For non-key scoring time periods, a time period compression protection program is activated. This program strictly limits the duration adjustment of non-core time periods to no more than 8% of the original duration. The compressed time periods use a scoring density enhancement algorithm to maintain review integrity.

[0036] The enterprise qualification acquisition unit connects to the industrial and commercial credit database during operation, and retrieves the registered capital verification report of bidding enterprises in real time through API interface. The registered capital information needs to be cross-verified by bank credit certificates. Industry experience data collection covers the bidding records of similar projects in the past five years. The system automatically filters performance with contract amount less than 30% of the current project budget. Historical performance evaluation uses an NLP text analysis engine to parse past owner evaluation reports, extract keywords to generate KPI performance scores. All qualification data are processed by a standardized conversion module. Registered capital is segmented and coded in the tens of millions range, industry experience is graded and quantified by the number of projects, and historical performance is converted into a percentage-based integrity index. The core of the constraint margin identification unit is the dynamic margin calculation model. This model performs a matrix comparison between enterprise qualification information and scoring rules requirements: the registered capital matching degree calculation uses a logarithmic function transformation. The registered capital of tens of millions corresponds to a basic margin coefficient of 1.0, and the coefficient increases by 0.2 for each order of magnitude increase; the industry experience matching degree is calculated linearly according to the number of projects, and the margin coefficient increases by 0.1 for every three additional similar projects; the historical performance matching degree is implemented in a segmented progressive manner, and the margin coefficient increases by 0.15 for every 5 points increase in the integrity index above 80. The final generated constraint margin value is converted into a preset adjustment rate adjustment range and a preset duration adjustment range through a weighted formula, where registered capital accounts for 40%, industry experience accounts for 35%, and historical performance accounts for 25%. The system is equipped with a margin threshold alarm mechanism, which automatically switches to conservative mode when the calculated adjustment range exceeds the safety boundary.

[0037] A progressive optimization strategy is adopted during the parameter adjustment execution phase. The system decomposes the entire optimization process into eight sequentially executed micro-batches, with each batch allowing adjustments to only 1 / 8 of the total parameters. Optimization operations during critical scoring periods are scheduled for the third and fifth batches, while adjustments to the rate of change are scheduled for the second and sixth batches. After each batch, parameter stability testing is initiated, with indicators including the standard deviation of the scoring distribution and the rate of change of the range. All parameter adjustment records are managed through a version tree, and the system retains complete parameter snapshots of the most recent five versions, supporting rollback to historical states at any time. A dedicated handling process is in place for abnormal enterprise qualifications: when registered capital verification fails, the system automatically switches to a security deposit verification mode; when industry experience data is missing, a related project substitution mechanism is activated; and when historical performance evaluation encounters obstacles, the industry average integrity value is temporarily used. The qualification data verification process includes triple verification: verification of business registration information, comparison of tax payment certificates, and scanning of related enterprises of the actual controller. Failure to pass any one of these verifications will trigger a yellow warning signal. Dynamic calibration is implemented during the constraint margin calculation process. The system automatically performs a margin model correction every three bidding projects: it collects the deviation between the actual review effect data and the predicted margin, and optimizes the matching coefficient using the least squares method; when the goodness of fit of three consecutive corrections is lower than 0.7, the model reconstruction process is triggered. The final output constraint parameters are stored in encrypted form using digital signatures, forming a tamper-proof record chain containing timestamps, operator IDs, and version hash values.

[0038] Taking a bidding project for a digital X-ray machine system purchased by a top-tier hospital as an example, after the system identifies the file type information as equipment bidding information, it initiates the process of Example 2. The review time acquisition unit retrieves the scoring details of the project, which was generated at 15:00 on November 5, 2023, and the current review time is 10:00 on November 8, calculating a review time of 64 hours. The preset review time module matches the equipment bidding threshold of 72 hours, and the system determines that the actual time is less than the preset value. The expert feedback acquisition unit then sends a data request to the medical equipment review expert database. Five experts submit commercial and technical scoring data on independent terminals: the commercial scoring focuses on the payment cycle (maximum score 20 points) and maintenance terms (maximum score 15 points), while the technical scoring involves core indicators such as imaging resolution (maximum score 25 points) and exposure accuracy (maximum score 20 points).

