CN-121981448-A - Distributed auditing method, system and storage medium for community content

CN121981448ACN 121981448 ACN121981448 ACN 121981448ACN-121981448-A

Abstract

The application provides a community content distributed auditing method, system and storage medium, which comprises the steps of receiving community content, performing machine primary screening on the community content to identify and mark a to-be-audited task which cannot confirm compliance, selecting a plurality of auditors from an auditor pool to form a distributed auditing group aiming at the to-be-audited task, distributing the to-be-audited task to each auditor in the distributed auditing group to obtain a plurality of independent, obtaining real-time secret dense credit scores of each auditor, determining dynamic weights of each auditor in the voting judgment according to the real-time secret dense credit scores, performing weighted aggregation calculation on final auditing results of the to-be-audited task based on voting judgment opinions of each auditor and the corresponding dynamic weights of the auditors, executing processing operation on the community content according to the final auditing results, and displaying the processing process and results of the to-be-audited task in the community. The application can improve the accuracy and the anti-interference capability of the auditing result.

Inventors

HAN XIAO
FAN LEI
LI SHUYING
XU CHUANTAI

Assignees

杭州小码教育科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260105

Claims (10)

1. A community content distributed auditing method, the method comprising: Receiving community content submitted by a user, and performing machine preliminary screening on the community content to identify and mark a task to be checked for which the machine cannot determine compliance; Selecting a plurality of auditors from a preset auditor pool to form a distributed audit group aiming at the task to be audited, and distributing the task to be audited to all auditors in the distributed audit group to acquire a plurality of independent auditors; acquiring real-time secret dense credit scores of auditors in the distributed audit group, and determining dynamic weights of each auditor in the voting judgment according to the real-time secret dense credit scores; Performing weighted aggregation calculation based on voting judgment opinions of each auditor in the distributed audit group and corresponding dynamic weights thereof to generate a final audit result of the task to be audited; and executing the processing operation of the community content according to the final auditing result, and displaying the processing process and the result of the task to be audited in the community.
2. The method of claim 1, wherein the selecting a plurality of auditors from a pre-set pool of auditors to form a distributed audit team for the task under audit comprises: Acquiring a multi-dimensional capability portrait vector of each auditor in an auditor pool, wherein the multi-dimensional capability portrait vector is dynamically generated based on historical judgment accuracy and expertise field understanding of the auditor on a plurality of violation types; Extracting a violation feature vector of the task to be checked, wherein the violation feature vector represents the probability or strength that the task to be checked belongs to various violations; calculating the matching degree of each auditor and the task to be audited based on the multi-dimensional capability image vector and the violation feature vector; And introducing auditors credit distribution constraint and capability view angle difference constraint on the basis of the matching degree, and processing the matching degree, the auditors credit distribution constraint and the capability view angle difference constraint through an optimization algorithm to select auditors from the auditor pool so as to form a distributed audit group with optimal overall matching degree and credit and view angle diversity.
3. The method of claim 1, wherein the obtaining a real-time secret dense reputation score for each auditor in the distributed audit team comprises: Obtaining a dynamic reputation score vector of each auditor, wherein the dynamic reputation score vector is generated based on historical audit behaviors and at least comprises an accuracy reputation score and a value discovery reputation score; Processing the dynamic credit component through a preset weight function to determine the basic voting weight of each auditor; Acquiring voting judgment opinions submitted by each auditor under the task to be audited, and analyzing voting similarity among auditors by combining the historical audit behaviors of each auditor so as to identify potential abnormal collaborative groups; Dynamically calculating a implicit consensus cohesive force penalty factor based on the consistency relation of the auditor and the abnormal collaborative community to which the auditor belongs on the voting opinion; And multiplying the basic voting weight and the consensus cohesive force penalty factor to generate a real-time secret dense credit score.
4. A method according to claim 3, wherein said determining the dynamic weight of each auditor in the current vote decision from the real-time secret dense reputation score comprises: Acquiring classification features or dispute assessment parameters of the task to be examined, and determining corresponding punishment factor adjustment parameters from a preset adjustment strategy set based on the classification features or the dispute assessment parameters; dynamically adjusting each consensus cohesive force penalty factor by using the penalty factor adjusting parameters to obtain a task adaptive penalty factor corresponding to each auditor; and multiplying the basic voting weight corresponding to the same auditor with the task adaptive penalty factor to obtain the dynamic weight of the auditor in the voting judgment.
