Search

CN-121833787-B - Dynamic rule driven multi-domain review method and system

CN121833787BCN 121833787 BCN121833787 BCN 121833787BCN-121833787-B

Abstract

The invention discloses a dynamic rule-driven multi-field review method and a dynamic rule-driven multi-field review system, which belong to the technical field of intelligent review. Whether talent selection, technical evaluation of scientific research projects and job title review are carried out, the dynamic construction and flexible adjustment of the differential review flow can be completed based on the same platform, the review efficiency and scientificity are effectively improved, a standardized and intelligent technical solution is provided for multi-subject and multi-scene collaborative review business, and the problems of flow stiffness, weak function, rough management, transparency loss and safety vulnerability of the conventional review system are solved.

Inventors

  • MA YUNQIANG
  • LI YAN
  • YANG YAHAN
  • ZHANG JIQIANG
  • CHEN SHUANGMEI
  • WANG NA
  • ZHOU WEI

Assignees

  • 云南这里信息技术有限公司

Dates

Publication Date
20260508
Application Date
20260313

Claims (9)

  1. 1. A dynamic rule driven multi-domain review method, comprising: The system comprises a dynamic rule engine, a dynamic configuration module, a visual configuration module and a control module, wherein the dynamic rule engine is used for flexibly adjusting an adaptation scene and a policy, and the dynamic configuration comprises a review flow, a review link, a scoring mode, a scoring dimension, rule parameters and configuration validation confirmation; S2, expert management, namely performing accurate portrait construction and dynamic update through expert information input, label system construction, portrait dynamic update and expert state management, realizing expert resource centralized management and accurate portrait, constructing a multi-dimensional expert capacity portrait model based on the expert accurate portrait and resource management, and guaranteeing expert qualification compliance; the expert management also comprises a parameter state linkage management architecture unit, wherein the parameter state linkage management architecture is as follows: a. The state machine model adopts a finite state machine to manage expert review life cycle, comprises four state stages, and is respectively draft, wherein the declaration books are distributed to the expert, the expert is evaluated to finish review after receiving the declaration books, the expert is submitted to submit the reviewed declaration books, the administrator is returned to return the declaration books submitted by the expert, and constraint conditions formed among the states are as follows: The draft to the assessed stage must fill in scores and comments, the assessed to submitted stage must satisfy scoring rules, and the submitted returns can only be operated by an administrator; b. the concurrency state synchronization algorithm comprises updating and reviewing state consistency and state consistency guarantee, wherein the updating and reviewing state consistency is guaranteed by using a java distributed lock, and the state consistency guarantee is realized as follows: The use AtomicInteger of the version number field to realize optimistic lock, the state change triggering expert scoring update event; c. the multi-project conflict detection algorithm is used for ensuring that the same declaration book can only be reviewed by one expert at the same time through time window overlapping detection so as to prevent data conflict; d. The state linkage triggering mechanism comprises the following steps: triggering an expert score updating event after submitting the comments, checking whether the declaration score is updated, checking whether all the experts are submitted, and notifying an administrator based on the checking result; The expert reviews the return event, the system deletes the matching result and notifies the manager of reassignment; The administrator refuses the grading, the grading state is updated to be ungraded, the average grade of the grading is calculated again, and the expert image is updated; e. the intelligent schedule management algorithm, load balancing scheduling, namely for effectively balancing the workload of the review expert, the algorithm does not adopt a simple alternate or random mode when assigning the review task, and introduces an assignable quantitative index to dynamically evaluate and preferentially select the most suitable expert, and the method is as follows: Allocability= (1-number of tasks currently allocated/maximum number of tasks carried by expert) × (1-time window overlap) ×expert qualification score In the above, (1-the number of tasks currently allocated/the maximum number of tasks carried by the expert) this part represents the load factor; (1-time window overlap) this part represents a time availability factor; expert qualification score, this part represents the quality factor; S3, automatic grouping matching is realized by comprising conflict rule setting, automatic matching triggering, grouping parameter configuration, manual adjustment and optimization, invitation information sending, expert evaluation feedback, judgment and re-matching and matching result confirmation, and the automatic grouping matching method is used for expert accurate grouping and avoids benefit conflicts; S4, expert online review, namely realizing the efficient on-line review of the expert, supporting the multiple scoring and the automatic calculation, and guaranteeing the standardization of the review through review material viewing, multiple scoring submission, score automatic calculation, multi-condition ordering recommendation, review opinion filling and result confirmation; And S5, big-party voting, namely realizing transparency of the voting process and real-time presentation of the result through voting item preparation, voting authority allocation, anonymous voting setting, real-time visual display and voting result confirmation.
