A dynamic rule-driven multi-domain review method and system

Abstract

The application discloses a kind of dynamic rule driven multi-field review method and system, belong to intelligent review technical field;Including: dynamic rule engine configuration, expert management configuration, automatic grouping matching, expert online review and general meeting voting;Realize cross-project type, cross organization department, cross business scene full-dimension coverage.No matter is talent selection, scientific research project technical evaluation, or title review, can be based on the same platform to complete the dynamic construction and flexible adjustment of differentiating review process, effectively improve the review efficiency and scientificity, provide standardized, intelligent technical solutions for multi-agent, multi-scene collaborative review business;Solve the problems of the current review system in process rigidification, weak function, extensive management, transparency loss and security vulnerability.

Description

Technical Field

[0001] This invention belongs to the field of intelligent review technology, specifically relating to a dynamic rule-driven multi-domain review method and system. Background Technology

[0002] The review process is becoming increasingly complex, with a constant stream of innovative achievements across disciplines and fields. This places higher demands on the adaptability and professionalism of review systems, and the need for fairness, efficiency, and accuracy in the review process is becoming more urgent. Traditional, relatively rigid review systems are struggling to effectively cope with this complexity. The current review landscape exhibits three core characteristics: First, significant differences exist across fields, with fundamentally different review processes, indicator systems, and rule logic across different domains; second, rules are becoming more dynamic, with frequent demands for policy adjustments and review standard optimization, leading to a continuously shortening rule iteration cycle; and third, data is becoming increasingly multimodal, with review materials expanding from single text to multimodal data such as images, audio, and structured reports, requiring cross-system collaborative verification. Expert reviews in complex application scenarios face numerous challenging issues. Traditional review system architectures are outdated and unable to adapt to dynamic and ever-changing business needs. From a macro perspective, the digital governance transformation of existing systems is lagging behind, developed based on a monolithic architecture, resulting in outdated system architecture, high module coupling, and the need to refactor core code for functional expansion; simultaneously, the lack of technical standards necessitates customized interface development for cross-departmental system integration. From a micro perspective, existing technologies face the following bottlenecks:

[0003] 1) Rigid rules and poor adaptability: a) Most existing systems adopt pre-set and fixed review processes, which make it difficult to flexibly adjust the review steps, order of links and rules according to the characteristics of different types of projects, review conditions and participating departments; b) The fixed process cannot flexibly adapt to different types of review needs and is difficult to respond to policy adjustments in a timely manner; c) The types of expert permissions are limited, which restricts the deep integration of professional perspectives.

[0004] 2) Weak cross-domain review capabilities and missing functions: a) Existing systems struggle to support multi-domain review environments. When review objects involve multiple disciplines or interdisciplinary fields, the system often lacks effective mechanisms to handle different review processes; b) Lack of multi-project parallel processing capabilities limits review efficiency; c) Different scoring standards exist across different domains. Most systems lack multiple types of calculation engines, and score calculation relies on manual methods, leading to increased costs; d) The scoring mode is singular, supporting only fixed formats and failing to implement multi-dimensional weighted scoring; e) Severe information barriers between departments create data silos, making cross-departmental collaboration difficult and inefficient; f) Lack of custom query and export functions limits data query and export capabilities.

[0005] 3) Disorder in expert management: a) Expert resources are unstructured and their profiles are vague, with key information such as expertise, historical experience, and conflict of interest missing, making it impossible to accurately match review needs; b) There is a lack of an automatic matching mechanism, and matching relies on human experience, which is inefficient and difficult to guarantee fairness; c) Key mechanisms such as double-blind review are missing, affecting the credibility of the review.

[0006] 4) Blind spots in process visualization: a) The review progress is not transparent and the status of key nodes is difficult to track; b) The conference projection cannot be dynamically presented, and the decision-making lacks real-time data support.

[0007] 5) Insufficient security: Weak data backup and desensitization mechanisms, lack of private deployment options and operational auditing capabilities, posing risks of data leakage and loss of process control. Summary of the Invention

[0008] This invention provides a dynamic rule-driven multi-domain review method and system. It breaks down rigidity through a dynamic process engine, supports customizable configuration of review paths and expert permissions for multiple scenarios, and achieves minute-level response to policy changes. A multi-dimensional intelligent review hub empowers complex business processes, integrating multi-format scoring, automatic weighted calculation, and rule verification, supporting parallel processing of multiple projects. A comprehensive expert database and intelligent profiling achieve precise matching, centralized resource management, and the construction of a multi-dimensional capability model, while embedding a double-blind selection mechanism to ensure fairness. Security is ensured through data backup, data anonymization, private deployment, and complete audit logs. This invention solves the problems of current review systems, including rigid processes, weak functionality, lax management, lack of transparency, and security vulnerabilities.

[0009] To achieve the above-mentioned technical objectives, the present invention is implemented through the following technical solution:

[0010] A dynamic rule-driven multi-domain review method includes:

[0011] S1: A dynamic rule engine that adapts flexibly to different scenarios and policies. Dynamic configuration includes: review process, review stages, scoring methods, scoring dimensions, rule parameters, and configuration confirmation. It allows for the rapid generation of review processes and rules that meet business needs through visual configuration.

[0012] S2: Expert Management. Through expert information entry, tag system construction, dynamic profile updating, and expert status management, accurate profiles are constructed and dynamically updated to achieve centralized management and accurate profiles of expert resources. Based on accurate expert profiles and resource management, a multi-dimensional expert capability profile model is constructed to ensure the compliance of expert qualifications.

[0013] S3: Automatic grouping and matching, through the following: conflict rule setting, automatic matching triggering, grouping parameter configuration, manual adjustment and optimization, invitation information sending, expert participation feedback, judgment and rematching, and matching result confirmation, achieves intelligent recommendation and flexible adjustment, which is used for accurate expert grouping and avoids conflicts of interest;

[0014] S4: Online expert review enables efficient online expert review through viewing review materials, submitting multi-dimensional scores, automatic score calculation, multi-condition sorting and recommendation, filling in review comments, and confirming results. It supports multi-dimensional scoring and automatic calculation to ensure the standardization of the review process.

[0015] S5: Conference voting, through the preparation of voting items, allocation of voting rights, setting of anonymous voting, real-time visualization and confirmation of voting results, to achieve transparency in the voting process and real-time presentation of results.

[0016] Preferably, the review process settings in the dynamic rule engine settings are as follows: Enter the backend configuration center and configure the review process according to different special types, including at least: talent review, job review, professional title review, and project review; select an existing process template according to the review type and then manually optimize and adjust it, or directly customize the configuration; set user roles according to review needs, including: review administrator and conference host.

[0017] The selection of the review stage is specifically as follows: select the stage to be enabled at the process node and configure it differently according to the review type;

[0018] The scoring method selects and configures different scoring modes; then, different scoring modes are set for different dimensions of the same project.

[0019] The scoring dimension configuration settings include scoring dimensions such as basic conditions, budget rationality, and project feasibility, setting the weight of each dimension and linking it to the scoring engine;

[0020] The rule parameter configuration process verifies the rules and completes the process transmission according to the configured rules; and directly modifies the rule parameters when the policy is adjusted.

[0021] Preferably, the expert information entered in the expert management system includes at least: basic information and qualification certificates; expert information is entered through two methods: manual entry or automatic import from a form template.

