CN-122022710-A - Intelligent examination system and method oriented to purchasing

Abstract

The invention provides an intelligent examination system and method for purchasing, wherein the system comprises an account management module, a rule management module, a dual-drive compliance verification module, a scoring table detection module, a specific category annotating module, a sensitive word management module and a standing book management module, wherein the dual-drive compliance verification module is used for completing core compliance detection of purchasing files by combining business rules and a knowledge base, outputting a problem list and a modification suggestion, supporting manual correction error suggestion, and the standing book management module is used for automatically recording detection process data, supporting self-defined Excel derivation and facilitating follow-up tracing and statistics. The system solves the problems of low manual detection efficiency, insufficient accuracy and single function of the traditional tool and lack of intelligent reasoning and self-defining capability in the conventional government purchase file compliance detection.

Inventors

• TIAN SHENGLI
• Liao Yaqiao
• MA JINGYUAN
• LUO MINGKANG
• Li Zhuomiao
• Zeng Yingyan
• ZOU XIAOHONG
• WU WENHUI
• WANG SHINA
• ZHAO JINGYING
• YE LUGAO
• ZHU HUAN
• WU YUGE
• WEN YIFENG
• WANG HONG
• OuYang Jinjie

Assignees

• 广州市政府采购中心
• 广东南方信息安全产业基地有限公司

Dates

Publication Date

20260512

Application Date

20251218

Claims (10)

