An AI-based intelligent auxiliary evaluation system for bidding and tendering

Abstract

This application relates to the field of computer application technology and discloses an intelligent auxiliary evaluation system for bidding based on artificial intelligence. A full-domain dynamic data support module generates a set of structured pointers with digital signatures and maintains a full-domain state machine for core resources, ensuring the authenticity and tamper-proof nature of the evaluation data source. An intelligent logic verification module incorporates a logic conflict verification network, performing multi-dimensional logic conflict deduction on bidding resources based on real-time resource status and physical transfer speed thresholds, automatically identifying qualification fraud and false resource allocation. A violation identification and adaptive feedback module possesses algorithmic adaptive evolution capabilities, capturing expert feedback signals through a dynamic weight correction mechanism, correcting text similarity calculation logic, accurately identifying bid rigging and collusion, and forming a human-machine collaborative learning feedback loop. This invention, through automated and intelligent assistance, significantly improves the objectivity, accuracy, and efficiency of bid evaluation, curbs bidding irregularities, and optimizes the business environment.

Description

Technical Field

[0001] This invention relates to the field of computer application technology, specifically to an intelligent auxiliary evaluation system for bidding based on artificial intelligence. Background Technology

[0002] With the expansion of engineering bidding and tendering business and the increasing complexity of business scenarios, traditional bid evaluation models are no longer sufficient to meet the demands for efficient and accurate business processing, given the massive and heterogeneous nature of bidding and tendering data from multiple sources. Current technologies mainly suffer from the following shortcomings: In the existing bidding process, the data management system is fragmented, with regulatory data often scattered across different business systems, creating data silos. This results in a lack of real-time and comprehensive assessment of the bidding entities and their committed resource status. For example, there is a lack of dynamic and centralized monitoring mechanisms for the resource status of bidding companies or their key personnel in other projects, making it difficult for bid evaluators to logically deduce and verify the feasibility of resource allocation and usage across different projects. This deficiency provides an opportunity for qualification-for-profit practices, thus compromising the objectivity of the bidding results.

[0003] Furthermore, when reviewing the content of bid documents, existing automated analysis tools primarily rely on static similarity thresholds to judge bid-rigging and collusion behaviors with highly similar texts. This lacks adaptability and a professional understanding of the review scenario. When bidders use minor modifications or synonym substitutions to circumvent these restrictions, the system is prone to false positives or false negatives. Because the system cannot translate expert opinions into algorithm parameter optimizations, its accuracy is difficult to improve through practical application feedback. This forces experts to invest significant effort in the review process, failing to effectively achieve human-machine collaboration and impacting the efficiency and fairness of the bid evaluation process.

[0004] Meanwhile, the lack of unified standardization and control mechanisms for data flow, business flow, and decision-making flow during the bidding process leads to insufficient transparency and traceability throughout the entire process. Unstructured data referencing and process uncertainty not only increase the burden of manual review but also create opportunities for external service providers to exploit information asymmetry for irregular pricing, directly hindering the optimization of the business environment. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent auxiliary evaluation system for bidding based on artificial intelligence, aiming to solve problems such as low efficiency, difficulty in identifying qualification fraud and bid rigging in traditional evaluation methods.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: an intelligent auxiliary evaluation system for bidding based on artificial intelligence, comprising: a full-process business management module, a full-domain dynamic data support module, an intelligent logic verification and objective review module, a violation identification and adaptive feedback module, and an expert review and dynamic decision-making module.

[0007] The full-process business management module, acting as the system's central scheduling mechanism, is primarily responsible for centralized control of the entire bidding process. Its configuration maintains a set of business process states, including configuration, bidding, review, and public announcement states, and coordinates the sequential transitions between these states with various functional modules. Simultaneously, this module distributes necessary algorithm control parameters to other computing modules, such as attenuation coefficients for text analysis and maximum speed thresholds for physical verification.

[0008] The dynamic data support module, controlled by the end-to-end business management module, serves as the system's data foundation. Its configuration allows access to multi-dimensional raw data sources, standardizing the dispersed raw data to generate a set of structured pointers for bidding reference, thus decoupling the data reference layer from the raw storage layer. Furthermore, this module maintains the global state machine of core resources in real time through a global state monitoring mechanism and outputs the real-time resource status as a benchmark for logical reasoning by downstream modules.

[0009] The intelligent logic verification and objective review module is primarily responsible for objective scoring calculation and in-depth logic verification. This module has a built-in logic conflict verification network, configured to call upon the real-time resource status output by the full-domain dynamic data support module. It performs multi-dimensional logic conflict deduction based on state mutual exclusion and spatiotemporal physical constraints on the resources promised in the bid documents, thereby determining the administrative compliance and physical feasibility of the bid and outputting the logic conflict verification results.

[0010] The violation identification and adaptive feedback module primarily analyzes the similarity of unstructured text and possesses adaptive learning capabilities. Its configuration is used to extract unstructured bidding text from the multi-dimensional raw data sources accessed by the full-domain dynamic data support module, calculate semantic similarity based on the feature vector space, and output the similarity calculation results. This module has algorithm adaptive evolution capabilities, configured to respond to algorithm control parameters distributed by the full-process business management module, and capture expert feedback confirmation signals from the expert review and dynamic decision-making modules. It updates feature word weights through a dynamic weight correction mechanism to achieve numerical fitting of the algorithm to expert decision preferences.

[0011] The expert review and dynamic decision-making module serves as the system's human-computer collaborative interaction terminal. Its configuration aggregates the logical conflict verification results output by the intelligent logic verification and objective review module, as well as the similarity calculation results output by the violation identification and adaptive feedback module. This module employs differentiated review logic, pushing only conflicting or outlier items exceeding warning thresholds to experts, reducing the workload of manual review. Simultaneously, this module is configured to collect manual adjudication instructions, generate the final adjudication result, and send expert feedback confirmation signals to the violation identification and adaptive feedback module, forming a learning feedback closed loop. Furthermore, this module possesses dynamic decision-making capabilities; when the final adjudication result determines the current candidate to be invalid, it automatically triggers dynamic replacement calculations in the candidate queue.