[0022] After adding expert domain tags, the system automatically extracts keywords from the expert's historical review records to supplement the tags, constructing the tag system according to the following method:

[0023] The tag building model, which combines multi-source fusion with human-machine collaboration, includes a three-layer tag structure. ：

[0024] The first layer consists of basic attribute tags, including research field, professional title, academic title, institution, and region.

[0025] The second layer consists of dynamic behavioral tags, including the number of reviews, review status, average scoring tendency, response speed tag, and return rate tag.

[0026] The third layer consists of quality assessment labels, including effective scoring rate, score adjustment rate, and rejection rate.

[0027] The automatic tag extraction method is as follows:

[0028] a) Historical review record mining

[0029] Extract all applications reviewed by experts from historical review records, use the TF-IDF algorithm to extract research field keywords from the applications, calculate keyword frequencies, and generate a candidate tag set;

[0030] b. Domain similarity calculation

[0031] Similarity (expert, application) = Σ(expert label i ∩ application field j) / | set of expert labels|. When the similarity > threshold θ, the application field is added as a candidate label.

[0032] c. Label confidence score

[0033] Confidence level = (Number of historical reviews × 0.4) + (Review quality score × 0.3) + (Domain matching degree × 0.3)

[0034] in:

[0035] Review quality score = (1 - Return rate) × (1 - Rejection rate) × Valid score rate

[0036] Domain matching score = Number of reviews in this domain / Total number of reviews

[0037] d. Manual review mechanism

[0038] Labels with a confidence level > 0.8 are automatically added;

[0039] Tags with a confidence level of 0.5 or less and a confidence level of ≤0.8 are pushed to the administrator for review;

[0040] Labels with a confidence level ≤ 0.5 are discarded;

[0041] Expert Profile Building Model

[0042] A star-shaped model is used to construct a multi-dimensional profile of experts. The basic information of the profile includes: a unique expert identifier, a set of research field tags, professional title, a list of titles, an institution ID, and provincial / international identification.

[0043] The dimensions of the dynamic profile include: cumulative number of reviews, number of reviews this year, average score, score variance, average response time, return rate, rejection rate, effective score rate, activity score, and review quality score.

[0044] The dynamic profile update trigger conditions include at least: experts submitting new review records, administrators adjusting or rejecting expert scores, experts voluntarily returning review tasks, and scheduled update tasks;

[0045] The update method is as follows: New portrait value = α × Old portrait value + (1-α) × New data, where α is the smoothing coefficient, with a default value of 0.7;

[0046] The newly generated expert profiles are evaluated for quality, including: overall quality score and activity score;

[0047] The overall quality score:

[0048] qualityScore = w1 × Response Speed ​​Score + w2 × Scoring Stability Score + w3 × Validity Score + w4 × (1 - Return Rate) + w5 × (1 - Rejection Rate)

[0049] The above response speed score = 1 / (1 + log(average response time / h))

[0050] Rating stability score = 1 / (1 + rating variance)

[0051] Weight vector [w1,w2,w3,w4,w5]=[0.2,0.25,0.25,0.15,0.15]

[0052] The activity score algorithm is as follows: activityScore = (Number of reviews this year / Average number of reviews in the system) × (1 - Number of idle days / 365)

[0053] Idle days = Current date - Last submission date for review.

[0054] Preferably, the expert management system's participation status linkage management architecture is as follows:

[0055] a. State Machine Model: The expert review lifecycle is managed using a Finite State Machine (FSM) model, comprising four state phases: Draft: The application is assigned to experts; Evaluated: Experts complete their review after receiving the application; Submitted: The reviewed application is submitted; Returned: The administrator returns the submitted application to the expert. The constraints between these states are as follows:

[0056] Drafts must be scored and commented on during the evaluation stage; drafts must meet the scoring rules before being submitted; submitted drafts can only be returned by the administrator.

[0057] b. Concurrent state synchronization algorithm, including update review state consistency and state consistency guarantee, wherein Java distributed lock is used to ensure update review state consistency; the state consistency guarantee is implemented as follows:

[0058] AtomicInteger is used to ensure the atomicity of the counter; optimistic locking is implemented at the database level using a version number field; and expert score update events are triggered by state changes.

[0059] c. Multi-project conflict detection algorithm, through time window overlap detection, is used to ensure that the same application can only be reviewed by one expert at a time, to prevent data conflicts.

[0060] d. Status linkage triggering mechanism, the process stages are as follows:

[0061] After an expert submits a review, an expert score update event is triggered. The system checks whether the application score has been updated and whether all experts have submitted their applications. Based on the check results, the administrator is notified.

[0062] When an expert returns a match for review, an expert review return event is triggered. The system deletes the matching result and notifies the administrator to reassign the match.

[0063] The administrator rejected the rating, the rating status was updated to unrated, the average review score was recalculated, and the expert profile was updated.

[0064] e. Intelligent schedule management algorithm, load balancing scheduling: To effectively balance the workload of review experts, the algorithm does not use a simple round-robin or random method when assigning review tasks. Instead, it introduces an allocatable metric to dynamically evaluate and prioritize the most suitable experts, as detailed below:

[0065] Allocability = (1 - Number of currently assigned tasks / Maximum number of tasks an expert can handle) × (1 - Time window overlap) × Expert qualification score

[0066] In the above formula, (1 - number of currently assigned tasks / maximum number of tasks an expert can handle): this part represents the load factor; the fewer tasks an expert currently handles, the closer its value is to 1, indicating that it has sufficient workload and higher priority, thus avoiding excessive concentration of tasks.

[0067] (1-Time Window Overlap): This part represents the time availability factor. The time window overlap is calculated by the proportion of the number of days that overlap between the review time windows of the expert's existing tasks and the review time windows of the tasks to be assigned to the total number of review days. The lower the overlap, the more idle the expert is during that time period. The higher the value of this factor, the higher the allocation priority.

[0068] Expert Qualification Rating: This section represents the quality factor; it is calculated based on factors such as the expert's professional field of expertise, past review quality, and reputation level to ensure that tasks are assigned to the most competent experts.

[0069] By integrating three core elements—load balancing, time conflict avoidance, and quality assurance—the algorithm can intelligently and dynamically select the "most available, most suitable, and most available" experts for each task to be reviewed, thereby achieving efficient, fair, and high-quality allocation of the overall review tasks.

[0070] Preferably, in the expert management, resources are centrally managed and precise profiles are created to construct a multi-dimensional capability model for verifying expert qualifications. The specific construction process is as follows:

[0071] Construct a multi-dimensional expert capability profile model, including the following core dimensions:

[0072] a. Research Field Codes: Adopting the Science and Technology Department's detailed classification coding system, supporting multi-field tags, and forming an expert professional capability map.

[0073] b. Unit Identifier (UnitId / UnitName): Used to implement unit mutual exclusion constraints in the double-blind mechanism.

[0074] c. Geographical Distribution Characteristics (in Yunnan): Differentiating between experts from within and outside the province to support a regional allocation strategy.

[0075] d. Job Title / Title: Constructing a dimension for evaluating expert authority

[0076] e. Dynamic load counter (ExpertCounter): Tracks the allocation of expert review tasks in real time to achieve load balancing.

[0077] Training methods

[0078] A hybrid training model combining rule-driven and heuristic optimization is adopted:

[0079] a. Core Matching Rules

[0080] Research Field Matching Rule: Achieves precise field alignment by calculating the intersection of the applicant's research field code and the expert's field label set;

[0081] Unit Mutex Matching Rule: In double-blind mode, it forces the filtering of experts from the same institution to ensure the impartiality of the review process;

[0082] Province Ratio: Experts from within and outside the province are allocated according to a preset ratio to meet policy requirements.