[0039] The parameter acquisition unit begins analyzing the operational characteristics of the first-level optimization program for scoring. The system detects an abnormal adjustment rate for the payment cycle parameter within the business scoring range. This parameter increased from a weighting coefficient of 1.2 to 1.8 over three consecutive review cycles, resulting in an adjustment rate of 0.025 per hour. The key scoring time period identification module scans the technical scoring details and finds that the equipment stability test accounts for 32% of the total technical score and 25% of the total review time, meeting the definition of a key time period. Based on the characteristics of equipment-type tenders, the system sets this time period as a 55-minute core window. The fair review optimization unit activates the time period optimization engine to process the key time period data. The system matches the upper limit of the time period expansion coefficient for equipment-type tenders by 20%, extending the original 55-minute technical demonstration time period to 66 minutes. Within the extended time period, a 1.4-fold weighting factor is applied to the imaging resolution index. This coefficient is derived from the back-calculation model of the best review results of similar historical projects. The adjustment rate control module detected that the payment cycle parameter adjustment rate deviated by 38% from the historical average, approaching the critical value of a preset adjustment rate adjustment range of 40%. The system automatically inserted a data processing delay and activated the damping adjustment mechanism to stabilize the adjustment rate within a reasonable range of 0.022 per hour. The enterprise qualification acquisition unit retrieved bidding enterprise data through the business registration information verification interface: the registered capital of 5 million yuan was verified by bank credit certification; the industry experience database showed that the company had won two similar medical equipment procurement bids in the past three years, but the single contract amount was less than 40% of the current budget; the historical performance analysis module analyzed past user reviews to generate a 78-point integrity index. The constraint margin identification unit performed matrix calculations: the registered capital matching degree was logarithmically transformed to obtain a base coefficient of 1.0; the industry experience matching degree was calculated to obtain an incremental coefficient of 0.08; and the historical performance matching degree generated an incremental coefficient of 0.12. After weighted fusion, a constraint margin value of 1.2 was generated. The system mapped the constraint margin value to the parameter control rules: the preset adjustment rate adjustment range was set to ±35%, and the preset duration adjustment range was limited to ±10%. During the optimization of critical scoring time periods, the technical demonstration period was extended from 55 minutes to 66 minutes (an increase of 20%). When this limit was exceeded, a protection mechanism was triggered, automatically reducing the time to 63 minutes (an actual increase of 14.5%). The adjustment rate of the payment cycle parameter was stabilized at 0.022 per hour after damping adjustment, with the change controlled within a safe range of 33%. The original duration of the business negotiation session during non-critical periods was 40 minutes, and the system compression adjustment reduced it by only 3 minutes (7.5%), which meets the preset protection threshold.

[0040] The qualification anomaly handling module detected a gap in industry experience data and automatically activated the associated project substitution mechanism: it retrieved similar-scale bidding records of the company in other provinces to supplement the experience of two in vitro diagnostic equipment procurement projects. Parameter adjustments were implemented using an eight-batch progressive strategy: the first batch optimized imaging resolution weights, the second batch adjusted payment cycle parameters, and the fourth batch optimized key time periods; stability testing was performed at intervals between each batch, and the standard deviation of the technical score distribution was consistently controlled within 0.3. The final parameter version management tree recorded five key changes: the key time period extension scheme version V1.2 → V1.5, the payment cycle parameter reverted from 1.8 to 1.75, and the technical weight coefficient was slightly adjusted from 1.4 to 1.38. All operational data was hashed and encrypted before being stored on the blockchain, forming a timestamped evidence chain containing 17 operational nodes.

[0041] Example 3: When the review database unit detects that the file type information is sensitive project bidding information, it immediately initiates a special protocol. The system first accesses the classified project database through a security authentication gateway. This database is protected by a three-level encryption system and only allows access from terminals authenticated by hardware keys. Sensitive project identification is based on a pre-set combination of 128 keywords, including category tags such as national defense security, major infrastructure, and core technology research and development. Each tag corresponds to a different data access permission level. When the system confirms that the review time is less than the preset review time, it triggers a basic optimization program call process. This process uses a distributed query engine to search for matching items in parallel across 17 special partitions of the review database. The search conditions include dimensions such as project sensitivity level, budget size threshold, and technical complexity coefficient.