[0012] Preferably, in the full-domain dynamic data support module, the structured pointer generation unit is configured to perform digital signature processing on the generated set of structured pointers to generate a unique verification fingerprint. During the review phase, the system uses a hash algorithm to compare the verification fingerprint with the digest value of the currently read data. After verifying consistency, it retrieves the source data from the main database for rendering, effectively ensuring the authenticity and tamper-proof nature of the evaluation data source.

[0013] Preferably, the state space defined by the global state machine includes at least idle, locked, under construction, and maintenance states. The global dynamic data support module responds to external event signals, such as bid-winning notifications or completion filings, triggering automatic state transitions to ensure dynamic updates of the resource's real-time status.

[0014] In one specific embodiment, the logic conflict verification network in the intelligent logic verification and objective review module includes a cascaded state retrieval and analysis subunit and a spatiotemporal constraint calculation subunit. The state retrieval and analysis subunit performs state mutual exclusion verification at the administrative compliance level based on the real-time resource status. The spatiotemporal constraint calculation subunit performs spatiotemporal logic conflict verification at the physical feasibility level for resource scheduling between different projects based on the physical transfer limit speed threshold distributed by the full-process business management module.

[0015] Furthermore, the specific logic for the spatiotemporal constraint calculation subunit to perform spatiotemporal logic conflict verification is as follows: Calculate the shortest physical path distance and available time window between the preceding project and the current project. If the available time window is less than or equal to zero, the physical transfer speed requirement is determined to be infinite.

[0016] When the available time window is greater than zero, calculate the physical transfer speed requirement value and compare it with the physical transfer limit speed threshold. If it exceeds the threshold, generate a physical conflict indication result.

[0017] Ultimately, the logical conflict verification network generates a solidified chain of logically contradictory evidence based on these indications.

[0018] In a preferred embodiment, the specific method by which the violation identification and adaptive feedback module performs algorithm adaptive evolution includes: First, the feature vector of the unstructured bid text is constructed using the TF-IDF algorithm; Secondly, upon receiving expert feedback confirmation signals, the weight values ​​of the feature words confirmed as reasonable citations are reduced using an exponential decay model. In subsequent similarity calculations, cosine similarity is calculated based on the updated feature word weights, enabling the algorithm to numerically fit the review experts' decision preferences.

[0019] Furthermore, the dynamic weight correction mechanism adopts an exponential decay formula, wherein the decay rate of the feature word weight is controlled by the decay coefficient configured by the full-process business management module, and the decay magnitude is positively correlated with the cumulative number of expert feedback confirmation signals.

[0020] In one specific embodiment, the differentiated review logic implemented by the expert review and dynamic decision-making module is as follows: Built-in warning filtering rules generate abnormal warning records and highlight them to the expert review interface only when a logically contradictory evidence chain is received or the similarity calculation result exceeds the threshold; for scoring items that do not trigger warnings, the objective scoring calculation result of the system is accepted by default.

[0021] In addition, this module is configured to perform the highest priority determination: When a valid expert input instruction is received, the final decision result is forcibly anchored to the value or status corresponding to the expert input instruction, overriding the system's automatic calculation result.

[0022] Furthermore, if the final decision determines that the current candidate's bid is invalid, a dynamic replacement operation is triggered to automatically remove invalid nodes and re-rank the remaining valid bidders based on their comprehensive scores, ensuring the efficiency of the bid evaluation and the validity of the results.

[0023] In one specific embodiment, the full-process business management module strictly coordinates the running order of system functional modules by maintaining a set of business process states: in the bidding state, the data access interface is opened but the violation identification and adaptive feedback module is locked; in the review state, the parallel computing capabilities of the intelligent logic verification and objective review module and the violation identification and adaptive feedback module are activated; in the public announcement state, the final decision result is solidified and the global state machine and all computing modules are locked to prevent data from being tampered with.

[0024] This invention provides an intelligent auxiliary evaluation system for bidding and tendering based on artificial intelligence. It has the following beneficial effects: 1. This invention constructs a multi-dimensional logical conflict deduction capability based on real-time resource status and physical transfer speed threshold through an intelligent logic verification and objective review module. The system can automatically and accurately identify the state mutual exclusion and spatiotemporal logical conflicts between bidders in cross-project resource scheduling, and generate a logical contradiction evidence chain. This enables the system to proactively discover and solidify violations such as qualification fraud and false resource allocation during the bid evaluation stage, greatly reducing the workload of experts in standardized review, concentrating manual review on anomalies, thereby shortening the bid evaluation time and enhancing the fairness of the bid evaluation.

[0025] 2. This invention constructs an expert feedback-driven algorithm adaptive evolution mechanism through the collaboration of a violation identification and adaptive feedback module and an expert review and dynamic decision-making module. When calculating text similarity, the system can adjust the feature word weights in real time based on expert feedback confirmation signals through a dynamic weight correction mechanism. This effectively solves the problems of misjudgment and omission in traditional text similarity analysis, enabling the system to more accurately identify homogeneous bid documents commonly found in bid rigging and collusion, ensuring the accuracy of bid evaluation, and improving the system's scenario adaptability.

[0026] 3. This invention generates a set of structured pointers with digital signatures through a global dynamic data support module and maintains the global state machine of core resources. This effectively prevents tampering with bidding data during the evaluation process and ensures the real-time and traceability of resource status. It achieves standardized and automated management of the entire process. This transparency and standardization of data and processes not only provides reliable data support for administrative supervision but also eliminates the space for service agencies to illegally charge fees by taking advantage of information asymmetry. This directly optimizes the business environment for bidding and reduces administrative costs. Attached Figure Description

[0027] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a schematic diagram of the structure of the global dynamic data support module of the present invention; Figure 3 This is a schematic diagram of the intelligent logic verification and objective review module of the present invention; Figure 4 This is a schematic diagram of the violation identification and adaptive feedback module of the present invention; Figure 5 This is a schematic diagram of the expert review and dynamic decision-making module of the present invention; Figure 6 This is a schematic diagram of the structure of the full-process business management module of the present invention.