[0083] Preferably, in the automatic grouping and matching:

[0084] S1: Conflict rule settings, configure avoidance rules, and use a multi-dimensional expert capability profile model to automatically verify experts and remove conflicting personnel in this review;

[0085] S2: Automatic matching triggering. Based on the review requirements, the multi-dimensional expert capability profile model selects experts with a matching degree of ≥80% from the expert database that has completed the avoidance rule matching to generate a candidate pool.

[0086] S3: Grouping parameter configuration, setting the number of groups and the number of experts in each group. The administrator selects the grouping conditions, and the system automatically and evenly distributes experts. When there are enough experts in the candidate pool, they are selected from high to low matching degree to accommodate secondary matching. If there are not enough experts in the candidate pool, they are supplemented according to matching degree. Experts with a matching degree lower than the set matching degree are highlighted after grouping.

[0087] S4: Manual adjustment and optimization. Administrators can manually add or remove experts. After adjustment, the system will automatically re-verify the grouping conditions.

[0088] S5: Invitation Information Sending. After the initial expert list is generated, the administrator automatically generates a standardized notification template based on the system. The administrator can edit and supplement the notes. The notification method adopts a dual mode of SMS sending and in-system message push. The system calls the SMS interface to send SMS to the expert's mobile phone number, and at the same time pushes a notification in the "Message Center" of the system, which includes a link to jump to the evaluation feedback with one click. After sending, the sending status is displayed in real time. Failed numbers can be manually retried or the contact information can be replaced.

[0089] S6: Expert Participation Feedback. After receiving the notification, if an expert selects "Confirm Participation," the system automatically records the confirmation status and updates the expert's schedule and status synchronously. If the expert confirms "Unable to Participate," a reason must be selected, and the system automatically sends the reason to the administrator's backend and marks the expert as "Unable to Participate." Administrators can view the expert's response progress in real time. For experts who do not respond within a set time, the system automatically triggers a second reminder. Experts who do not respond to notifications within a set number of times are considered unable to participate by default.

[0090] S7: Rematching and matching: After the invitation and feedback process for the participants ends, the system automatically locks the experts who have been matched and the experts who cannot participate. For the vacant number of participants, repeat the above steps S2~S6 to match again until the matching is completed and the rules are met.

[0091] S8: Matching results confirmed. Check the final confirmation status of the group and officially enter the review stage.

[0092] Preferably, the online expert review process is as follows:

[0093] S1: Review materials viewing. After logging into the system, experts can go to "My Reviews" to view project materials, which support online annotation and note-taking.

[0094] S2: Multi-dimensional scoring submission. Scores are submitted according to the configured dimensions on the scoring interface. The corresponding mode is selected, and the system displays the scoring progress in real time.

[0095] S3: Automatic score calculation. After scoring is completed, the system automatically calculates the total score according to the preset weights and supports real-time viewing of score details.

[0096] S4: Multi-condition ranking recommendation. After the scoring is completed and submitted, experts can rank the scores based on the total score, the highest score in a single item, and the items that meet the criteria. The final ranking table can be exported.

[0097] S5: Fill in the review comments. Fill in the written comments for each dimension. The system provides comment templates for reference. After completing the comments, you can export the review comment document.

[0098] S6: Result confirmation. Based on the scores, rankings, and expert review opinions, select and confirm the "Proposed Recommendation" list and submit it to proceed to the conference voting stage.

[0099] Preferably, the general assembly voting includes:

[0100] S1: Voting item preparation. In the system backend, the administrator selects whether to vote in groups, whether to vote by topic, and whether all members should participate in the meeting, confirms the voting items, and sets the voting rules.

[0101] S2: Voting permission allocation, specifying the scope of participants in the vote, and sending voting notifications through the system;

[0102] S3: Anonymous voting setting, hiding voter information and only recording voting results and time; S4: Real-time visualization display, dynamically showing the vote share of each option during the voting process.

[0103] Supports filtering and viewing detailed data by group;

[0104] S5: Voting results confirmed. After the voting deadline, the system automatically summarizes the results, generates a voting report, and exports it.

[0105] Another objective of this invention is to provide a dynamic rule-driven multi-domain review system, comprising:

[0106] The dynamic rules engine module includes a review process configuration unit, which is used to configure the review process module according to different projects;

[0107] The review process involves selecting different stages based on the type of project being reviewed.

[0108] The scoring unit allows for the setting of corresponding scoring methods for different projects and dimensions;

[0109] The scoring dimension configuration unit sets the scoring dimensions and assigns a weight to each scoring dimension, which is then used to associate the scoring units.

[0110] The rule parameter setting unit allows you to set process rules at process nodes and corresponding steps, and complete the review process transfer according to the rules.

[0111] The confirmation unit confirms the configuration of each of the above units;

[0112] The expert management module includes an expert information entry unit, which allows users to enter basic information and qualification certificates of experts; and sets up multiple entry methods.

[0113] The tag system construction unit establishes a tag system from basic attribute tags, dynamic behavior tags, and quality assessment tags respectively;

[0114] The profile dynamic update unit builds expert profiles based on the tag system and dynamically updates the expert profiles according to the actual review status results.

[0115] The expert status management unit allocates experts based on factors such as multiple project conflicts, review status matching, and review schedules.

[0116] The grouping and matching module includes a conflict rule setting unit, which identifies whether there is a conflict of interest between experts and the project, and requests expert recusal.

[0117] The automatic matching trigger unit selects and matches suitable experts as candidates from among those who have been recused, based on the review requirements;

[0118] The grouping parameter configuration unit allocates the appropriate number of people from the candidate experts in a reasonable and balanced manner according to the appropriate number of groups and the appropriate number of people in each group.

[0119] In addition to the automatic grouping completed by the automatic matching trigger unit and the grouping parameter configuration unit, the administrator can also manually group the units. After manual grouping, the system will then perform grouping condition verification.

[0120] The invitation information sending unit sends invitation information to the preliminarily selected review experts and displays the information sending status; after receiving the invitation information, the unit receives feedback information from the experts.

[0121] After matching and confirming the expert's response, the participating experts are finally confirmed.

[0122] The online expert review module includes a review materials viewing unit, where experts can log in to the system and view the materials for the review projects they are responsible for.

[0123] The system features multiple scoring units that score project materials according to scoring configurations, and the scoring progress is displayed in real time.

[0124] The automatic score calculation unit completes the scoring, and the system automatically calculates the total score according to the preset weights.

[0125] The multi-dimensional ranking recommendation unit allows experts to rank items based on total score, highest score for a single item, and matching items after the scoring is completed and submitted. It also supports exporting the final ranking table.

[0126] The section for filling in review comments allows experts to enter their comments.

[0127] The conference voting module includes a voting item preparation unit, an administrator operation setting for the voting mode, and a setting for the voting rules;

[0128] The voting rights allocation unit sets the scope of personnel who can participate in voting;

[0129] An anonymous voting feature hides the voter's explicit information.

[0130] The display unit shows the actual voting results in real time.

[0131] The beneficial effects of this invention are:

[0132] 1) Significantly improved rule flexibility and multi-domain adaptability

[0133] It supports visual process configuration, allowing users to quickly customize or adjust the review process without code development, adapting to different types of review scenarios.

[0134] Through the dynamic rule management of the rule engine, review standards and workflows can be quickly adjusted according to needs, greatly improving the system's adaptability to review scenarios in different fields.