  2. 2. The method for dynamically driving the multi-domain review according to claim 1, wherein the review flow in the dynamic rule engine setting is specifically set by entering a background configuration center, configuring the review flow at least comprising talent review, post review, job review and project review according to different special types, selecting the existing flow templates according to the review types, then manually optimizing and adjusting or directly customizing the configuration, and setting user roles comprising a review manager and a meeting host according to the review requirements; Selecting links to be started at a flow node, and performing differentiated configuration according to the review type; The scoring mode is selected and configured with different scoring modes, and then different scoring modes are set for different dimensions of the same item; The scoring dimension configuration setting comprises scoring dimensions including basic conditions, budget rationality and project feasibility, setting each dimension weight, and associating to a scoring engine; The rule parameter configures the flow verification rule, completes flow transmission according to the configured rule, and directly modifies the rule parameter when the policy is adjusted.
  3. 3. The dynamic rule-driven multi-domain review method of claim 1 wherein expert information entered in the expert management comprises at least basic information, qualification proof; after the expert domain label is added, the system automatically extracts keywords of the expert history review record to carry out label supplementation, and a label system is constructed according to the following method: The model is constructed by adopting multi-source fusion and combining with a man-machine cooperative tag, and comprises a three-layer tag structure: The first layer is a basic attribute label comprising a research field, a title, a unit and a region; The second layer is a dynamic behavior label comprising a review number, a review state, an average scoring tendency, a response speed label and a return rate label; The third layer is a quality evaluation label comprising an effective scoring rate, a score adjustment rate and a reject rate; the automatic label extraction method comprises the following steps: a. History review record mining Extracting all declarations of expert history review from the history review record, extracting research field keywords in the declarations by using a TF-IDF algorithm, calculating keyword frequency, and generating a candidate tag set; b. domain similarity calculation Similarity (expert, declaration) =Σ (expert label i n declaration book field j)/|expert label set|, and when similarity > threshold θ, adding declaration book field as candidate label; c. Tag confidence score Confidence = (history number of reviews×0.4) + (quality score of review×0.3) + (field matching degree×0.3) Wherein: Review quality score= (1-reject rate) × (1-reject rate) ×effective score rate Domain matching degree = number of domain reviews/total number of reviews D. manual auditing mechanism Automatically adding a label with a confidence level of > 0.8; the label with the confidence coefficient of 0.5< less than or equal to 0.8 is pushed to an administrator for auditing; discarding labels with confidence less than or equal to 0.5; Expert portrait construction model Constructing an expert multidimensional portrait by adopting a star model, wherein basic information of the portrait comprises an expert unique identifier, a research field tag set, a title list, a unit ID and an inside-outside identifier; The dimension information of the dynamic image comprises accumulated review times, annual review times, average scoring, scoring variance, average response time, return rate, rejected rate, effective scoring rate, liveness score and review quality score; The dynamic portrait update triggering conditions at least comprise that an expert submits a new review record, an administrator adjusts or rejects the expert score, the expert actively returns to the review task and the timed update task; The updating method is based on new image value = alpha x old image value + (1-alpha) x newly added data, wherein alpha is a smoothing coefficient, and defaults to 0.7.
  4. 4. The dynamic rule-driven multi-domain review method of claim 3 wherein the newly generated expert representation is quality scored comprising a composite quality score and an liveness score; The comprehensive quality score: qualityScore = w1 x response speed score + w2 x score stability score + w3 x effective rate score + w4 x (1-reject rate) +w5 x (1-reject rate) The above-mentioned response speed is divided into 1/(1+log (average response time length/h)) Score stability score = 1/(1 + score variance) Weight vector [ w1, w2, w3, w4, w5] = [0.2,0.25,0.25,0.15,0.15] The liveness scoring algorithm is activityScore = (number of annual reviews/average number of system reviews) × (1-days of inactivity/365) Idle day = current date-last submitted review date.