[0028] Among them, 10. Full-domain dynamic data support module; 20. Intelligent logic verification and objective review module; 30. Violation identification and adaptive feedback module; 40. Expert review and dynamic decision-making module; 50. Full-process business management module. Detailed Implementation

[0029] The technical solutions in 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.

[0030] See attached document Figure 1 The present invention provides an intelligent auxiliary evaluation system for bidding based on artificial intelligence. The system includes: a dynamic data support module 10, an intelligent logic verification and objective review module 20, a violation identification and adaptive feedback module 30, an expert review and dynamic decision-making module 40, and a full-process business management module 50.

[0031] The full-domain dynamic data support module 10 is used to build the basic database for transportation engineering bidding and maintain the status of full-domain resources. This module does not directly store unstructured bid document images or PDF documents, but instead stores a set of structured data pointers. The full-domain dynamic data support module 10 establishes a main database, where bidders pre-enter personnel, equipment, and performance information. When generating bid document data, the full-domain dynamic data support module 10 generates a database containing a unique enterprise identifier. Personnel reference pointer Device reference pointer and performance reference indicators Structured reference collection .

[0032] In addition, the global dynamic data support module 10 is also used to maintain the global state machine of core resources, including key personnel. and key equipment The resource set is denoted as Any resource At any moment status Belongs to the state space ,in ; Corresponding to idle time; The corresponding winning candidate has been locked in; Corresponding to the ongoing construction and locking; Corresponding maintenance status. The full-domain dynamic data support module 10 updates the above status in real time by connecting to the data interface of the trading platform. .

[0033] The intelligent logic verification and objective evaluation module 20 is connected to the global dynamic data support module 10, and is used to perform deterministic rule calculations and logic conflict verification based on global resource constraints. The intelligent logic verification and objective evaluation module 20 first calculates the price score and credit score based on a preset scoring function. More importantly, the intelligent logic verification and objective evaluation module 20 uses the data provided by the global dynamic data support module 10 to evaluate the resources committed to the bid documents. Perform dual logical checks across projects.

[0034] The first layer of verification is a state mutual exclusion verification. The intelligent logic verification and objective review module 20 calculates the state conflict indication function. To determine resource availability, the state conflict indicator function is defined as follows: ; in, This indicates the time window for the current bidding project. For resources at all times The state. If If so, then a state conflict is determined to exist; Indicates if time exists Within the current project timeframe, and with resources available at all times. The status is either locked as a winning candidate or locked under construction, which is a conditional judgment.

[0035] The second layer of verification is a spatiotemporal physical constraint verification. Even when the resource status shows as idle, the intelligent logic verification and objective review module 20 still needs to verify the feasibility of physical transfer. The location of the preceding project is set as... The current project location is The completion time of the preceding project is The current project requires an on-site arrival time of [date / time]. The intelligent logic verification and objective review module 20 calculates the physical conflict indication function. : ; in, Indicates the geographical distance between two places. Representing resources The physical transfer limit velocity threshold.

[0036] like The intelligent logic verification and objective review module 20 determines the existence of physical logic conflicts and generates a logical contradiction evidence chain. The final logic verification result... This is the sum of all resource conflict values.

[0037] The violation identification and adaptive feedback module 30 is used to perform multimodal feature analysis and parameter correction based on expert feedback. The module extracts the text feature vectors of the bid documents and calculates different features for different bid documents. and Cosine similarity between To address the false alarm issue caused by industry standard citations, the violation identification and adaptive feedback module 30 introduces an expert feedback-driven dynamic weight correction mechanism. The violation identification and adaptive feedback module 30 receives feedback signals transmitted from the expert review and dynamic decision-making module 40. When an expert selects a violation containing feature words... When a text segment is marked as a valid citation, the module updates the feature words. Real-time weights The weight updates follow the exponential decay formula: ; in, Characteristic words The initial inverse document frequency value, Characteristic words The cumulative number of times it has been marked as fair citation by experts. The pre-set attenuation coefficient is used. The violation detection and adaptive feedback module 30 utilizes the updated weights. Subsequent similarity calculations are performed to automatically filter standard clause content.

[0038] The expert review and dynamic decision-making module 40 connects to the intelligent logic verification and objective evaluation module 20 and the violation identification and adaptive feedback module 30, providing a human-computer interaction interface and executing dynamic supplementation algorithms. The expert review and dynamic decision-making module 40 is only displayed to the terminal. Warning items with similarity exceeding a threshold are also included. The expert review and dynamic decision-making module 40 is configured with the highest priority judgment logic, meaning that when the expert review result is inconsistent with the system calculation result, the expert's input instruction shall prevail. In addition, the expert review and dynamic decision-making module 40 includes a dynamic replacement unit. When an invalid bid confirmation instruction is received for a certain winning candidate, the dynamic replacement unit retrieves the substitute with the highest score from the set of remaining valid bidders and automatically updates the winning candidate queue without the need for manual re-aggregation of scores.

[0039] The end-to-end business management module 50 connects to each of the above modules and is used to control the state flow of the system. The end-to-end business management module 50 provides a configuration interface for users to set logic verification parameters, including time windows. and maximum speed and setting the attenuation coefficient The full-process business management module 50 controls the system to switch between configuration, bidding, review, and public announcement states, ensuring that each module executes data processing tasks according to the preset time sequence.

[0040] See attached document Figure 2 In this embodiment, the global dynamic data support module 10 is specifically configured as the system's underlying data processing unit, responsible for constructing the basic database for transportation engineering bidding and maintaining the global resource status. This global dynamic data support module 10 mainly includes a data standardization access unit, a structured pointer generation unit, a global status monitoring unit, and a state machine maintenance unit.

[0041] The data standardization access unit is equipped with a data cleaning interface for connecting to multiple dimensions of raw data sources, including but not limited to historical data from public resource trading platforms in various provinces and cities, data from enterprise credit information disclosure systems, and data from industry regulatory platforms. The data standardization access unit performs field mapping and cleaning operations on the collected raw data to build a unified format main database.