[0135] 2) Enhanced functionality and intelligent support capabilities

[0136] The system's functions cover the entire review process and have deep intelligent support;

[0137] It supports multiple scoring methods and automatic calculation, covering a variety of scoring modes such as ten-point system, hundred-point system, and grade system. It has a built-in automatic calculation engine that can automatically generate the total score, reducing the error of manual calculation.

[0138] Intelligent verification and logical control are improved by using a rule engine to verify material integrity and data logic, providing real-time feedback on errors and indicating the reasons, thus reducing the cost of manual review.

[0139] To achieve multi-dimensional and precise management, a multi-dimensional scoring framework is constructed, supporting the customization of indicator systems according to scenarios, and meeting the needs of complex reviews for refined scoring;

[0140] 3) Enhanced precision in expert management and improved fairness in review processes

[0141] Integrate scattered expert resources, construct expert profiles, and combine with search engines to enable rapid selection of experts based on needs;

[0142] Experts are automatically assigned based on expert profiles, reducing the bias of manual assignment.

[0143] Double-blind review is achieved through data encryption and access control isolation, which technically avoids human interference and ensures the fairness of the review process.

[0144] 4) Improved efficiency in data collaboration and ease of querying and exporting

[0145] A unified review platform integrates data from multiple departments, breaks down information barriers, and enables cross-departmental data flow.

[0146] It enables flexible querying and custom exporting, supports multi-condition combined queries, and is equipped with custom report export function to meet data analysis and archiving needs;

[0147] 5) Enhanced end-to-end visualization and real-time monitoring capabilities

[0148] By displaying the progress and node status of each stage in real time through dynamic flowcharts, the review progress is visualized, and managers can have a global grasp of the review dynamics.

[0149] The voting results of the conference are displayed in real time, and the voting process and results are presented in real time in the form of pie charts, etc., to improve the transparency of decision-making.

[0150] 6) The security system has been comprehensively strengthened.

[0151] The system strengthens security capabilities across the entire chain from data storage and transmission to operation, enabling sensitive information desensitization and data backup.

[0152] Supports private deployment to meet data localization requirements;

[0153] Log auditing records all operational behaviors and supports multi-dimensional traceability by operator, time, and type, ensuring data security and compliance. Attached Figure Description

[0154] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0155] Figure 1 This is a system structure block diagram of the present invention. Detailed Implementation

[0156] 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.

[0157] Example 1

[0158] A dynamic rule-driven multi-domain review method includes:

[0159] S1: A dynamic rule engine that adapts flexibly to different scenarios and policies. Dynamic configuration includes: review process, review stages, scoring methods, scoring dimensions, rule parameters, and configuration confirmation. It allows for the rapid generation of review processes and rules that meet business needs through visual configuration.

[0160] The review process is set up as follows: Enter the backend configuration center and configure the review process according to different specializations, including at least: talent review, job review, professional title review, and project review; select an existing process template according to the review type and then manually optimize and adjust it, or directly customize the configuration; set user roles according to review needs, including: review administrator and conference host.

[0161] The selection of review stages is as follows: select the stages to be enabled at each process node, such as enabling the document review stage and skipping the on-site defense stage; and make differentiated configurations according to the review type.

[0162] The scoring method selection configuration offers ten-point system, percentage system, grade system, pass or fail; different scoring modes can be set for different dimensions of the same project, such as using a percentage system for technical scores and a grade system for comprehensive scores;

[0163] The scoring dimensions configuration includes: basic conditions, budget rationality, and project feasibility. The weights of each dimension are set as follows: basic conditions 30%, budget rationality 30%, and project feasibility 40%. The system will automatically associate the newly added dimensions and weights with the scoring engine.

[0164] The rule parameter configuration process verifies the rules, and the process is completed according to the configured rules. For example, the process can only proceed to the next stage after all experts have completed their scoring, and the process can only proceed to the next stage after all applicants have completed their review comments. The rule parameters can also be directly modified when policies are adjusted.

[0165] S2: Expert Management. Through expert information entry, tag system construction, dynamic profile updating, and expert status management, accurate profiles are constructed and dynamically updated to achieve centralized management and accurate profiles of expert resources. Based on accurate expert profiles and resource management, a multi-dimensional expert capability profile model is constructed to ensure the compliance of expert qualifications.

[0166] The expert information entered includes: basic information such as name, organization, and professional title, as well as qualification certificates such as academic certificates and professional qualification certificates; the system will identify key information such as the date and validity period of the imported qualification certificates by manually importing the above expert information or by automatically importing the above expert information through a form template.

[0167] Before building the tagging system, expert tags need to be added, categorized by expert field such as: Artificial Intelligence, Financial Auditing, Cutting-Edge Technology, etc. The system will then automatically extract keywords from experts' historical review records to supplement the tags, constructing the tagging system as follows:

[0168] A tag building model employing multi-source fusion combined with human-machine collaboration is adopted, comprising a three-layer tag structure:

[0169] The first layer consists of basic attribute tags, including research field, professional title, academic title, institution, and region.

[0170] The second layer consists of dynamic behavioral tags, including the number of reviews, review status, average scoring tendency, response speed tag, and return rate tag.

[0171] The third layer consists of quality assessment labels, including effective scoring rate, score adjustment rate, and rejection rate.

[0172] The method for automatically extracting historical review tags is as follows:

[0173] a. Historical review record mining

[0174] Extract all applications reviewed by experts from historical review records, use the TF-IDF algorithm to extract research field keywords from the applications, calculate keyword frequencies, and generate a candidate tag set;

[0175] b. Domain similarity calculation

[0176] Similarity (expert, application) = Σ(expert tag i ∩ application field j) / |expert tag set|. When the similarity > threshold θ, in this embodiment the threshold θ is 0.6, the application field is added as a candidate tag.

[0177] c. Label confidence score

[0178] Confidence level = (Number of historical reviews × 0.4) + (Review quality score × 0.3) + (Domain matching degree × 0.3)

[0179] in:

[0180] Review quality score = (1 - Return rate) × (1 - Rejection rate) × Valid score rate

[0181] Domain matching score = Number of reviews in this domain / Total number of reviews

[0182] d. Manual review mechanism

[0183] Labels with a confidence level > 0.8 are automatically added;

[0184] Tags with a confidence level of 0.5 or less and a confidence level of ≤0.8 are pushed to the administrator for review;

[0185] Labels with a confidence level ≤ 0.5 are discarded;

[0186] After extracting expert tags, construct expert profiles.

[0187] A star-shaped model is used to construct a multi-dimensional profile of experts. The basic information of the profile includes: a unique expert identifier, a set of research field tags, professional title, a list of titles, an institution ID, and provincial / international identification.

[0188] The dimensions of the dynamic profile include: cumulative number of reviews, number of reviews this year, average score, score variance, average response time, return rate, rejection rate, effective score rate, activity score, and review quality score.

[0189] When an expert submits a new review record, an administrator adjusts or rejects an expert's score, an expert voluntarily returns a review task, or a scheduled task is updated, the expert profile will be updated accordingly. The update method is as follows:

[0190] New portrait value = α × old portrait value + (1-α) × new data, where α is the smoothing coefficient, default 0.7;

[0191] The updated expert profiles are scored for quality, including an overall quality score and an activity score.