  5. 5. The dynamic rule-driven multi-domain review method according to claim 1, wherein in the expert management, resources and accurate portraits are centrally managed, a multi-dimensional capability model is built for verifying expert qualification, and the building process comprises the following steps: constructing a multidimensional expert ability image model, which comprises the following core dimensions: a. Research field vector, adopting science and technology hall subclass coding system to support multi-field labels and form expert professional ability spectrum B. Unit attribution identification for realizing unit mutual exclusion constraint in double-blind mechanism C. regional distribution characteristics, namely distinguishing between provincial and extravehicular experts and supporting regional proportioning strategy D. title and title construction expert authority assessment dimension E. Dynamic load counter for real-time tracking expert review task allocation amount and realizing load balancing Training method A rule driven and heuristic optimization combined hybrid training mode is adopted: a. Core matching rules The research domain matching rule is that accurate domain alignment is realized through intersection operation of declaration research domain codes and expert domain label sets; the unit mutual exclusion rule is that the same unit expert is forcedly filtered under a double-blind mode, so that the fairness of review is ensured; the ratio rule inside and outside the province: the experts in and out of the province are distributed according to the preset proportion, meeting the policy requirements.
  6. 6. The dynamic rule-driven multi-domain review method of claim 1 wherein, in the automatic packet matching: s1, setting conflict rules, configuring avoidance rules, automatically checking experts by adopting a multidimensional expert capacity portrait model, and eliminating conflict personnel of the current review; S2, automatic matching triggering, namely screening the experts with the matching degree of more than or equal to 80% from an expert library which is matched with the avoidance rule by adopting a multidimensional expert ability portrait model according to the evaluation requirement to generate a candidate pool; S3, grouping parameter configuration, namely setting grouping quantity and the number of expert persons in each group, selecting grouping conditions by an administrator, automatically balancing and distributing the expert by a system, and selecting from high to low according to the matching degree when the number of expert persons in a candidate pool is sufficient so as to adapt to the secondary matching condition; S4, manually adjusting and optimizing, wherein an administrator can manually add or remove experts, and the system automatically rechecks grouping conditions after adjustment; S5, after inviting information to be sent and generating a preliminary expert list, an administrator automatically generates a standardized notification template according to the system, the administrator can edit remark supplementary content, the notification mode adopts a double mode of combining short message sending and in-system message pushing, and the system calls a short message interface to send short messages to expert mobile phone numbers, and simultaneously pushes notification in the system, wherein the notification comprises links for one-key jump parameter and comment feedback; s6, expert evaluation feedback is carried out, after the expert receives the notification, the expert selects ' confirm evaluation ', the system automatically records the confirm state and synchronously updates the expert schedule and the expert state, if the ' can't evaluate ' needs to be selected, the system automatically feeds back the reason to the background of the administrator, marks the expert as ' can't evaluate ', the administrator checks the response progress of the expert in real time, the system automatically triggers secondary reminding for the expert which is not fed back in the set time, and the expert which is not fed back in the set times is notified and defaults as can't evaluate; s7, judging re-matching, automatically locking the matched expert and the expert incapable of being evaluated by the system after the invitation and feedback of the evaluation are finished, and repeating the steps S2-S6 for the vacant number to re-match until the matching is finished and the rule is met; and S8, confirming the matching result, checking the final confirmation state of the group, and formally entering a review link.
  7. 7. The dynamic rule-driven multi-domain review method of claim 1 wherein the expert on-line review process is as follows: S1, checking the materials, and checking the project materials after an expert logs in a system; s2, multi-element scoring submission, scoring respectively according to configured dimensions at a scoring interface, selecting a corresponding mode, and displaying scoring progress in real time by a system; s3, automatically calculating the score, and automatically calculating total score according to preset weight by the system after scoring is completed, so as to support real-time checking of score details; S4, multi-condition sorting recommendation, wherein after scoring is finished and submitted, an expert can sort based on total score, single highest score and coincidence item and support to derive a final sorting table; s5, filling in the review comments, filling in text comments aiming at each dimension, providing a comment template for reference by the system, and supporting to export the review comment document after finishing filling in the comments; and S6, confirming the result, checking and confirming a 'quasi-recommended' list by expert review opinion according to the score and the sequence, submitting the list, and entering a conference voting link.