[0042] Within this main database, each bidding entity is assigned a unique enterprise identification code. Its associated qualification certificates, financial statements, performance records, and personnel resumes are all converted into structured metadata entries, rather than electronic documents in image format. For data storage and retrieval technologies, those skilled in the art can employ relational databases or distributed file systems. The specific storage media and read / write protocols are well-known technologies in the field and will not be elaborated upon here.

[0043] The structured pointer generation unit is used to generate a set of structured data references to replace traditional bid documents. During the bid document generation phase, the structured pointer generation unit does not perform file encapsulation operations; instead, in response to the bidder's selection instructions, it extracts the index key values ​​of the corresponding entries in the main database and constructs a set of data pointers. This data pointer set Defined as: ; In the formula, A unique identification code representing the bidding company; This represents a set of personnel reference pointers, containing the index addresses of records in the main database for key personnel such as project managers and technical leads. This represents a set of equipment reference pointers that point to the inbound records of key mechanical equipment such as tunnel boring machines and bridge erecting machines. This represents a set of performance reference pointers, pointing to the index of the filing records of completed and accepted projects.

[0044] The structured pointer generation unit uses a hash algorithm to generate... Perform digital signature processing to generate a unique verification fingerprint. During the review phase, when the system reads the data pointer set, it will again use the same hash algorithm to calculate the digest value of the currently read data and compare it with the stored verification fingerprint. A comparison is performed; if the two are inconsistent, the system determines that the data source has been tampered with during transmission or storage, refuses to load it, and triggers a security alert. This process ensures that the data source accessed during the review process is accurate. satisfy ,in This refers to the main database verification set that has been publicly disclosed and is currently locked. Through this referencing and anchoring mechanism, the system directly retrieves the source data from the main database for rendering during the review process. This achieves logical decoupling between the data presentation layer and the original storage layer, avoids OCR recognition errors, and prevents bidders from creating false materials for a single project.

[0045] The global status monitoring unit is used to track the usage of critical resources across platforms. Critical resources include key personnel. and key equipment Collectively referred to as resource collection The global status monitoring unit periodically synchronizes bidding result announcements, construction permit issuance records, and completion acceptance filing information nationwide through API interfaces or data crawlers, and maps the acquired external events into internal system status change signals.

[0046] The state machine maintenance unit is connected to the global state monitoring unit to provide services for the resource set. Each individual resource Maintain a real-time global state machine. The state machine maintenance unit defines resources. At any moment status Belongs to a finite state space : ; In the formula, This indicates an idle state, meaning the resource is not currently occupied by any project and is eligible for bidding. This indicates that the resource has been included in the list of candidates for a project, but a contract has not yet been signed and the project is in the public notice and objection period. This indicates a locked-in status, meaning that the project to which the resource is located has signed a contract or is already under construction. Resources in this status cannot be reused for bidding on other projects within the scope permitted by laws and regulations. This indicates a state of maintenance or unavailability, mainly for large machinery and equipment that are under maintenance or personnel whose certificates are temporarily suspended.

[0047] The state machine maintenance unit executes state transition operations based on received external event signals. The specific transition logic configuration is as follows: When the state machine maintenance unit receives information about resources... When the "Announcement of Successful Bidders" signal is received, the status of the resource will be updated. Depend on Updated to ; When a "notice of award issued" or "construction contract filed" signal is received, the status of the resource will be changed. Depend on Updated to ; When a "Completion Acceptance Report Filing" or "Personnel Change / Unlock Application Approved" signal is received, the status of this resource will be changed. Depend on Updated to ; When a "device sent for repair record" or "personnel certificate temporarily suspended administrative penalty" signal is received, the status of this resource will be updated. Updated to The state machine maintenance unit writes the updated state data to the global resource state table in real time, which is then used by subsequent modules for logical verification. Through the dynamic maintenance of this state machine, the system achieves exclusive management of physical resources in the time dimension.

[0048] See attached document Figure 3 In this embodiment, the intelligent logic verification and objective review module 20 is communicatively connected to the aforementioned global dynamic data support module 10, and is configured to perform objective score calculation based on deterministic rules and logic verification based on global constraints. Specifically, the intelligent logic verification and objective review module 20 includes an objective score calculation unit and a logic conflict verification network 22, which will be described in detail in subsequent embodiments.

[0049] The objective scoring calculation unit is used to quantify the price and credit factors of bidders according to the scoring method preset in the bidding documents. This unit is further divided into price scoring calculation subunit 2, credit scoring mapping subunit 2, and comprehensive score summary subunit 2.

[0050] Price scoring calculation subunit 2 is used to process bid price data. After the review is initiated, price scoring calculation subunit 2 first obtains the set of bid prices from all valid bidders, and calculates the benchmark price for evaluation according to preset rules (such as the reasonable low price method or the lowest evaluated price method).

[0051] Taking the reasonable low price method commonly used in traffic engineering as an example, the price scoring calculation sub-unit 2 calculates the benchmark price for bid evaluation. This benchmark price is typically taken as the arithmetic mean or weighted average of all valid bids. Subsequently, the price scoring calculation subunit 2 is based on the bid prices. Compared with the benchmark price Deviation rate calculation price score The specific computational model is expressed as follows: ; In the formula, This indicates the maximum score for the price item; This represents the deduction coefficient, which is set according to the tender documents, and is usually when... Time value ,when Time value ; This indicates the absolute value operation. If the calculated result is less than the preset minimum score threshold, then... Take the lowest score threshold.

[0052] Credit scoring mapping subunit 2 is used to access the main database in the full-domain dynamic data support module 10 and extract the credit rating data of the bidding entities. Unlike the traditional manual review of credit reports, credit scoring mapping subunit 2 directly reads the credit index from the structured data. And map it to a credit score The mapping relationship is defined by a piecewise function: ; In the formula, This indicates the maximum score for the credit item; and These represent the high and low thresholds of the credit index, respectively. This is the credit score decay coefficient. For cases where the credit rating levels (such as AA, A, B, C) issued by the administrative authorities are directly used, the credit score mapping subunit 2 directly outputs the corresponding fixed score by looking up a pre-set key-value pair mapping table.