[0192] Overall quality score:

[0193] qualityScore = w1 × Response Speed ​​Score + w2 × Scoring Stability Score + w3 × Validity Score + w4 × (1 - Return Rate) + w5 × (1 - Rejection Rate)

[0194] The above response speed score = 1 / (1 + log(average response time / h))

[0195] Rating stability score = 1 / (1 + rating variance)

[0196] Weight vector [w1,w2,w3,w4,w5]=[0.2,0.25,0.25,0.15,0.15]

[0197] Activity score algorithm: activityScore = (Number of reviews this year / System average number of reviews) × (1 - Number of idle days / 365)

[0198] Idle days = Current date - Last submission date for review;

[0199] In expert management, the collaborative management framework for participation status is as follows:

[0200] a. State Machine Model: The expert review lifecycle is managed using a Finite State Machine (FSM) model, comprising four state phases: Draft: The application is assigned to experts; Evaluated: Experts complete their review after receiving the application; Submitted: The reviewed application is submitted; Returned: The administrator returns the submitted application to the expert. The constraints between these states are as follows:

[0201] Drafts must be scored and commented on during the evaluation stage; drafts must meet the scoring rules before being submitted; submitted drafts can only be returned by the administrator.

[0202] b. Concurrent state synchronization algorithm, including update review state consistency and state consistency guarantee, wherein Java distributed lock is used to ensure update review state consistency; the state consistency guarantee is implemented as follows:

[0203] AtomicInteger is used to ensure the atomicity of the counter; optimistic locking is implemented at the database level using a version number field; and expert score update events are triggered by state changes.

[0204] c. Multi-project conflict detection algorithm, through time window overlap detection, is used to ensure that the same application can only be reviewed by one expert at a time, to prevent data conflicts.

[0205] d. Status linkage triggering mechanism, the process stages are as follows:

[0206] After an expert submits a review, an expert score update event is triggered. The system checks whether the application score has been updated and whether all experts have submitted their applications. Based on the check results, the administrator is notified.

[0207] When an expert returns a match for review, an expert review return event is triggered. The system deletes the matching result and notifies the administrator to reassign the match.

[0208] The administrator rejected the rating, the rating status was updated to unrated, the average review score was recalculated, and the expert profile was updated.

[0209] e. Intelligent schedule management algorithm, load balancing scheduling: To effectively balance the workload of review experts, the algorithm does not use a simple round-robin or random method when assigning review tasks. Instead, it introduces an allocatable metric to dynamically evaluate and prioritize the most suitable experts, as detailed below:

[0210] Allocability = (1 - Number of currently assigned tasks / Maximum number of tasks an expert can handle) × (1 - Time window overlap) × Expert qualification score

[0211] In the above formula, (1 - number of currently assigned tasks / maximum number of tasks an expert can handle): this part represents the load factor; the fewer tasks an expert currently handles, the closer its value is to 1, indicating that it has sufficient workload and higher priority, thus avoiding excessive concentration of tasks.

[0212] (1-Time Window Overlap): This part represents the time availability factor. The time window overlap is calculated by the proportion of the number of days that overlap between the review time windows of the expert's existing tasks and the review time windows of the tasks to be assigned to the total number of review days. The lower the overlap, the more idle the expert is during that time period. The higher the value of this factor, the higher the allocation priority.

[0213] Expert Qualification Rating: This section represents the quality factor; it is calculated based on factors such as the expert's professional field of expertise, past review quality, and reputation level to ensure that tasks are assigned to the most competent experts.

[0214] By integrating three core elements—load balancing, time conflict avoidance, and quality assurance—the algorithm can intelligently and dynamically select the "most available, most suitable, and most available" experts for each task to be reviewed, thereby achieving efficient, fair, and high-quality allocation of the overall review tasks.

[0215] The expert management module centralizes resource management and precise profiling, constructing a multi-dimensional capability model to verify expert qualifications, screen experts to be avoided, and generate an expert candidate pool. The multi-dimensional expert capability profile model includes the following core dimensions:

[0216] a. Research Field Codes: Adopting the Science and Technology Department's detailed classification coding system, supporting multi-field tags, and forming an expert professional capability map.

[0217] b. Unit Identifier (UnitId / UnitName): Used to implement unit mutual exclusion constraints in the double-blind mechanism.

[0218] c. Geographical Distribution Characteristics (in Yunnan): Differentiating between experts from within and outside the province to support a regional allocation strategy.

[0219] d. Job Title / Title: Constructing a dimension for evaluating expert authority

[0220] e. Dynamic load counter (ExpertCounter): Tracks the allocation of expert review tasks in real time to achieve load balancing;

[0221] Model training methods

[0222] A hybrid training model combining rule-driven and heuristic optimization is adopted:

[0223] a. Core Matching Rules

[0224] Research Field Matching Rule: Achieves precise field alignment by calculating the intersection of the applicant's research field code and the expert's field label set;

[0225] Unit Mutex Matching Rule: In double-blind mode, it forces the filtering of experts from the same institution to ensure the impartiality of the review process;

[0226] Province Ratio: Experts from within and outside the province are allocated according to a preset ratio to meet policy requirements;

[0227] The application process of the multi-dimensional expert capability profile model is as follows:

[0228] a. Matching mechanism selection

[0229] The system supports three matching modes:

[0230] Point-to-point fully random matching: randomly selects qualified experts from the global expert pool for each applicant;

[0231] Group-to-group matching: Applicants and experts are pre-grouped and matched within the group, which is suitable for field-specific review scenarios;

[0232] Manual import: Supports manual intervention and handling of special cases;

[0233] b. Execution process

[0234] Expert pool initialization: Load active experts for the current year from the global expert database and build a concurrent and secure counter mapping table;

[0235] Parallel matching execution: Java parallel streams are used to process the list of applicants in a multi-threaded manner;

[0236] Multi-stage screening:

[0237] Phase 1: Domain Matching Filtering

[0238] Phase Two: Unit Exclusivity Check

[0239] Phase 3: Stratified sampling within and outside the province

[0240] Phase 4: Load Balancing Selection

[0241] Anomaly Handling and Compensation: When there is a shortage of experts, the matching error is recorded and manual supplementation is supported later;

[0242] c. Double-blind mechanism guarantee

[0243] Innovation: Achieving three layers of double-blind protection:

[0244] Pre-matching phase: Stakeholders are automatically excluded using unit mutual exclusion rules;

[0245] Matching results are isolated: information is visible only to the applicant and the expert; experts can only view anonymized application materials.

[0246] Dynamic avoidance mechanism: Supports experts to voluntarily withdraw, and the system automatically triggers re-matching;

[0247] To complete the above-described profiles and qualification verification of experts, automatic grouping and matching of experts is required, including:

[0248] S1: Conflict rule settings, configure avoidance rules, and use a multi-dimensional expert capability profile model to automatically verify experts and remove conflicting personnel in this review;

[0249] S2: Automatic matching triggering. Based on the review requirements, the multi-dimensional expert capability profile model selects experts with a matching degree of ≥80% from the expert database that has completed the avoidance rule matching to generate a candidate pool.

[0250] S3: Grouping parameter configuration. Set the system to conduct parallel reviews in 3 groups, with 5 experts in each group. The administrator selects the grouping conditions, and the ratio of technical experts to financial experts in each group is 4:1. The system automatically distributes experts evenly. When there are enough experts in the candidate pool, the system selects experts from high to low matching degree to accommodate secondary matching situations. If there are not enough experts in the candidate pool, the system supplements them according to matching degree, and experts with a matching degree lower than the set matching degree are highlighted after grouping.

[0251] S4: Manual adjustment and optimization. Administrators can manually add or remove experts. After adjustment, the system will automatically re-verify the grouping conditions.