[0042] The scoring foundation optimization procedure is generated using a modular construction approach. The system extracts core review elements from the basic rule base and combines them into an initial framework. This framework includes 22 mandatory national security clause verification modules and 35 technical confidentiality check modules. Subsequently, customized rules are injected based on the specific characteristics of sensitive projects. For example, a weapons and equipment qualification verification chain is added for military projects, and an aerospace-grade standard compliance testing unit is injected for aerospace projects. The parameter settings of the foundation optimization procedure follow a conservative principle, with scoring thresholds generally 12-18 percentage points higher than ordinary projects, and a zero-tolerance policy is set for the rejection conditions of key qualification items. During program execution, a real-time monitoring thread is started, which continuously monitors the data flow during the scoring process and records the operation logs of all logical judgment nodes. When the review time exceeds the preset value, the system activates the upgrade optimization mechanism. The fair review optimization unit first conducts a timeliness assessment of the foundation optimization procedure. The assessment method uses a sliding window comparison technique: six months prior to the current time point is used as the baseline window, and the revision frequency and importance score of relevant legal clauses within this window are calculated. The timeliness assessment score is calculated using the following formula:

[0043]

[0044] in: Indicates timeliness score, The weighting coefficient representing the i-th type of regulatory clause. It is the time difference between the latest revision time of this type of clause and the current time. It is the maximum time difference threshold allowed by the system. Score the relevance of the revised terms to the current project. This is the maximum relevance score. When... When the value drops below 0.7, the system will automatically trigger a rule update process.

[0045] The secondary optimization program is constructed using a dual-track architecture: the main track inherits the core logic framework of the basic optimization program and injects a timeliness compensation factor, which is based on... The system dynamically adjusts the stringency of rules; a parallel verification channel is established for the auxiliary track, which connects to the latest industry standard database and policy and regulation database, ensuring the cutting-edge nature of review standards through real-time rule comparison. The review results from both tracks are merged by a consistency arbitrator using a weighted voting mechanism, with the main track result accounting for 60% and the auxiliary track result for 40%. When the divergence exceeds 15%, the results are automatically submitted to the expert committee for adjudication. Data processing for sensitive projects is protected by end-to-end encryption. All scoring operations are performed in a secure, isolated environment, data transmission uses quantum encryption protocols, and temporary data storage uses one-time write-once memory. The system establishes an operator behavior audit trail, recording the operator's identity, timestamp, and operation content for each review step. These records are synchronized in real-time to three off-site backup nodes. Before the review results are output, they undergo three verifications: the first verification checks the compliance of basic clauses; the second verification verifies the consistency of scoring logic; and the third verification involves multi-party countersigning of the final result. Differentiated processing strategies are implemented for projects with different sensitivity levels: high-sensitivity projects initiate a full-process manual review mode, with the system automatically assigning three independent experts for back-to-back review; medium-sensitivity projects adopt a human-machine collaborative review, with the system handling 70% of the standardized review content and the remaining 30% being focused on by professionals; low-sensitivity projects are allowed to operate autonomously by the system but require an additional random sampling review step. The review process for all sensitive projects generates tamper-proof blockchain evidence, including the review program version hash value, data integrity proof, and operation time-series fingerprint. A five-level response mechanism is in place for handling anomalies: connection isolation is immediately initiated when abnormal data access is detected; a logic freeze procedure is triggered when rule matching conflicts occur; a major event warning is automatically escalated when the review time exceeds the expected value by 50%; a full-process backtracking is initiated when the scoring result fluctuates by more than 20% of the historical benchmark value; and service is immediately stopped and switched to a backup system when the system self-checks and discovers security vulnerabilities. Intermediate data generated throughout the implementation process is subject to a tiered destruction strategy: ordinary process data is automatically deleted within 24 hours after the review is completed, and core review data is encrypted and archived in a national-level confidential data center.