[0053] Sub-unit 2, the comprehensive score aggregation unit, combines the price score and credit score to generate an objective total score. It outputs the objective total score based on a weighted allocation formula. : ; In the formula, and These are the price score weighting coefficient and the credit score weighting coefficient, respectively, and they satisfy the following conditions: (Considering only objective scores). The comprehensive score summary sub-unit 2 will calculate the... The value is written into the review results database and locked, only to be corrected by the expert account with the highest authority if an error in the calculation basis is found during subsequent expert review. Through the execution of the above deterministic algorithm, the system completes the quantitative evaluation of the basic objective data of the tender documents, laying the scoring foundation for subsequent intelligent logic verification.

[0054] In addition to the aforementioned objective scoring calculation unit, the intelligent logic verification and objective review module 20 also features a core logic conflict verification network. This network is not limited to isolated checks of a single bid document, but is constructed as a cross-dimensional verification architecture based on global data references. Specifically, the logic conflict verification network includes a state retrieval and analysis subunit, a spatiotemporal constraint calculation subunit, and an evidence chain generation subunit.

[0055] The status retrieval and analysis subunit is configured to perform state-based mutual exclusion resource availability checks. During the review process, the status retrieval and analysis subunit parses the set of data pointers in the tender documents. Extract a list of all key resources committed to the investment. For any resource in the list... The status retrieval and analysis subunit initiates a cross-project retrieval request to the global dynamic data support module 10 to obtain the time occupancy records of the resource in all other incomplete projects or projects currently under public notice. The status retrieval and analysis subunit not only verifies the current status but also verifies the overlap between the project duration of the proposed bidding project and the existing resource occupancy period.

[0056] The state retrieval and analysis subunit is based on the state conflict indication function. This function determines whether there are multiple promises made in violation of administrative regulations regarding resources. The logical definition of this function is as follows: ; In the formula, The planned construction period window specified in the tender documents for the current bidding project is defined as a closed interval. ; Representing resources The moments recorded in the global state machine The state; The candidate for winning the bid is locked. The system is in a locked state. The state retrieval and analysis sub-units are traversed. Within each time unit, once a resource is detected... At any point in time or The status indicates that a time conflict has occurred, and will... Assign a value of 1 if the value is 1, otherwise assign a value of 0.

[0057] This mechanism can identify resource conflicts that are currently idle but will be in the process of other projects under construction during the commencement of the project to be tendered.

[0058] The spatiotemporal constraint calculation subunit is configured to perform physical-logical conflict verification based on spatiotemporal constraints. This verification targets hidden violation scenarios where "administrative status is compliant but physical conditions are impossible." When the status retrieval and analysis subunit determines resources... If there is no conflict in the status (e.g., a preceding project finishes just before this project starts), the spatiotemporal constraint calculation subunit initiates physical feasibility verification. The spatiotemporal constraint calculation subunit extracts resources. In the preceding project Geographic coordinates and the estimated completion time of the preceding project. Simultaneously extract the current project Construction site coordinates and required entry time .

[0059] The spatiotemporal constraint computation subunit determines conflicts by calculating the difference between the theoretical velocity required for resource transfer and the physical limit velocity. The physical conflict indicator function... The structure is as follows: ; In the formula, This represents the shortest physical path distance between the previous project location and the current project location. This distance value is obtained by calling the path planning interface of the Geographic Information System (GIS), or calculated using the spherical distance formula between two points in simplified mode. Indicates the length of the time window during which resources can be transferred; Representing resources The corresponding physical transfer limit velocity threshold.

[0060] Before performing the division operation, the spatiotemporal constraint calculation subunit first determines the value of the time window length. If This means that the current project requires an entry time earlier than or equal to the completion time of the preceding project. The spatiotemporal constraint calculation subunit directly determines that the physical transfer speed requirement is infinite. Force the value to 1; if the time window length is greater than 0, perform the above division operation and combine it with the result. Compare them.

[0061] for The values ​​of are stored internally in the spatiotemporal constraint calculation subunit, which contains a classification threshold table: when resources When a key person, Set as the combined maximum speed limit for civil aviation transportation and ground connections; when resources For large-scale specialized equipment (such as tunnel boring machines and bridge erecting machines), This is set as the legally mandated maximum speed or physical driving limit for heavy-duty special transport vehicles on the relevant road grade, and includes necessary conversion factors for disassembly and assembly buffer time. If the calculated necessary transfer speed exceeds... or denominator If the value is negative or zero, the spatiotemporal constraint calculation subunit will... If the value is set to 1, it is determined to be a physical logic conflict.

[0062] The evidence chain generation subunit is used to solidify evidence when a conflict is detected. or At that time, the evidence chain generation subunit does not just output a simple rejection signal, but constructs a logically contradictory evidence chain data package containing complete traceability information.

[0063] The data packet contains: the name of the conflicting resource, the name and code of the preceding project, the time and geographical location of the preceding project, the time and geographical location of the current project, the conflict type (overlapping status or physically unreachable), and specific calculated parameter values. The evidence chain generation subunit transmits this evidence chain data packet to the subsequent expert review module as the objective basis for experts to determine whether to reject the bid, thus realizing a closed-loop support from data calculation to fact-finding.

[0064] See attached document Figure 4 In this embodiment, the violation identification and adaptive feedback module 30 is communicatively connected to the expert review and dynamic decision-making module 40, and is configured to perform deep feature analysis based on multimodal data and adaptive adjustment of algorithm parameters. The violation identification and adaptive feedback module 30 is not limited to simple keyword matching, but rather quantifies the homology between different bid documents by constructing a high-dimensional feature space. Specifically, the violation identification and adaptive feedback module 30 includes a feature vector construction unit and a similarity calculation unit.