[0252] S5: Invitation Information Sending. After a preliminary expert list is generated, the administrator automatically generates a standardized notification template based on the system. The information includes the project name, review time, and feedback link. The administrator can edit and supplement the content with remarks. The notification method adopts a dual mode of SMS sending and in-system message push. The system connects to the operator's SMS gateway to send SMS to the expert's mobile phone number, and at the same time pushes a notification in the system's "Message Center" including a link to jump to the review feedback with one click. After sending, the sending status is displayed in real time, including two statuses: sent or failed to send. Failed numbers can be manually retried or the contact information can be replaced.

[0253] S6: Expert Participation Feedback. After receiving the notification, experts can access the "Participation Confirmation" page via SMS link or system message. The page displays anonymized project information, including basic information, review time, and required time, filtered according to conflict avoidance rules. If an expert selects "Confirm Participation," the system automatically records the confirmation status and updates the expert's schedule and status. If an expert confirms "Unable to Participate," they must select and provide a reason, such as time conflict or field mismatch. The system automatically sends the reason to the administrator and marks the expert as "Unable to Participate." Administrators can view the expert response progress in real time through the "Feedback Monitoring Panel." If 5 out of 10 experts have confirmed and 2 are unable to participate, the system automatically triggers a second reminder for experts who do not provide feedback within a set time. Administrators can customize the feedback time limit. Experts who do not provide feedback within a set number of notifications are considered unable to participate by default. Administrators can customize the number of notifications.

[0254] S7: Rematching and matching: After the invitation and feedback process for the participants ends, the system automatically locks the experts who have been matched and the experts who cannot participate. For the vacant number of participants, repeat the above steps S2~S6 to match again until the matching is completed and the rules are met.

[0255] S8: Matching Result Confirmation. View the final confirmation status of the grouping, which includes the list of confirmed experts and the reasons for non-participation. Exporting the expert grouping table is supported. After confirming that the grouping meets the review requirements, "Lock" the system. The system will then bind the grouping results to the project and officially enter the review phase.

[0256] After group matching, the process proceeds to the online expert review stage, including:

[0257] S1: Review materials viewing. After logging into the system, experts can go to "My Reviews" to view project materials, which support online annotation and note-taking.

[0258] S2: Submit multiple scores. On the scoring interface, score according to the technical feasibility and budget rationality of the configuration. Select the corresponding scoring mode, such as 85 points on a percentage scale and "good" on a grade scale. The system displays the scoring progress in real time.

[0259] S3: Automatic score calculation. After scoring is completed, the system automatically calculates the total score according to the preset weights, such as 85×50% for technical score + 90×50% for budget score = 87.5. Real-time viewing of score details is supported.

[0260] S4: Multi-condition ranking recommendation. After the scoring is completed and submitted, experts can rank the scores based on the total score, the highest score in a single item, and the items that meet the criteria. The final ranking table can be exported, and the ranking table includes basic information, scores, rankings, etc.

[0261] S5: Fill in the review comments. Fill in the written comments for each dimension. The system provides comment templates for reference. After completing the comments, you can export the review comment document.

[0262] S6: Result confirmation. Based on the scores, rankings, and expert review opinions, select and confirm the "Proposed Recommendation" list and submit it to proceed to the conference voting stage.

[0263] The general meeting vote included:

[0264] S1: Voting item preparation. In the system backend, the administrator selects whether to vote in groups, whether to vote by topic, and whether all members should attend the meeting. The administrator confirms the voting items, such as selecting 10 people in the artificial intelligence group and 15 people in the medical technology group, and sets voting rules such as single selection, multiple selection, equal number of votes, differential number of votes, and voting deadline.

[0265] S2: Voting permission allocation, specifying the scope of participants in the vote can be set to members of the review committee, and voting notifications can be sent through the system;

[0266] S3: Anonymous voting setting, hiding voter information and only recording voting results and time; S4: Real-time visualization display, dynamically displaying the vote share of each option during the voting process.

[0267] The data can be displayed as a statistical bar chart or pie chart; it supports filtering and viewing detailed data by group.

[0268] S5: Voting results confirmed. After the voting deadline, the system automatically summarizes the results, generates a voting report, and exports it.

[0269] Example 2

[0270] A dynamic rule-driven multi-domain review system includes:

[0271] The dynamic rules engine module includes a review process configuration unit, which is used to configure the review process module according to different projects;

[0272] The review process involves selecting different stages based on the type of project being reviewed.

[0273] The scoring unit allows for the setting of corresponding scoring methods for different projects and dimensions;

[0274] The scoring dimension configuration unit sets the scoring dimensions and assigns a weight to each scoring dimension, which is then used to associate the scoring units.

[0275] The rule parameter setting unit allows you to set process rules at process nodes and corresponding steps, and complete the review process transfer according to the rules.

[0276] The confirmation unit confirms the configuration of each of the above units;

[0277] The expert management module includes an expert information entry unit, which allows users to enter basic information and qualification certificates of experts; and sets up multiple entry methods.

[0278] The tag system construction unit establishes a tag system from basic attribute tags, dynamic behavior tags, and quality assessment tags respectively;

[0279] The profile dynamic update unit builds expert profiles based on the tag system and dynamically updates the expert profiles according to the actual review status results.

[0280] The expert status management unit allocates experts based on factors such as multiple project conflicts, review status matching, and review schedules.

[0281] The grouping and matching module includes a conflict rule setting unit, which identifies whether there is a conflict of interest between experts and the project, and requests expert recusal.

[0282] The automatic matching trigger unit selects and matches suitable experts as candidates from among those who have been recused, based on the review requirements;

[0283] The grouping parameter configuration unit allocates the appropriate number of people from the candidate experts in a reasonable and balanced manner according to the appropriate number of groups and the appropriate number of people in each group.

[0284] In addition to the automatic grouping completed by the automatic matching trigger unit and the grouping parameter configuration unit, the administrator can also manually group the units. After manual grouping, the system will then perform grouping condition verification.

[0285] The invitation information sending unit sends invitation information to the preliminarily selected review experts and displays the information sending status; after receiving the invitation information, the unit receives feedback information from the experts.

[0286] After matching and confirming the expert's response, the participating experts are finally confirmed.

[0287] The online expert review module includes a review materials viewing unit, where experts can log in to the system and view the materials for the review projects they are responsible for.

[0288] The system features multiple scoring units that score project materials according to scoring configurations, and the scoring progress is displayed in real time.

[0289] The automatic score calculation unit completes the scoring, and the system automatically calculates the total score according to the preset weights.

[0290] The multi-dimensional ranking recommendation unit allows experts to rank items based on total score, highest score for a single item, and matching items after the scoring is completed and submitted. It also supports exporting the final ranking table.

[0291] The section for filling in review comments allows experts to enter their comments.

[0292] The conference voting module includes a voting item preparation unit, an administrator operation setting for the voting mode, and a setting for the voting rules;

[0293] The voting rights allocation unit sets the scope of personnel who can participate in voting;

[0294] An anonymous voting feature hides the voter's explicit information.

[0295] The display unit shows the actual voting results in real time.