1. 1. An intelligent audit system oriented to purchasing is characterized by comprising the following specific steps: The account management module is used for realizing account system construction, login authentication, intelligent agent platform access and hierarchical authority management, and ensuring that different roles have corresponding operation authorities; The rule management module is used for realizing rule template generation, rule uploading, paging inquiry, deletion and hierarchical classification management, providing rule support for compliance verification, and supporting custom multi-purchasing package parameter transparent transmission and mixed verification; The dual-drive compliance verification module is used for combining the business rules and the knowledge base dual-drive to finish core compliance detection of the purchase file, outputting a problem list and a modification suggestion and supporting manual correction error suggestion; the scoring table detection module is used for finishing scoring logic detection, scoring object detection, semantic detection and related content extraction and arrangement aiming at the scoring table in the purchasing file, so as to ensure compliance of the scoring table; Specific category annotating module for completing exclusive compliance detection and annotating for different purchase categories and purchase package types; The sensitive word management module is used for realizing addition, deletion and labeling of sensitive words, supporting specified detection positions and avoiding illegal expression in purchase files; and the ledger management module is used for automatically recording detection process data, supporting the self-defined Excel to be exported and facilitating the follow-up tracing and statistics.
2. 2. The intelligent examination system for purchasing is characterized in that the account management module completes login authentication and intelligent agent platform access based on OAuth2.0 protocol, adopts RBAC model design hierarchical rights, stores account information, role information and rights mapping relation through a database, and supports flexible configuration of rights details.
3. 3. The procurement-oriented intelligent review system of claim 1 characterized by the logic of implementation of the rule management module comprises: the rule template is generated by analyzing a standard document provided by a purchasing center, automatically extracting standard document elements and generating a standardized rule template; hierarchical classified storage, namely managing rules by adopting a three-level folder structure, and realizing quick retrieval of the rules through database index optimization; and the mixed verification is that the design rule engine supports the uploading of the custom rules by users, links with the built-in general rules of the system, preferentially executes the custom rules during verification, automatically triggers the general rules in the uncovered scene, and ensures the comprehensiveness of the verification.
4. 4. The intelligent inspection system for purchasing is characterized in that the dual-drive compliance verification module comprises a business rule verification sub-module and a knowledge base reasoning verification sub-module, wherein the business rule verification sub-module is used for universal rule matching and realizing multiple matching logics, key information in a purchasing file is extracted through a text analysis technology and is compared with business rules in a rule management module to generate a verification result, a BERT pre-training model is adopted for semantic matching, the knowledge base reasoning verification sub-module is used for integrating a law regulation library, a purchasing file compiling negative list, a historical case library, a certificate knowledge base, an item category knowledge base and an industry category knowledge base, building a correlation relationship through a knowledge graph technology, carrying out logic reasoning based on the knowledge base, automatically correlating rules that an admission class certificate cannot be scored when the admission class certificate appears in a scoring table as a scoring item, outputting a violation prompt, and intelligently generating a modification suggestion based on a case similarity algorithm by comparing the purchasing file with a high-quality case base.
5. 5. The intelligent, procurement-oriented screening system of claim 1 characterized by the dual-drive compliance verification module further comprising a manual correction module for providing a "correction portal" on the test results page that a sponsor can modify the system-generated error advice, the correction records automatically synchronized to a knowledge base, optimizing subsequent inference accuracy by machine learning algorithms.
6. 6. The intelligent purchasing-oriented screening system of claim 1, wherein the logic for implementing the score table detection module comprises: The scoring logic detection, namely extracting scoring values, scoring points and withholding point data of the scoring factors in the scoring table through a table analysis technology, and automatically checking logic relations; The method comprises the steps of detecting a review object, associating purchasing demand content, checking the correspondence of the scoring object, automatically filtering objects which are not required to be detected in the same category project experience and comprehensive credit evaluation, and reversely checking the position of the review object in purchasing demand, and generating exclusive endorsements aiming at the problems of inconsistent sequence numbers and object deletion; Semantic detection, namely analyzing scoring directivity and scoring principle through an NLP technology, and identifying illegal expression; And (3) extracting contents, namely automatically extracting data ' and ' qualification requirements ' required by suppliers in a scoring table, generating a comparison table, filling the data requirements and the prompting contents of the suppliers in the table through a large model, supporting the increase, decrease and deletion of notes of common problems, identifying asterisk clauses and triangular clauses, and outputting a structured table.
7. 7. The intelligent examination system of claim 1, wherein the sensitive word management module comprises logic for establishing a sensitive word library, maintaining the word library by a user through a background management interface, rapidly positioning the sensitive word according to the position appointed by the user during detection, marking the illegal position and outputting a reminder.
8. 8. The intelligent examination system for purchasing according to claim 1, wherein the ledger administration module is implemented with logic that, after the detection is completed, the system automatically writes the detection data into a database, designs a configurable export template, and the user can select the field to be exported, and generates an Excel file through POI technology to support batch exporting and single item exporting.
9. 9. The intelligent examination method for purchasing is characterized by comprising the following steps: S1, data input and initialization, wherein a user logs in a system through an account management module, uploads a purchase file to be detected, selects project related parameters, and the system automatically loads rules and a knowledge base of a corresponding module; s2, loading rules and a knowledge base, wherein a rule management module calls corresponding rules in a three-level folder according to project parameters, and a knowledge base reasoning and checking sub-module synchronously loads related legal regulations and case base data; S3, multi-module collaborative detection, namely a dual-drive compliance verification module, a comment table detection module, a specific category annotation module and a sensitive word management module work in parallel to respectively finish business rule verification, knowledge base reasoning, comment table logic and object detection, specific scene annotation and sensitive word annotation, and the detection results of the modules are summarized to a central processing unit in real time; s4, result integration and suggestion generation, wherein the central processing unit performs de-duplication and classification on detection results of all modules, and generates targeted modification suggestions based on historical cases and rules in a knowledge base to form a preliminary detection report; S5, manual correction and result confirmation.
10. 10. The intelligent review method for purchasing according to claim 9, wherein in step S3, the specific workflow of the multi-module collaborative detection is as follows: S31, a double-drive compliance verification module adopts a double-drive mode of a rule engine and a machine learning model, when detecting compliance of contract terms in government purchasing files, the rule engine automatically matches key contents such as contract payment modes, performance deadlines and the like in the files according to preset government purchasing rules, judges whether hard violations exist, such as that a payment period exceeds a legal upper limit, and simultaneously, the machine learning model identifies hidden risks which are difficult to define by rules such as a fuzzy acceptance standard by analyzing a large number of historical compliance and violation file data, and the combination of the rule engine and the rule ensures compliance detection without dead angles; S32, taking a government purchasing bid-drawing scoring table as an example, the scoring table detection module automatically identifies each index in the scoring table, and detects whether the scoring standard has contradiction and whether the scoring weight setting accords with industry convention through a built-in logic verification algorithm; s33, a specific class annotating module carries out targeted annotating according to a preset rule aiming at a special scene in a government purchase file; S34, a sensitive word management module constructs a dynamically updated sensitive word library containing legal and legal forbidden words and industry sensitive expressions, when government procurement files are detected, when words which violate the fair competition principle appear in the files or domain discrimination expressions are related, the module immediately marks and highlights the words and the industry sensitive expressions, and simultaneously provides replacement suggestions.