  8. 8. The dynamic rule-driven multi-domain review method of claim 1 wherein the big vote comprises: S1, preparing voting items, wherein an administrator selects whether to group voting, whether to vote according to special terms, whether to meet on a whole member in a system background, confirms the voting items and sets voting rules; S2, voting authority allocation, designating a range of participation voters, and sending a voting notice through a system; s3, setting anonymous voting, hiding voter information, only recording voting result and time, S4, dynamically displaying the vote rate of each option in the voting process in real time, Support screening by group to view the subdivision data; and S5, confirming the voting result, automatically gathering the result by the system after the voting is ended, generating a voting report and exporting the voting report.
  9. 9. A dynamic rule driven multi-domain review system for implementing the dynamic rule driven multi-domain review method of any one of claims 1-8, comprising: the dynamic rule engine module comprises a review flow configuration unit and a dynamic rule module, wherein the review flow configuration unit is used for configuring the review flow module according to different projects; The review link selection unit is used for selecting different links according to the types of the review items at the review flow node; The scoring unit is used for setting corresponding scoring modes for different projects and different dimensions; The grading dimension configuration unit is used for setting grading dimensions, setting dimension weights corresponding to the grading dimensions and associating the grading dimensions with the grading units by the dimension weights; the rule parameter setting unit sets a process rule at the process node and the corresponding link, and completes the review process transmission according to the rule; a confirmation unit for confirming the configuration of each unit; the expert management module comprises an expert information input unit, an expert information input unit and an expert information input unit, wherein the expert information input unit is used for inputting basic information and qualification evidence of an expert; The label system construction unit is used for respectively constructing a label system from the basic attribute label, the dynamic behavior label and the quality evaluation label; a portrait dynamic updating unit for constructing an expert portrait based on the label system and dynamically updating the expert portrait according to the expert actual evaluation status result; The expert state management unit is used for allocating the expert from the aspects of multi-project conflict, review state matching and review schedule; A packet matching module including a conflict rule setting unit, identifying whether the expert collides with the project or not, and applying for the expert to avoid; the automatic matching triggering unit is used for screening and matching proper experts from the avoided experts to be used as candidates according to the review requirement; the grouping parameter configuration unit reasonably and evenly distributes proper numbers of people in the group numbers from candidate owners according to proper group numbers; the manual adjustment optimizing unit not only automatically matches the automatic grouping completed by the triggering unit and the grouping parameter configuration unit, but also can manually group the grouping parameters, and the grouping condition is checked by the system after the manual grouping; An invitation information transmitting unit that transmits invitation information to the preliminary determination review expert and displays an information transmission state; after receiving the invitation information, the expert receives feedback information of the expert; the matching confirmation unit is used for finally confirming the evaluation specialist after obtaining the response feedback of the specialist; The expert online review module comprises a review material viewing unit, an expert login system and a review project material responsible for the expert online review module; the multi-element scoring unit scores the project materials according to scoring configuration, and the system displays scoring progress in real time; the score automatic calculation unit finishes scoring, and the system automatically calculates total score according to preset weight; The multidimensional sorting recommendation unit is used for sorting the expert based on the total score, the single highest score and the coincidence item after the scoring is finished and submitted, and supporting to derive a final sorting table; the expert inputs the review comments; the big meeting voting module comprises a voting item preparation unit, wherein an administrator operates to set a voting mode and set a voting rule; a voting authority allocation unit for setting a range of persons who can participate in voting; An anonymous voting setting unit for hiding the voting humanized information; And the display unit displays the actual voting result in real time.