[0065] The feature vector construction unit is used to transform unstructured bid text content into a computer-computable mathematical vector representation. In bidding scenarios, the feature vector construction unit mainly processes highly subjective sections such as construction organization design, technical solutions, and project management organization setup. The feature vector construction unit first preprocesses the collected text data, removing stop words, punctuation marks, and format control characters, and then uses a Chinese word segmentation algorithm to serialize the continuous text into a set of terms. For the specific word segmentation algorithm, those skilled in the art can use word segmenters based on Hidden Markov Models (HMMs) or Conditional Random Fields (CRFs), which are well-known technologies in the field and will not be elaborated upon here.

[0066] After word segmentation, the feature vector construction unit uses the Term Frequency-Inverse Document Frequency (TF-IDF) algorithm to calculate the weight value of each feature word. The feature vector construction unit constructs a feature vector space, where the vector dimension is... The size is determined by the feature dictionary maintained by the system. For any given tender document... Its corresponding feature vector Represented as , where each component The value is determined by the feature words Word frequency in the document Its inverse document frequency The product determines the weights. In this embodiment, to support subsequent dynamic weight correction, the feature vector construction unit uses the original weights of the feature words. Stored independently and maintained as a variable parameter.

[0067] The similarity calculation unit is connected to the feature vector construction unit and is used to quantify the content overlap between bid documents submitted by different bidders. The similarity calculation unit receives the feature vectors output by the feature vector construction unit and uses the cosine similarity algorithm as the core evaluation index.

[0068] Cosine similarity measures the consistency of the orientation of two vectors by calculating the cosine of the angle between them in a multidimensional space, thus effectively avoiding calculation biases caused by differences in document length. For the two tender documents to be compared... and The similarity calculation unit executes the following similarity calculation formula: ; In the formula, Document With Documents The similarity value ranges from 1 to 1. ; This represents the total dimension of the feature vector, which is the total number of feature words included in the feature dictionary. Indicates the index number of the feature word; Indicating characteristic words The weighting coefficients are initially set based on the IDF values ​​of the general corpus and are adjusted by the feedback mechanism during subsequent operation. Indicating characteristic words In the document The frequency of occurrence (TermFrequency) or Boolean presence flag in the text; Indicating characteristic words In the document Frequency of occurrence in. represent and Weighted dot product across all feature dimensions; They are and The length of the weighted vector; The similarity calculation unit calculates the similarity. Then, compare it with the preset warning threshold. Perform a comparison. If... The similarity calculation unit determines the bid documents. and If any abnormal similarities are found, it is identified as suspected bid-rigging or collusion, and a plagiarism report containing the location information of the similar paragraphs is generated. This calculation process covers different granularities from the paragraph level to the chapter level, and can identify plagiarism behaviors that are concealed by adjusting the paragraph order, replacing synonyms, etc., providing quantitative and objective evidence for subsequent expert review.

[0069] To address the issue of false text similarity reports in the field of transportation engineering due to the extensive use of mandatory national standards and construction specifications, the violation identification and adaptive feedback module 30, in addition to the aforementioned feature vector construction unit and similarity calculation unit, is further equipped with a dynamic weight correction unit. This dynamic weight correction unit constructs a closed-loop channel for expert decision data to feed back to the algorithm parameters, enabling online iterative updates of feature word weights.

[0070] The dynamic weight correction unit specifically includes a feedback signal capture subunit and a parameter decay update subunit. The feedback signal capture subunit communicates with the expert review and dynamic decision-making module 40, receiving real-time processing instructions from experts for warning items. When a high similarity warning output by the similarity calculation unit is marked as "reasonable citation" or "whitelist exemption" by an expert, the feedback signal capture subunit generates a feedback trigger signal. This signal contains the index information of the exempted text paragraph and the set of feature words contained in that paragraph. The feedback signal capture subunit maintains a global feedback counter for each feature word. Record the cumulative number of times the text paragraph containing the citation has been confirmed as a valid citation by experts. Each valid exemption operation will trigger the corresponding... Increase the value by one.

[0071] The parameter decay update subunit is used to dynamically adjust the weights based on feedback counts. In traditional text analysis, feature word weights are usually static or updated only based on corpus statistics, while this embodiment introduces an exponential decay mechanism based on expert experience. The parameter decay update subunit uses an exponential decay model to calculate feature word weights. Real-time weights The specific calculation formula is defined as follows: ; The parameter decay update subunit updates via decay coefficient. Control the decay rate of feature word weight values. The value is configured by the system administrator in the end-to-end business management module 50, and is usually set between 0.1 and 0.5. When When the setting is large, the weight value decreases faster with increasing feedback frequency; a small number of verification checks can quickly reduce the weight of relevant feature words. When the setting is smaller, the weight reduction process is smoother.

[0072] Through the above mechanism, for standard terms such as "concrete strength grade" and "subgrade compaction degree" that must be cited in bidding documents and tender documents and cannot be modified, as the number of review sessions increases, their corresponding... The value will accumulate continuously, leading to its weight. It rapidly approaches zero according to an exponential law. The cosine similarity formula is then applied in subsequent operations. During calculation, the contribution of these high-frequency standard words to the total similarity score is mathematically suppressed, thus enabling the algorithm to automatically ignore standard template content and focus on abnormal similarities in non-standard content such as construction plans and project management. This process does not require manual maintenance of a complex stop word list, but rather achieves numerical fitting of algorithm parameters to industry characteristics through actual review behavior.

[0073] See attached document Figure 5 The expert review and dynamic decision-making module 40, serving as the human-computer interaction hub and final decision-making execution mechanism of the entire system, is communicatively connected to the intelligent logic verification and objective review module 20 and the violation identification and adaptive feedback module 30. The expert review and dynamic decision-making module 40 does not simply display all review data; instead, it is configured to execute anomaly-driven differentiated review logic and possesses automated candidate queue reconstruction capabilities. Specifically, the expert review and dynamic decision-making module 40 includes a differentiated interaction control unit parameter decay update subunit and a dynamic supplementary operation unit parameter decay update subunit.