[0046] Example 4: When processing a certain type of aerospace sensor procurement project, the parameter weight acquisition unit initiated multi-dimensional parameter analysis. This project was identified as a high-sensitivity tender. The system first scanned all 328 parameters in the secondary optimization program for scoring. Through correlation analysis, it identified core correlated parameters including Business Weight (BW), Technical Weight (TW), and Key Scoring Optimization Time Period (KT). The Business Weight covers six sub-items such as price reasonableness and payment method; the Technical Weight includes nine indicators such as sensor accuracy and environmental adaptability; and the Key Scoring Optimization Time Period is locked within the 45-minute core window of the technical presentation. Non-correlated parameters identified 27 auxiliary control parameters, such as log recording frequency, result output format, and data backup cycle. The system calculated the adjustment weight of each parameter using a parameter influence assessment model. This model employs regression analysis based on historical project data to establish a correlation coefficient between parameter changes and the accuracy of the final score. The sensitivity level acquisition unit extracts the classification identifier of the aerospace project from the project filing database, analyzes whether the project involves key defense technologies and exceeds the warning threshold, and automatically classifies it as a high-sensitivity project. The system accesses the sensitive project database to retrieve weight configuration templates for similar projects and implements a differentiated weight allocation strategy based on the sensitivity level. The weight allocation of related parameters for high-sensitivity projects follows the principle of "technology first," with technology weight receiving the highest configuration value, followed by commercial weight, and the weight for key time periods is dynamically adjusted according to the complexity of the technology. The weight of unrelated parameters adopts a fixed-proportion reduction mode to ensure that auxiliary parameters do not excessively affect the core review logic.

[0047] After receiving the weight configuration, the fair review optimization unit activates the parameter adjustment engine, which uses a progressive optimization algorithm to calculate the specific adjustment value for each parameter. The calculation of related parameter adjustment values ​​is based on three constraints: technical weight adjustments must ensure 100% coverage of core technical indicators; business weight adjustments must maintain a reasonable ratio between price and technical scores; and adjustments during critical time periods must ensure sufficient time for core review stages. Non-related parameter adjustments adopt a simplified processing mode, primarily involving fine-tuning based on system load and operational specifications. All parameter adjustment processes are verified in real time; after each parameter adjustment is completed, the review process is immediately simulated to detect changes in the stability of the scoring results. Referring to Table 1, the system generates a parameter weight configuration table during execution to record the complete adjustment process.

[0048] Table 1: Parameter Weighting Configuration Table for Aerospace Sensor Projects

[0049]

[0050] During the parameter adjustment execution phase, a step-by-step verification mechanism is adopted. The system decomposes the adjustment process into four orderly stages: the first stage adjusts the technical weight parameters to ensure that core technical indicators are fully reflected; the second stage optimizes the business weight parameters to balance the weight distribution of price factors and technical performance; the third stage processes parameters for critical time periods and rationally allocates review time resources; and the fourth stage uniformly adjusts unrelated parameters to optimize system operating efficiency. After each stage, a specific verification test is conducted to ensure that parameter adjustments do not cause system performance degradation or review logic anomalies. For the special requirements of highly sensitive projects, the system implements enhanced security control measures: all parameter adjustment operations require two-factor authentication, and must simultaneously obtain electronic signature authorization from both the system administrator and the project leader; the parameter transmission process uses an end-to-end encryption protocol to prevent data theft or tampering during transmission; the system establishes a complete audit log for parameter adjustments, recording the operator, timestamp, pre-adjustment value, and post-adjustment value for each adjustment, and these logs are synchronized to the security audit server in real time. When encountering parameter conflicts or contradictory weight allocations, the system initiates an intelligent arbitration procedure: first, it attempts to use a historical best-solution matching algorithm to find the optimal parameter configuration for similar projects; if no matching solution is found, it initiates a multi-objective optimization algorithm to seek the optimal parameter combination that satisfies all constraints; when automatic arbitration fails to resolve the conflict, the system automatically generates a special report and submits it to an expert committee for manual adjudication. The entire parameter adjustment process maintains a high degree of transparency and traceability, with all decision-making basis and adjustment reasons recorded in detail in the project file. The system also has a parameter rollback mechanism; when adjusted parameters are detected to cause abnormal scoring or system performance degradation, it automatically reverts to the previous stable version of the parameter configuration. Rollback decisions are based on real-time monitoring data; when key indicators deviate from the normal range beyond a preset threshold, a rollback operation is immediately triggered to ensure the stability and reliability of the review process. The final version of all parameter configurations undergoes rigorous verification testing to ensure that they meet the stringent requirements of highly sensitive project reviews.