Description

Dynamic rule driven multi-domain review method and system Technical Field The invention belongs to the technical field of intelligent evaluation, and particularly relates to a dynamic rule-driven multi-domain evaluation method and system. Background The review activities are increasingly complicated, innovation results of the interdisciplinary and interdisciplinary fields are continuously emerging, higher requirements are put on the adaptability and the professional performance of the review system, and the requirements on fairness, efficiency and accuracy of the review process are urgent. Conventional, relatively cured review systems have difficulty effectively addressing this complexity. The current evaluation situation presents three core characteristics, namely obvious field diversity, essential difference between evaluation flows, index systems and rule logics in different fields, frequent requirement of rule dynamic enhancement, policy adjustment, evaluation standard optimization and the like, continuous shortening of rule iteration period, data polymorphism development, and cross-system collaborative verification of multi-mode data such as images, audios and structured reports of evaluation materials which are expanded from single texts. Expert review in complex application scenes faces a plurality of troublesome problems, and the traditional review system architecture is old and is difficult to adapt to the dynamically changeable business requirements. From the macroscopic view, the digital control transformation hysteresis of the existing system is based on the development of a single-body architecture, the system architecture is old, the coupling degree of the modules is high, the function expansion needs to reconstruct core codes, and meanwhile, the technical standard is lost, so that the cross-department system docking needs to customize development interfaces. From a microscopic point of view, the prior art has the following technical bottlenecks: 1) The rule is stiff, the adaptability is low, a, most of the existing systems adopt preset and fixed review processes, and when facing different types of projects, the review steps, link sequences and rules are difficult to flexibly adjust according to the project characteristics, the review conditions and the difference of participating departments; 2) The cross-domain review capability is weak, the functions are lost, a, the existing system is difficult to support a multi-domain review environment, when a review object relates to a plurality of disciplines or cross domains, the system often lacks an effective mechanism to deal with different review flows, b, the multi-project parallel capability is lost, the review efficiency is limited, c, the scoring standards of different domains are different, most of the systems lack a multi-type calculation engine, the score calculation is dependent on manpower, the cost is increased, d, the scoring mode is single, only a fixed mode is supported, and multi-dimensional weight scoring cannot be realized at the same time, e, the inter-department information barrier is serious, the cross-department coordination difficulty is caused by forming a data island, the coordination efficiency is low, f, the custom query and export functions are lost, and the data query and export functions are limited; 3) Expert management out-of-order, namely, a, unstructured expert resources, fuzzy images, lack of key information such as expertise, historical experience, avoidance relation and the like, and incapability of accurately matching review requirements; 4) A, a process visualization blind area, wherein the review progress is opaque, the state of a key node is difficult to track, a large-scale screen cannot be dynamically presented, and the decision lacks instant data support; 5) The security guarantee is insufficient, the data backup and desensitization mechanism is weak, the privately-arranged options and the operation audit capability are lacked, and the risks of data leakage and flow runaway exist. Disclosure of Invention The invention provides a dynamic rule-driven multi-field review method and a dynamic rule-driven multi-field review system, which are characterized in that stiffness is broken through a dynamic flow engine, custom configuration of multi-scene review paths and expert authorities are supported, minute-level response of policy change is realized, a multi-dimensional intelligent review center enables complex business, multi-system scoring is integrated, automatic weighted calculation and rule verification can support multi-project parallel processing, a global expert library and intelligent portraits are accurately matched, resources are centrally managed, a multi-dimensional capacity model is constructed, a double-blind review mechanism is embedded to ensure fairness, and safety is ensured through data backup, data desensitization, privately deployment and complete audit logs. In order to achieve the technic