[0074] The differentiated interaction control unit parameter decay update subunit filters routine compliance items, pushing only potentially risky items requiring review to the review expert's terminal. This subunit incorporates early warning filtering logic that monitors the calculation results of preceding modules in real time. Specifically, the differentiated interaction control unit parameter decay update subunit only updates when it detects a logical verification result. (i.e., resource conflicts exist) or similarity calculation results (i.e., similarity exceeds a preset threshold) When an anomaly warning is triggered, an anomaly warning record is generated and rendered to the expert review interface. For scoring items that do not trigger a warning, the differential interactive control unit parameter decay update subunit is marked as passed by default, but experts still retain the right to actively review them.

[0075] During the review and decision-making process, the parameter decay update subunit of the differentiated interactive control unit executes the highest priority judgment logic, establishing the veto power of human decision-making over the algorithm results. The preliminary review result vector automatically calculated by the system is defined as follows: The judgment result vector entered by the expert on the review interface is: The differential interaction control unit parameter decay update subunit generates the final decision result based on the following logic. : ; In the formula, It includes objective scores and compliance markers derived from the system's algorithm; The initial state is NULL (empty value); when an expert performs a "veto", "modify score", or "confirm violation" operation, The specific numerical value or status code is written into the input. The formula above indicates that once the system receives a valid expert input instruction, regardless of the system's original calculation result, the final output will have legal validity. All are forcibly anchored to This mechanism ensures that the review results comply with the legal requirements regarding the "reviewer responsibility system," while also utilizing the feedback mechanism in the aforementioned embodiments to correct the algorithm model.

[0076] The dynamic replacement operation unit's parameter decay update subunit handles ranking changes caused by bidders being deemed invalid. In traditional electronic bidding systems, removing the top candidate typically requires manually recalculating the scores of all subsequent bidders and manually adjusting their rankings. However, the dynamic replacement operation unit's parameter decay update subunit is equipped with a triggered queue reconstruction algorithm. This subunit maintains a real-time updated queue of winning bidders in memory. In the initial state Sort by overall score in descending order, among which The number of candidates specified in the tender documents (usually 3).

[0077] When the parameters of the differentiated interaction control unit decay, the final decision result of the updated subunit output is obtained. Determine a candidate in the current queue When it is an "invalid mark" or "rejected mark" (i.e., a validity mark) (If the value is flipped to 0), the dynamic replenishment operation unit parameter decay update subunit immediately responds to this state change signal and executes an automated replenishment process.

[0078] First, the parameter decay update subunit of the dynamic replenishment operation unit will... From queue Remove from the list; then, the unit iterates through the remaining set of valid bidders. Retrieve items that are not currently in the queue. And the bidder with the highest overall score Finally, the parameter decay update subunit of the dynamic replenishment operation unit will... Insert into the queue and update the set based on the overall score. Execute the quicksort algorithm to generate a new queue of winning candidates. This process is automatically triggered and calculated by the system's backend logic, replacing the traditional method of manually re-summarizing scores and adjusting rankings, thus ensuring the accuracy of the calibration results.

[0079] See attached document Figure 6 In this embodiment, the full-process business management module 50 serves as the central scheduling mechanism of the system. It establishes communication connections with the full-domain dynamic data support module 10, the intelligent logic verification and objective review module 20, the violation identification and adaptive feedback module 30, and the expert review and dynamic decision-making module 40 to achieve time-series control and parameter configuration throughout the entire lifecycle of the bidding process. The full-process business management module 50 does not directly participate in specific scoring calculations; instead, it coordinates the operational order of various functional modules by maintaining a deterministic finite state machine. Specifically, the full-process business management module 50 includes a parameter configuration interaction unit and a process status control unit.

[0080] The parameter configuration interaction unit is configured to provide a digital parsing and parameter setting interface for the tender documents. During the project initiation phase, the parameter configuration interaction unit receives key control parameters input by the tenderer and distributes them to the corresponding calculation modules.

[0081] Specifically, the parameter configuration interaction unit receives and stores the planned duration time window of the current bidding project. This parameter is transmitted to the intelligent logic verification and objective review module 20 for state mutual exclusion verification. Simultaneously, the parameter configuration interaction unit allows users to adjust the physical transfer limit speed threshold based on the geographical characteristics of the project location (e.g., plains, mountains). And adjust the similarity weight decay coefficient used for violation identification according to industry review practices. .

[0082] In addition, the parameter configuration interaction unit is also responsible for defining the set of review items, dividing the review methods in the bidding documents into a set of intelligent review items. Set of non-intelligent review items ,in Associated with the system's automatic calculation logic, This will then link to the manual scoring interface.

[0083] The process status control unit is used to maintain the system's operational status and ensure that business processes comply with legal and regulatory procedures. The process status control unit defines the system's set of business process states. as follows: ; In the formula, This indicates the configuration state, corresponding to the stage of preparing and setting parameters for the tender documents; This indicates the bidding status, corresponding to the bid submission stage; This indicates the review status, corresponding to the bid evaluation stage after the bid opening; This indicates a public announcement phase, corresponding to the stage after the successful bidder has been determined.

[0084] The process status control unit performs status switching based on a time-triggered mechanism or external command signals. When the system is in... In this state, the system locks the calculation module and only grants read / write permissions to the parameter configuration interaction unit. Once the tender announcement is published, the process status control unit switches the status to... At this point, the system opens the data interface of the full-domain dynamic data support module 10 to receive the set of structured data pointers submitted by the bidder. and a commitment letter regarding the authenticity of the data Until the preset deadline is reached. .

[0085] Response to the bid closing time Upon receiving the signal, the process status control unit automatically switches the system status to [normal]. In this state, the process status control unit activates the calculation modules sequentially according to the preset dependencies: first, it triggers the intelligent logic verification and objective review module 20 to perform objective score calculation and logic conflict verification; then, it triggers the violation identification and adaptive feedback module 30 to perform similarity analysis; and finally, it activates the human-computer interaction interface of the expert review and dynamic decision-making module 40.