Claims

1. A dynamic rule-driven multi-domain review method, characterized in that, include: S1: A dynamic rule engine that adapts flexibly to different scenarios and policies. Dynamic configuration includes: review process, review stages, scoring methods, scoring dimensions, rule parameters, and configuration confirmation. It allows for the rapid generation of review processes and rules that meet business needs through visual configuration. S2: Expert Management. Through expert information entry, tag system construction, dynamic profile updating, and expert status management, accurate profiles are constructed and dynamically updated to achieve centralized management and accurate profiles of expert resources. Based on accurate expert profiles and resource management, a multi-dimensional expert capability profile model is constructed to ensure the compliance of expert qualifications. The expert management system also includes a participation status linkage management architecture unit, which is as follows: a. State Machine Model: A finite state machine is used to manage the expert review lifecycle, comprising four state phases: Draft: The application is assigned to experts; Evaluated: Experts receive the application and complete the review; Submitted: The reviewed application is submitted; Returned: The administrator returns the submitted application to the expert. The constraints between these states are as follows: Drafts must be scored and commented on during the evaluation stage; drafts must meet the scoring rules before being submitted; submitted drafts can only be returned by the administrator. b. Concurrent state synchronization algorithm, including update review state consistency and state consistency guarantee, wherein Java distributed lock is used to ensure update review state consistency; the state consistency guarantee is implemented as follows: AtomicInteger is used to ensure the atomicity of the counter; optimistic locking is implemented at the database level using a version number field; and expert score update events are triggered by state changes. c. Multi-project conflict detection algorithm, through time window overlap detection, is used to ensure that the same application can only be reviewed by one expert at a time, to prevent data conflicts. d. Status linkage triggering mechanism, the process stages are as follows: After an expert submits a review, an expert score update event is triggered. The system checks whether the application score has been updated and whether all experts have submitted their applications. Based on the check results, the administrator is notified. When an expert returns a match for review, an expert review return event is triggered. The system deletes the matching result and notifies the administrator to reassign the match. The administrator rejected the rating, the rating status was updated to unrated, the average review score was recalculated, and the expert profile was updated. e. Intelligent schedule management algorithm, load balancing scheduling: To effectively balance the workload of review experts, the algorithm does not use a simple round-robin or random method when assigning review tasks. Instead, it introduces an allocatable metric to dynamically evaluate and prioritize the most suitable experts, as detailed below: Allocability = (1 - Number of currently assigned tasks / Maximum number of tasks an expert can handle) × (1 - Time window overlap) × Expert qualification score In the above formula, (1 - number of currently assigned tasks / maximum number of tasks that an expert can handle): this part represents the load factor; (1 - Time window overlap): This part represents the time availability factor; Expert Qualification Rating: This section represents the quality factor; S3: Automatic grouping and matching, through the following: conflict rule setting, automatic matching triggering, grouping parameter configuration, manual adjustment and optimization, invitation information sending, expert participation feedback, judgment and rematching, and matching result confirmation, achieves intelligent recommendation and flexible adjustment, which is used for accurate expert grouping and avoids conflicts of interest; S4: Online expert review enables efficient online expert review through viewing review materials, submitting multi-dimensional scores, automatic score calculation, multi-condition sorting and recommendation, filling in review comments, and confirming results. It supports multi-dimensional scoring and automatic calculation to ensure the standardization of the review process. S5: Conference voting, through the preparation of voting items, allocation of voting rights, setting of anonymous voting, real-time visualization and confirmation of voting results, to achieve transparency in the voting process and real-time presentation of results.

2. The dynamic rule-driven multi-domain review method according to claim 1, characterized in that, The dynamic rule engine settings for the review process are as follows: Enter the backend configuration center and configure the review process according to different specializations, including at least: talent review, job review, professional title review, and project review; select an existing process template according to the review type and then manually optimize and adjust it, or directly customize the configuration; set user roles according to review requirements, including: review administrator and conference host. The selection of the review stage is specifically as follows: select the stage to be enabled at the process node and configure it differently according to the review type; The scoring method selects and configures different scoring modes; then, different scoring modes are set for different dimensions of the same project. The scoring dimension configuration settings include scoring dimensions such as basic conditions, budget rationality, and project feasibility, setting the weight of each dimension and linking it to the scoring engine; The rule parameter configuration process verifies the rules and completes the process transmission according to the configured rules; and directly modifies the rule parameters when the policy is adjusted.

3. The dynamic rule-driven multi-domain review method according to claim 1, characterized in that, The expert information entered in the expert management system shall include at least: basic information and qualification certificates; After adding expert domain tags, the system automatically extracts keywords from the expert's historical review records to supplement the tags, constructing the tag system according to the following method: A tag building model employing multi-source fusion combined with human-machine collaboration is adopted, comprising a three-layer tag structure: The first layer consists of basic attribute tags, including research field, professional title, academic title, institution, and region. The second layer consists of dynamic behavioral tags, including the number of reviews, review status, average scoring tendency, response speed tag, and return rate tag. The third layer consists of quality assessment labels, including effective scoring rate, score adjustment rate, and rejection rate. The automatic tag extraction method is as follows: a. Historical review record mining Extract all applications reviewed by experts from historical review records, use the TF-IDF algorithm to extract research field keywords from the applications, calculate keyword frequencies, and generate a candidate tag set; b. Domain similarity calculation Similarity (expert, application) = Σ(expert label i ∩ application field j) / | set of expert labels|. When the similarity > threshold θ, the application field is added as a candidate label. c. Label confidence score Confidence level = (Number of historical reviews × 0.4) + (Review quality score × 0.3) + (Domain matching degree × 0.3) in: Review quality score = (1 - Return rate) × (1 - Rejection rate) × Valid score rate Domain matching score = Number of reviews in this domain / Total number of reviews d. Manual review mechanism Labels with a confidence level > 0.8 are automatically added; Tags with a confidence level of 0.5 or less and a confidence level of ≤0.8 are pushed to the administrator for review; Labels with a confidence level ≤ 0.5 are discarded; Expert Profile Building Model A star-shaped model is used to construct a multi-dimensional profile of experts. The basic information of the profile includes: a unique expert identifier, a set of research field tags, professional title, a list of titles, an institution ID, and provincial / international identification. The dimensions of the dynamic profile include: cumulative number of reviews, number of reviews this year, average score, score variance, average response time, return rate, rejection rate, effective score rate, activity score, and review quality score. The dynamic profile update trigger conditions include at least: experts submitting new review records, administrators adjusting or rejecting expert scores, experts voluntarily returning review tasks, and scheduled update tasks; The update method is as follows: New portrait value = α × old portrait value + (1-α) × new data, where α is the smoothing coefficient, which is 0.7 by default.

4. The dynamic rule-driven multi-domain review method according to claim 3, characterized in that, The newly generated expert profiles are scored for quality, including: overall quality score and activity score; The overall quality score: qualityScore = w1 × Response Speed ​​Score + w2 × Scoring Stability Score + w3 × Validity Score + w4 × (1 - Return Rate) + w5 × (1 - Rejection Rate) The above response speed score = 1 / (1 + log(average response time / h)) Rating stability score = 1 / (1 + rating variance) Weight vector [w1,w2,w3,w4,w5]=[0.2,0.25,0.25,0.15,0.15] The activity score algorithm is as follows: activityScore = (Number of reviews this year / Average number of reviews in the system) × (1 - Number of idle days / 365) Idle days = Current date - Last submission date for review.