Description

Intelligent examination system and method oriented to purchasing Technical Field The invention relates to the technical field of intelligent text examination, in particular to an intelligent examination system and method for purchasing. Background In the government purchasing field, the compliance of purchasing files directly affects the fairness, legitimacy and efficiency of purchasing activities. At present, the compliance detection of purchase files mainly comprises the following two modes: (1) Manual detection mode The business sponsor needs to check the contents of the clauses, the scoring list, the purchasing demands and the like of the purchasing file one by contrasting the business rules, the related laws and regulations, the purchasing file compiling negative lists, the historical cases and the like issued by the purchasing center. The specific operation flow is that A acquires purchasing files, B combing key information, such as project attributes, review factors and the like, C compares against a compliance standard table, D checks compliance of the review table, E checks specific category, F checks context, and G checks sensitive words. Manual detection has the limitation that in the manual detection process, the application of rules depends on the experience of the sponsor. With the complexity of the service, the manual detection has the problems of low exposure efficiency (the detection of complex files takes several hours), difficult guarantee of accuracy (easily influenced by factors such as fatigue), insufficient rule uniformity (different people understand and execute with deviation), lack of intelligence (the analysis cannot be inferred and the accurate suggestion can be given, and the recording process is tedious and easy to miss), and the like. (2) Existing simple compliance detection tool The product constitution and function can only realize the fixed content verification of a single dimension. The application limitation is that the uploading of the custom rule is not supported, the multi-module linkage detection (such as the cooperation of rule table management and scoring table detection) is not supported, and the reasoning capability based on a knowledge base is not provided. It is difficult to adapt to complex scenarios involving huge rule systems and multidimensional requirements for government procurement file compliance detection. Disclosure of Invention Aiming at the defects of the prior art, the invention provides an intelligent examination system and method for purchasing, which solve the problems of low manual detection efficiency, insufficient accuracy and single function of the traditional tool and lack of intelligent reasoning and self-definition capability in the conventional government purchasing file compliance detection. The intelligent inspection system for purchasing comprises an account management module, a rule management module, a dual-drive compliance verification module, a comment table detection module, a special class annotating module, a sensitive word management module and an automatic expression and verification module, wherein the account management module is used for realizing account system construction, login authentication, intelligent agent platform access and hierarchical authority management, ensuring that different roles have corresponding operation authorities, the rule management module is used for realizing rule template generation, rule uploading, paging inquiry, deletion and hierarchical classification management, providing rule support for compliance verification, supporting custom multi-purchasing package parameter transparent transmission and hybrid verification, combining business rule and knowledge base dual-drive, completing core compliance detection of purchasing files, outputting problem list and modification advice, supporting manual correction error advice, aiming at a comment table in the purchasing files, finishing score logic detection, comment object detection, semantic detection and related content extraction and arrangement, ensuring compliance of the comment table, the special class annotating module is used for different purchasing classes and purchasing package types, completing proprietary compliance detection and annotation, the sensitive word management module is used for realizing sensitive addition, deletion and annotation, supporting specified detection positions, automatically defining and self-defining the data of the purchasing management platform, and facilitating the self-defining process. Preferably, the account management module completes login authentication and intelligent agent platform access based on an oauth2.0 protocol, adopts an RBAC model to design hierarchical rights, stores account information, role information and rights mapping relation through a database, and supports flexible configuration of rights details. Preferably, the logic for implementing the rule management module includes: the rule tem