[0051] Example 5: When analyzing the bidding documents for a smart park construction project in a certain city, the scoring and evaluation unit initiated score fluctuation monitoring. The project was identified as a general project bidding information type. The system first extracted the scoring benchmark dataset of similar municipal greening projects in the past three years from the historical project database. This dataset contains the technical score distribution, the average commercial score, and the median of the comprehensive score for 27 completed projects. The system uses sliding window comparison technology to calculate the score decline of the current bid document. The specific method is as follows: the current technical score is compared with the same item in the benchmark dataset to obtain the single item deviation; the commercial score is calculated as the percentage difference with the historical average; and the comprehensive score is calculated as the standard deviation distance with the median. The three indicators are weighted and integrated to generate the comprehensive score decline value, with the weight allocation as 40% for technical, 30% for commercial, and 30% for comprehensive. When the system detects that the overall score has dropped to a preset threshold of 8.5 points and the review time is less than the preset value of 36 hours, the review database unit immediately activates the optimization program retrieval module. This module searches for suitable solutions in the basic optimization program library using a project feature vector matching algorithm. The search criteria include project scale classification (this project is a medium-sized municipal engineering project), the main cause of the score drop (analysis shows that the technical solution score is abnormally low), and the type of bidding enterprise (dominated by private enterprises). The system ultimately matches the basic optimization program with the number OP-07. This program is specifically designed for scenarios with sudden drops in technical solution scores and includes three core adjustments: a temporary 5% increase in the weight of the technical score, a 10% reduction in the threshold of the innovation indicator, and the addition of a cross-checking process in the feasibility verification stage. When the review time exceeds the preset value, a secondary optimization mechanism is triggered. The fair review optimization unit first performs a timeliness correction on the basic optimization program: it accesses the latest 2024 version of the "Technical Specifications for Smart Landscape Construction" and automatically compares the updates of the 12 key indicators in the technical scoring details; it also simultaneously scans the feature library of winning bids for similar projects recently to extract the frequency distribution data of technical innovation points. The system generates a secondary optimization program for scoring based on these dynamic data. The main adjustments include: restoring the threshold of innovation indicators to the standard level but adding an expert explanation and clarification process; changing the feasibility verification of the scheme to a three-party back-to-back review mode; and reducing the increase in the technical weight to 3% but adding an additional sub-item for technical potential.

[0052] The downgrade level determination unit continuously monitors score changes during program execution. When it detects that the technical solution score is still lower than the benchmark value of 6.2 points, it initiates a level reassessment. The system maps the score decline value to a three-level classification system: 0-3 points indicate a slight decline corresponding to the L1 optimization program; 3.1-6.5 points indicate a moderate decline triggering the L2 version; and scores above 6.6 points indicate a severe decline triggering the L3 version. This assessment confirms the downgrade level as moderate (L2 level), and the system automatically switches to the more stringent OP-07L2 optimization program. This version is characterized by an increased technical score weighting of 7%, the addition of a special technical risk assessment module, and the use of interval segmentation for calculating the price score in the business assessment. A dynamic monitoring strategy is adopted during the program execution phase. The system updates the score decline trend chart after each scoring stage: the horizontal axis displays the percentage of review progress, and the vertical axis marks the change in the gap between the real-time score and the benchmark value. When the score gap narrows to 4.8 points after the technical solution defense stage, the system automatically downgrades to the L1 version optimization program. If a new score drop occurs during the business review stage, a second upgrade assessment is immediately triggered. All version switching operations are recorded in the decision log, including timestamps, reasons for switching, and expected effect analysis. A specific handling procedure is in place for abnormal score fluctuations: when a single score mutation is detected exceeding three standard deviations of the historical fluctuation range, the system pauses the review process and initiates data verification; when multiple bidders experience a simultaneous decline in scores, the horizontal comparison module is activated; when two consecutive version switches fail to improve the score, manual review is mandatory. The entire process generates a score repair trajectory report, detailing the input parameters, processing logic, and output results for each decision node.