5. The method of claim 2, wherein the generating the final audit result for the task under audit based on the voting judgment opinions of each auditor in the distributed audit group and their corresponding dynamic weights by weighted aggregate calculation comprises: obtaining a secret letter scoring vector submitted by each auditor in the distributed auditing group for the task to be audited, wherein the secret letter scoring vector comprises a violation confidence score for representing the violation degree of content, a hazard grade score for representing the hazard grade and an urgency score for representing the processing priority; weighting calculation is carried out on the submitted violation confidence score, the hazard grade score and the urgency score based on the dynamic weight corresponding to each auditor to obtain a weighted violation confidence set, a weighted hazard grade set and a weighted urgency set; Respectively carrying out aggregation operation on the violation confidence coefficient set, the hazard level set and the urgency set to obtain a comprehensive violation index, a comprehensive hazard level and a comprehensive urgency level of the task to be checked; Comparing the comprehensive violation index, the comprehensive hazard level and the comprehensive urgency level with a preset gradient judging threshold to obtain a comparison result, and mapping the comparison result to a corresponding processing operation level to generate a final auditing result of the task to be audited, wherein the processing operation level at least comprises content folding, content replacement prompting and direct deleting.
6. The method of claim 5, wherein the exposing the process and results of the pending task within the community further comprises: Receiving a complaint request initiated by a user aiming at a public final auditing result and associated point mortgage information, responding to the complaint request, and distributing a complaint judging task to a public juggler set randomly selected from community users; And receiving a voting result of support original judgement or support complaints returned by the public juggle set, processing the voting result by using a preset game reward punishment rule to judge whether the complaint request is satisfied, if so, executing a content processing operation opposite to the final auditing result, and carrying out community integral reward punishment on the complaint initiator or the public juggle of the voting according to the game reward punishment rule.
7. The method of claim 6, wherein the method further comprises: acquiring high-value disputed cases accumulated in the complaint request processing process, and carrying out natural language processing and cluster analysis on the high-value disputed cases to identify new implied violation expression patterns and audit strategy blind areas; Generating at least one of updating characteristics of a machine primary screening model, updating questions of an auditor capability test and optimization suggestions of audit strategy parameters based on the new violation expression mode and the audit strategy blind area; And dynamically adjusting portrait vectors of the machine prescreening, the auditor pool or the gradient decision threshold according to the updating characteristics, the updating titles or the optimization suggestions.
8. The method of claim 7, wherein generating at least one of an update feature of a machine prescreening model, an update topic of an auditor capability test, and an optimization suggestion of audit policy parameters based on the new violation expression pattern and audit policy blind zone comprises: extracting key semantic features of the new violation expression pattern and context association rules to generate updated features of a machine preliminary screening model; Converting the auditing strategy blind area into auditing capability test questions comprising specific scene descriptions and standard answers to generate updated questions of an auditor capability test; And obtaining comparison analysis of the processing result distribution of the high-value dispute case and the gradient arbitration threshold value, and determining a threshold value adjustment direction and amplitude based on the processing result distribution and the comparison analysis to generate optimization suggestions of auditing strategy parameters.
9. A community content distributed auditing system is characterized by comprising an initial auditing module, an auditing personnel module, an auditing processing module and an auditing display module, The initial auditing module is used for receiving community contents submitted by users, performing machine primary screening on the community contents to identify and mark to-be-audited tasks for which the machine cannot determine compliance; the auditor module is used for selecting a plurality of auditors from a preset auditor pool to form a distributed audit group aiming at the task to be audited, and distributing the task to be audited to all auditors in the distributed audit group so as to obtain a plurality of independent voting judgment opinions; The auditing processing module is used for acquiring real-time secret dense credit scores of auditors in the distributed auditing group, determining dynamic weights of each auditor in the voting judgment according to the real-time secret dense credit scores, carrying out weighted aggregation calculation based on voting judgment opinions of each auditor in the distributed auditing group and the corresponding dynamic weights thereof to generate a final auditing result of the task to be audited, and executing processing operation on the community content according to the final auditing result; and the auditing and displaying module is used for displaying the processing process and the result of the task to be audited in the community.
10. A computer readable storage medium having stored thereon a computer program executable on a processor, wherein the computer program when executed by the processor implements a community content distributed auditing method according to any of claims 1 to 8.