5. The dynamic rule-driven multi-domain review method according to claim 1, characterized in that, In the aforementioned expert management, resources are centrally managed and precise profiles are created to construct a multi-dimensional capability model for verifying expert qualifications. The specific construction process is as follows: Construct a multi-dimensional expert capability profile model, including the following core dimensions: a. Research Field Vector: Adopting the Science and Technology Department's detailed category coding system, supporting multi-field tags, and forming an expert professional capability map. b. Unit Attribution Identifier: Used to implement unit mutual exclusion constraints in the double-blind mechanism. c. Geographical Distribution Characteristics: Differentiate between experts from within and outside the province to support a regional allocation strategy. d. Professional Titles and Honors: Constructing Dimensions for Evaluating Expert Authority e. Dynamic load counter: Tracks the allocation of expert review tasks in real time to achieve load balancing. Training methods A hybrid training model combining rule-driven and heuristic optimization is adopted: a. Core Matching Rules Research field matching rules: Accurate field alignment is achieved by calculating the intersection of the applicant's research field code and the expert field label set; Institutional mutual exclusion rule: In double-blind mode, experts from the same institution are forcibly filtered out to ensure the impartiality of the review; The allocation rule between experts from within and outside the province is to allocate experts from within and outside the province according to a predetermined ratio to meet policy requirements.

6. The dynamic rule-driven multi-domain review method according to claim 1, characterized in that, In the automatic grouping and matching: S1: Conflict rule settings, configure avoidance rules, and use a multi-dimensional expert capability profile model to automatically verify experts and remove conflicting personnel in this review; S2: Automatic matching triggering. Based on the review requirements, the multi-dimensional expert capability profile model selects experts with a matching degree of ≥80% from the expert database that has completed the avoidance rule matching to generate a candidate pool. S3: Grouping parameter configuration, setting the number of groups and the number of experts in each group. The administrator selects the grouping conditions, and the system automatically and evenly distributes experts. When there are enough experts in the candidate pool, they are selected from high to low matching degree to accommodate secondary matching. If there are not enough experts in the candidate pool, they are supplemented according to matching degree. Experts with a matching degree lower than the set matching degree are highlighted after grouping. S4: Manual adjustment and optimization. Administrators can manually add or remove experts. After adjustment, the system will automatically re-verify the grouping conditions. S5: Invitation information is sent. After the initial list of experts is generated, the administrator automatically generates a standardized notification template based on the system. The administrator can edit the remarks and supplement the content. The notification method adopts a dual mode of SMS sending and in-system message push. The system calls the SMS interface to send SMS to the expert's mobile phone number, and at the same time pushes the notification in the system, including a link to jump to the evaluation feedback with one click. After sending, the sending status is displayed in real time. Failed numbers can be manually retried or the contact information can be replaced. S6: Expert feedback: After receiving the notification, if an expert selects "Confirm Participation," the system will automatically record the confirmation status and update the expert's schedule and status synchronously. If the expert selects "Unable to Participate," a reason must be selected, and the system will automatically send the reason to the administrator's backend and mark the expert as "Unable to Participate." The administrator can view the expert's response progress in real time. For experts who do not respond within a set time, the system will automatically trigger a second reminder. Experts who do not respond to notifications within a set number of times are considered unable to participate by default. S7: Rematching and matching: After the invitation and feedback process for the participants ends, the system automatically locks the experts who have been matched and the experts who cannot participate. For the vacant number of participants, repeat the above steps S2~S6 to match again until the matching is completed and the rules are met. S8: Matching results confirmed. Check the final confirmation status of the group and officially enter the review stage.

7. The dynamic rule-driven multi-domain review method according to claim 1, characterized in that, The online expert review process is as follows: S1: Review materials review. Experts can log in to the system to review project materials. S2: Multi-dimensional scoring submission. Scores are submitted according to the configured dimensions on the scoring interface. The corresponding mode is selected, and the system displays the scoring progress in real time. S3: Automatic score calculation. After scoring is completed, the system automatically calculates the total score according to the preset weights and supports real-time viewing of score details. S4: Multi-condition ranking recommendation. After the scoring is completed and submitted, experts can rank the scores based on the total score, the highest score in a single item, and the items that meet the criteria. The final ranking table can be exported. S5: Fill in the review comments. Fill in the written comments for each dimension. The system provides comment templates for reference. After completing the comments, you can export the review comment document. S6: Result confirmation. Based on the scores, rankings, and expert review opinions, select and confirm the "Proposed Recommendation" list and submit it to proceed to the conference voting stage.

8. The dynamic rule-driven multi-domain review method according to claim 1, characterized in that, The voting at the general meeting included: S1: Voting item preparation. In the system backend, the administrator selects whether to vote in groups, whether to vote by topic, and whether all members should participate in the meeting, confirms the voting items, and sets the voting rules. S2: Voting permission allocation, specifying the scope of participants in the vote, and sending voting notifications through the system; S3: Anonymous voting setting, hiding voter information and only recording voting results and time; S4: Real-time visualization display, dynamically showing the vote share of each option during the voting process. Supports filtering and viewing detailed data by group; S5: Voting results confirmed. After the voting deadline, the system automatically summarizes the results, generates a voting report, and exports it.

9. A dynamic rule-driven multi-domain review system, used to implement the dynamic rule-driven multi-domain review method according to any one of claims 1 to 8, characterized in that, include: The dynamic rules engine module includes a review process configuration unit, which is used to configure the review process module according to different projects; The review process involves selecting different stages based on the type of project being reviewed. The scoring unit allows for the setting of corresponding scoring methods for different projects and dimensions; The scoring dimension configuration unit sets the scoring dimensions and assigns a weight to each scoring dimension, which is then used to associate the scoring units. The rule parameter setting unit allows you to set process rules at process nodes and corresponding steps, and complete the review process transfer according to the rules. The confirmation unit confirms the configuration of each of the above units; The expert management module includes an expert information entry unit, where basic information and qualification certificates of experts are entered. It also sets up multiple methods for data entry; The tag system construction unit establishes a tag system from basic attribute tags, dynamic behavior tags, and quality assessment tags respectively; The profile dynamic update unit builds expert profiles based on the tag system and dynamically updates the expert profiles according to the actual review status results. The expert status management unit allocates experts based on factors such as multiple project conflicts, review status matching, and review schedules. The grouping and matching module includes a conflict rule setting unit, which identifies whether there is a conflict of interest between experts and the project, and requests expert recusal. The automatic matching trigger unit selects and matches suitable experts as candidates from among those who have been recused, based on the review requirements; The grouping parameter configuration unit allocates the appropriate number of people from the candidate experts in a reasonable and balanced manner according to the appropriate number of groups and the appropriate number of people in each group. In addition to the automatic grouping completed by the automatic matching trigger unit and the grouping parameter configuration unit, the administrator can also manually group the units. After manual grouping, the system will then perform grouping condition verification. The invitation information sending unit sends invitation information to the preliminarily selected review experts and displays the information sending status; after receiving the invitation information, the unit receives feedback information from the experts. After matching and confirming the expert's response, the participating experts are finally confirmed. The online expert review module includes a review materials viewing unit, where experts can log in to the system and view the materials for the review projects they are responsible for. The system features multiple scoring units that score project materials according to scoring configurations, and the scoring progress is displayed in real time. The automatic score calculation unit completes the scoring, and the system automatically calculates the total score according to the preset weights. The multi-dimensional ranking recommendation unit allows experts to rank items based on total score, highest score for a single item, and matching items after the scoring is completed and submitted. It also supports exporting the final ranking table. The section for filling in review comments allows experts to enter their comments. The conference voting module includes a voting item preparation unit, an administrator operation setting for the voting mode, and a setting for the voting rules; The voting rights allocation unit sets the scope of personnel who can participate in voting; An anonymous voting feature hides the voter's explicit information. The display unit shows the actual voting results in real time.

Images