[0053] Before the final review, the system performs credibility verification of the results: It analyzes the differences between the optimized program output and the original scoring program results; if the difference in technical score exceeds 8% or the difference in commercial score exceeds 5%, a review process must be initiated. Simultaneously, it compares the results with the winning score range of similar recent projects to ensure the results conform to a reasonable market range. All verification data is compiled into a digitally signed confirmation document, a mandatory attachment to the review report. Full-cycle tracking is implemented for data management: original scoring data is stored in read-only mode to prevent tampering; the entire parameter adjustment sequence is recorded during the optimization program's operation; and alternative solutions are retained for comparative analysis during version switch decisions. When a project is archived, a special analysis chapter on scoring optimization is automatically generated, including the basis for selecting the benchmark value, the calculation process for the score reduction, the matching logic of the optimization program, and the final effect verification data, forming a closed-loop evidence chain.

[0054] 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.

Claims

1. A fair review system for bidding documents based on artificial intelligence, characterized in that, The system includes a document acquisition unit, a scoring model unit, and a fair review optimization unit; The document acquisition unit is used to acquire the bidding document information input by the user and to acquire the scoring rules information corresponding to the bidding document; wherein, the bidding document information includes document type information and scoring weight information, and the scoring rules information includes commercial scoring rules and technical scoring rules. The scoring model unit is used to determine the scoring procedure based on file type information; The fair review optimization unit is used to determine the scoring optimization procedure based on the document type information, scoring details information, and scoring procedure, and to complete the fair review through the scoring optimization procedure. The system also includes a review duration acquisition unit and an expert feedback acquisition unit. The review duration acquisition unit is used to acquire the review duration corresponding to the scoring rules information, where the review duration is the time span for collecting the scoring rules information. When the document type information is material bidding information, equipment bidding information, or subcontracting bidding information, and when the review duration is less than the preset review duration, the expert feedback acquisition unit is used to acquire the business score and technical score from expert feedback. Through the first-level optimization model unit, a first-level optimization procedure for scoring is determined based on the business score, technical score, and scoring procedure. When the review duration is greater than or equal to the preset review duration, through the fair review optimization unit, a second-level optimization procedure for scoring is determined based on the document type information, scoring rules information, and the first-level optimization procedure for scoring, and a fair review is completed through the second-level optimization procedure for scoring. After receiving expert scoring data, the first-level optimization model unit starts the multi-source scoring fusion engine, which contains three core processing layers: the first layer performs difference analysis between the scoring program and the expert scores, and uses a dynamic weight allocation algorithm to calculate the confidence weight of each scoring item; the second layer constructs a scoring mapping matrix, establishes a correlation mapping relationship between the business scoring rules and the technical scoring rules, and eliminates the differences in scoring dimensions through matrix transformation; the third layer implements rule reconstruction, and automatically generates supplementary scoring rules and embeds them into the original scoring program for special cases marked in the expert feedback as not covered by the scoring rules. The fair review optimization unit initiates the secondary optimization process, which includes two core modules: a time decay compensation mechanism and a rule update mechanism. The time decay compensation mechanism analyzes the specific number of days the review time exceeds the preset value to correct the time-sensitive parameters in the primary optimization process. The rule update mechanism accesses the latest regulatory database in real time and automatically compares the updates of the industry standards applicable to the project under review. When it detects that the regulatory clauses on which the scoring rules are based have been revised, the system automatically converts the new clauses into scoring rules and overwrites the old rules.