Description

Distributed auditing method, system and storage medium for community content Technical Field The application relates to the technical field of internet information processing, in particular to a community content distributed auditing method, a community content distributed auditing system and a storage medium. Background With the rapid growth of the content scale of the internet community, the traditional content auditing mode has difficulty in considering efficiency, accuracy and fairness. The conventional auditing method mainly comprises the following two types of pure machine auditing, namely that the auditing is carried out automatically by depending on keyword matching or a pre-training model, but the misjudgment or missed judgment is easily caused due to lack of understanding of a complex context, and the auditing is carried out by platform professional staff, so that the auditing can be carried out on complex scenes, but is high in cost and slow in response, and is difficult to adapt to massive content scenes. In order to balance efficiency and quality, a crowdsourcing auditing mechanism is introduced in the prior art, namely community users participate in content auditing. For example, by publicly recruiting administers, multiple users vote on the same content and assign a fixed weight based on majority rules or simple historical accuracy rates, thereby forming audit conclusions. However, the method still has the remarkable defects that firstly, the voting weight of the auditors is often statically set based on the historical expression, the current professional ability and judgment credibility of the auditors cannot be reflected in real time, and secondly, the voting result is easily influenced by small group manipulation or group deviation due to the lack of a dynamic inhibition mechanism for potential collusion, scoring or popular behavior among the auditors, so that the fairness and reliability of the auditing result are influenced. Disclosure of Invention In order to improve the accuracy and the anti-interference capability of the auditing result. The embodiment of the application provides a community content distributed auditing method, a community content distributed auditing system and a storage medium. In a first aspect, there is provided a community content distributed auditing method, the method comprising: Receiving community content submitted by a user, and performing machine preliminary screening on the community content to identify and mark a task to be checked for which the machine cannot determine compliance; Selecting a plurality of auditors from a preset auditor pool to form a distributed audit group aiming at the task to be audited, and distributing the task to be audited to all auditors in the distributed audit group to acquire a plurality of independent auditors; acquiring real-time secret dense credit scores of auditors in the distributed audit group, and determining dynamic weights of each auditor in the voting judgment according to the real-time secret dense credit scores; Performing weighted aggregation calculation based on voting judgment opinions of each auditor in the distributed audit group and corresponding dynamic weights thereof to generate a final audit result of the task to be audited; and executing the processing operation of the community content according to the final auditing result, and displaying the processing process and the result of the task to be audited in the community. In some embodiments, the selecting a plurality of auditors from a preset auditor pool to form a distributed audit group for the task to be audited includes: Acquiring a multi-dimensional capability portrait vector of each auditor in an auditor pool, wherein the multi-dimensional capability portrait vector is dynamically generated based on historical judgment accuracy and expertise field understanding of the auditor on a plurality of violation types; Extracting a violation feature vector of the task to be checked, wherein the violation feature vector represents the probability or strength that the task to be checked belongs to various violations; calculating the matching degree of each auditor and the task to be audited based on the multi-dimensional capability image vector and the violation feature vector; And introducing auditors credit distribution constraint and capability view angle difference constraint on the basis of the matching degree, and processing the matching degree, the auditors credit distribution constraint and the capability view angle difference constraint through an optimization algorithm to select auditors from the auditor pool so as to form a distributed audit group with optimal overall matching degree and credit and view angle diversity. In some of these embodiments, the obtaining a real-time secret dense reputation score for each auditor in the distributed audit team includes: Obtaining a dynamic reputation score vector of each auditor, wherein the dynamic reputat