[0086] The process status control unit monitors the expert review progress in real time, and only allows the review to end after the expert review and dynamic decision-making module 40 outputs confirmation signals for all items. When the review end command is triggered, the process status control unit locks the status to [state missing]. In this state, the system solidifies the final ruling. The system generates an immutable bid evaluation report and publishes the list of successful bidders through an external interface. During this process, any requests to modify the original data are rejected by the system, thus ensuring the seriousness and traceability of the review results. Through [the following text is incomplete and requires further context: "through [the following text is incomplete and requires further context: "to generate an immutable bid evaluation report and publish the list of successful bidders through an external interface. During this process, any requests to modify the original data are rejected by the system, thus ensuring the seriousness and traceability of the review results."] With strict flow control of each state, this system enables the compliant implementation of technical solutions in actual business scenarios and prevents human factors from illegally interfering with the review process.

Claims

1. An intelligent auxiliary evaluation system for bidding and tendering based on artificial intelligence, characterized in that, include: The full-process business management module, as the central scheduling mechanism of the system, is configured to maintain the set of business process states and distribute algorithm control parameters, including attenuation coefficients, to other modules, and coordinate the time-series flow of each module between the configuration state, bidding state, review state and public announcement state. The full-domain dynamic data support module, controlled by the full-process business management module, accesses multi-dimensional raw data sources and generates a set of structured pointers with unique digital signatures. At the same time, it outputs the real-time status of resources through the full-domain state machine that monitors and maintains the core resources. The state space defined by the full-domain state machine includes at least idle state, locked state, under construction state, and maintenance state. The full-domain dynamic data support module responds to external event signals such as winning bid notification or completion filing to trigger state transition, thereby outputting and updating the real-time status of resources in real time. The intelligent logic verification and objective review module, through its built-in logic conflict verification network, invokes the real-time status of the resources to perform multi-dimensional logic conflict deduction on the bidding data based on state mutual exclusion and spatiotemporal physical constraints, and outputs logic conflict verification results. The logic conflict verification network includes a cascaded state retrieval and analysis subunit and a spatiotemporal constraint calculation subunit. The state retrieval and analysis subunit performs state mutual exclusion verification at the administrative compliance level based on the real-time status of the resources, and the spatiotemporal constraint calculation subunit performs spatiotemporal logic conflict verification at the physical feasibility level for resource scheduling between different projects based on the physical transfer limit speed threshold distributed by the full-process business management module. The violation identification and adaptive feedback module extracts unstructured bidding text from the multidimensional original data source, performs semantic similarity calculation based on the feature vector space, and outputs the similarity calculation results. In response to the algorithm control parameters, the expert feedback confirmation signal is captured, and the feature word weights are updated through a dynamic weight correction mechanism; The expert review and dynamic decision-making module aggregates the logical conflict verification results and the similarity calculation results, and pushes abnormal warning items only to experts through differentiated review logic; The system collects data from human decision-making to generate the final decision, feeds back expert feedback confirmation signals to form a learning feedback loop, and triggers the automatic reconstruction of the candidate queue when an invalid label is generated. 2.The intelligent bidding evaluation system based on artificial intelligence according to claim 1, characterized in that, The structured pointer generation unit in the global dynamic data support module is configured as follows: In the review state, the system compares the verification fingerprint of the structured pointer set with the digest value of the read data using a hash algorithm. After verifying consistency, it directly retrieves the source data from the main database for rendering. 3.The intelligent bidding evaluation system based on artificial intelligence according to claim 1, characterized in that, The specific logic for performing spatiotemporal logic conflict verification in the spatiotemporal constraint calculation subunit is as follows: Calculate the physical path distance and available time window between the previous project and the current project; When the available time window is less than or equal to zero, the physical transfer speed requirement is determined to be infinite, and a physical conflict indication result is generated. When the available time window is greater than zero, calculate the physical transfer speed requirement value and compare it with the physical transfer limit speed threshold. If it exceeds the threshold, generate a physical conflict indication result. The logical conflict verification network ultimately generates a solidified chain of logical contradiction evidence based on the physical conflict indication results.

4. The intelligent bidding evaluation system based on artificial intelligence according to claim 1, characterized in that, The specific methods of adaptive evolution of the algorithm include: The TF-IDF algorithm is used to construct the feature vector of unstructured bid text; Upon receiving the expert feedback confirmation signal, the weight values ​​of the feature words confirmed as reasonable citations are reduced using a dynamic weight correction mechanism. In subsequent similarity calculations, cosine similarity is calculated based on the updated feature word weights.

5. The intelligent bidding evaluation system based on artificial intelligence according to claim 1, characterized in that, The dynamic weight correction mechanism adopts an exponential decay formula, wherein the decay rate of the feature word weight is controlled by the decay coefficient configured by the full-process business management module, and the decay magnitude is positively correlated with the cumulative number of expert feedback confirmation signals.

6. The intelligent bidding evaluation system based on artificial intelligence according to claim 1, characterized in that, The specific differential review logic is as follows: Built-in warning filtering rules generate abnormal warning records and highlight them to the expert review interface only when a logically contradictory evidence chain is received or the similarity calculation result exceeds the threshold; for scoring items that do not trigger warnings, the objective scoring calculation result of the system is accepted by default.

7. The intelligent bidding evaluation system based on artificial intelligence according to claim 1, characterized in that, The expert review and dynamic decision-making module is also configured to perform the highest priority judgment: When a valid expert input instruction is received, the final decision result is forcibly anchored to the value or status corresponding to the expert input instruction, overriding the system's automatic calculation result. If the final decision determines that the current candidate's bid is invalid, a dynamic replacement operation is triggered, invalid nodes are automatically removed, and the remaining valid bidders are reordered based on their comprehensive scores.

8. The intelligent auxiliary evaluation system for bidding based on artificial intelligence according to claim 1, characterized in that, The full-process business management module coordinates the execution order of various functional modules by maintaining a deterministic finite state machine, specifically as follows: In the bidding process, the data access interface is open, but the violation identification and adaptive feedback modules are locked. In the review state, the intelligent logic verification and objective review module is activated to perform objective score calculation and logic conflict verification, and the violation identification and adaptive feedback module is triggered to perform similarity analysis. In the public announcement phase, the final decision is solidified, and an unalterable bid evaluation report is generated.

Images