2. The artificial intelligence-based fair review system for bidding documents as described in claim 1, characterized in that, The system also includes a parameter acquisition unit; the parameter acquisition unit is used to acquire the adjustment rate of the first-level scoring optimization program within the business scoring range, and to acquire the key scoring time periods of the first-level scoring optimization program in the business and technical scoring; wherein, the key scoring time period is the time period with the highest scoring weight, and different file types correspond to key scoring time periods of different durations; through the fair review optimization unit, based on the file type information, scoring rules information, and the key scoring time periods of the first-level scoring optimization program, the key scoring optimization time periods in the second-level scoring optimization program are determined, limiting the adjustment rate adjustment range to be less than the preset adjustment rate adjustment range, and limiting the duration adjustment range of other scoring time periods of the first-level scoring optimization program to be less than the preset duration adjustment range.

3. The artificial intelligence-based fair review system for bidding documents as described in claim 2, characterized in that, The system also includes an enterprise qualification acquisition unit and a constraint margin identification unit; the enterprise qualification acquisition unit is used to acquire the user's enterprise qualification information, which includes registered capital, industry experience and historical performance information; through the constraint margin identification unit, the preset adjustment rate adjustment range and preset duration adjustment range are determined according to the scoring rules information and enterprise qualification information.

4. The artificial intelligence-based fair review system for bidding documents as described in claim 3, characterized in that, The system also includes a review database unit; when the file type information is sensitive project bidding information, and when the review time is less than the preset review time, the review database unit obtains the scoring basis optimization program corresponding to the sensitive project bidding information. The scoring basis optimization program is used to meet the basic review requirements of the sensitive project corresponding to the sensitive project bidding information; when the review time is greater than or equal to the preset review time, the fair review optimization unit determines the scoring secondary optimization program based on the sensitive project bidding information, scoring details information, and scoring basis optimization program, and completes the fair review through the scoring secondary optimization program.

5. The artificial intelligence-based fair review system for bidding documents as described in claim 4, characterized in that, The system further includes a parameter weight acquisition unit; the parameter weight acquisition unit is used to acquire the associated and unassociated parameters of the scoring secondary optimization procedure, and to acquire the adjustment weights corresponding to the associated and unassociated parameters respectively; wherein, the associated parameters include business weights, technical weights, and key scoring optimization time periods, and the unassociated parameters include control parameters other than the associated parameters in the scoring secondary optimization procedure; through the fair review optimization unit, based on sensitive project bidding information, scoring rules information, scoring basic optimization procedure, and the adjustment weights corresponding to the associated and unassociated parameters respectively, the system determines the associated parameter adjustment values ​​corresponding to the associated parameters and the unassociated parameter adjustment values ​​corresponding to the unassociated parameters in the scoring secondary optimization procedure.

6. The artificial intelligence-based fair review system for bidding documents as described in claim 5, characterized in that, The system also includes a sensitivity level acquisition unit; the sensitivity level acquisition unit is used to acquire the project sensitivity level corresponding to the bidding information of sensitive projects; wherein, the project sensitivity level includes the sensitivity levels corresponding to high-sensitivity projects, medium-sensitivity projects and low-sensitivity projects respectively; according to the project sensitivity level, the adjustment weights corresponding to the associated parameters and non-associated parameters are determined respectively.

7. The artificial intelligence-based fair review system for bidding documents as described in claim 6, characterized in that, The system also includes a scoring and evaluation unit; when the document type information is ordinary project bidding information, the scoring and evaluation unit determines the score reduction score according to the scoring rules information; when the score reduction score is greater than or equal to the preset score reduction score, and when the review time is less than the preset review time, the scoring base optimization program corresponding to the preset score reduction score is obtained through the review database unit. When the review duration is greater than or equal to the preset review duration, the fair review optimization unit determines the secondary optimization procedure for scoring based on document type information, scoring details, and scoring basic optimization procedure, and completes the fair review through the secondary optimization procedure for scoring.

8. The artificial intelligence-based fair review system for bidding documents as described in claim 7, characterized in that, The system also includes a degradation level determination unit; the degradation level determination unit is used to determine the degradation level corresponding to the degradation score; and obtains different rating base optimization programs corresponding to the degradation level by reviewing the database unit, the parameter thresholds of the different rating base optimization programs being different.

9. A method for fair review of bidding documents based on artificial intelligence, characterized in that, It includes all modules and method processes of the artificial intelligence-based fair review system for bidding documents as described in any one of claims 1 to 8.

Images