CN-122019581-A - Library reference consultation question-answering method based on generated model for educational scene

CN122019581ACN 122019581 ACN122019581 ACN 122019581ACN-122019581-A

Abstract

The application discloses a library reference consultation question-answering method based on a generated model for an educational scene, which comprises the steps of obtaining and analyzing a reference consultation request text to obtain a structured analysis result containing task types, honest intention clues and the like; the learner portrait is built by combining the course policy parameter set, the comprehensive risk score is calculated, the risk level is determined, the highest allowable output level is determined according to the risk level, the course policy and the task completion degree, the reasoning input containing constraint conditions is built, candidate answers are generated, the score is obtained through the submittability judgment model, and the candidate answers are subjected to degradation and rewriting if the score reaches the standard, so that the compliance answer output is generated. The method realizes the balance of compliance and teaching, effectively avoids academic risk, and improves reference consultation service quality.

Inventors

KONG XIANGYAN
GAO WANQIN
Qin Yancui
TAO LAN
ZHANG LULU

Assignees

江南大学

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. The library reference consultation question-answering method based on the generated model for the education scene is characterized by comprising the following steps of: The method comprises the steps of obtaining a reference consultation request text, and analyzing the request text to obtain a structural analysis result, wherein the structural analysis result comprises a task type, a delivery type, a completion clue and an integrity intention clue; constructing a learner representation based on the structured analysis result and a preset course policy parameter set; Calculating a comprehensive risk score based on the learner representation and the course policy parameter set, and determining a risk level according to the comprehensive risk score and a preset risk threshold; Determining a highest allowable output level according to the risk level, the course policy parameter set and the task completion degree in the learner profile; providing an inference input to a generative model, the inference input comprising constraints extracted from the structured parsing result, the learner representation, and the risk level, generating a candidate answer; processing the candidate answers through a submittability judgment model to obtain a submittability score; And if the submittability score is greater than or equal to a preset submittability threshold, performing a degrading rewrite operation on the candidate replies, generating compliance replies and outputting the compliance replies.
2. The method of claim 1, wherein the step of determining the position of the substrate comprises, The learner portrait comprises an integrity risk tendency score and a task completion degree, wherein the integrity risk tendency score is calculated based on an integrity intention clue in the structural analysis result, and the task completion degree is calculated based on a completion degree clue in the structural analysis result.
3. The method of claim 2, wherein the step of determining the position of the substrate comprises, The comprehensive risk score is obtained by weighted summation of the honest risk tendency score, the complement of the task completion degree and an approximation index, and the approximation index is obtained by calculation based on the transfer times and the transfer amplitude of the user request type in a multi-round reference consultation session.
4. The method of claim 1, wherein the step of determining the position of the substrate comprises, The method comprises the steps of determining an initial upper limit according to a preset highest output level in a course policy parameter set, and restraining the initial upper limit according to the risk level, wherein the highest output level is not higher than a retrieval policy level if the risk level is high, and the output level sequentially comprises a clarification interview level, a retrieval policy level, a reading understanding level, a writing scaffold level and a manuscript-level text level from low to high according to output priority.
5. The method of claim 1, wherein the step of determining the position of the substrate comprises, The generative model reasoning input comprises a set of constraint fields comprising at least the allowed highest output level and a list of forbidden output items derived from the allowed highest output level.
6. The method of claim 1, wherein the step of determining the position of the substrate comprises, The step of processing the candidate answers through the submittability judging model to obtain a submittability score specifically comprises the steps of extracting text features including a continuity index, a arguments integrity index and a request text fitting degree index from the candidate answers, fusing the text features with the risk level and the highest allowable output level, and inputting the fused text features to a classifier to obtain the submittability score.
7. The method of claim 1, wherein the step of determining the position of the substrate comprises, The degrading rewrite operation comprises locating high submittable fragments in the candidate replies according to the trigger reason vector, and rewriting complete arguments in the high submittable fragments into paragraph function descriptions and placeholders, or reserving the structural framework of fragments and removing complete expressions therein.
8. The method of claim 7, wherein the step of determining the position of the probe is performed, The degrading rewrite operation further includes temporarily downregulating the allowed highest output level after the rewrite is performed and outputting a compliance reply containing the required user input field templates.
9. The method of claim 1, wherein the step of determining the position of the substrate comprises, The method further comprises the step of generating a learning step chain, wherein before the candidate answer is generated, the learning step chain comprising a plurality of step nodes is generated according to the task type in the structural analysis result and the highest allowable output level, each step node comprises a required input field and a corresponding verification rule, and current node information of the learning step chain is written into the generated model reasoning input.
10. An educational scenario based library reference consultation question-answering system of a generative model, comprising: the request analysis module is used for acquiring a reference consultation request text, analyzing the request text and obtaining a structural analysis result, wherein the structural analysis result comprises a task type, a delivery type, a completion degree clue and an integrity intention clue; The portrait construction module is used for constructing a learner portrait based on the structural analysis result and a preset course policy parameter set; The risk assessment module is used for calculating a comprehensive risk score based on the learner portrait and the course policy parameter set, and determining a risk level according to the comprehensive risk score and a preset risk threshold; The hierarchy control module is used for determining an allowable highest output hierarchy according to the risk level, the course policy parameter set and the task completion degree in the learner portrait; A candidate generation module for providing an inference input to a generative model, the inference input comprising constraints extracted from the structured parsing result, the learner representation, and the risk level, generating a candidate answer; The submittability judging module is used for processing the candidate answers through a submittability judging model to obtain a submittability score; And the output rewrite module is used for executing a degrading rewrite operation on the candidate answers if the submittability score is greater than or equal to a preset submittability threshold value, generating a compliance answer and outputting the compliance answer.

Description

Library reference consultation question-answering method based on generated model for educational scene Technical Field The application relates to the technical field of educational information service and natural language processing, in particular to a library reference consultation question-answering method based on a generated model for an educational scene. Background In educational scenes such as libraries in universities and middle schools, disciplines, service platforms in disciplines, and the like, online reference consulting service has become an important means for supporting the autonomous learning of learners, and the core requirement is to avoid the risk of the academic inadequacy while providing academic support. Currently, with the wide application of generative models in the field of natural language processing, related technologies are gradually introduced into reference consulting services to improve response efficiency and content richness. However, the prior art has significant drawbacks in the practical application of educational scenes. On one hand, with reference to the boundary blurring of the inadequacy of consultation and academic, a user can gradually approach the work finalization which can be directly submitted through multiple rounds of dialogue, and the existing model usually adopts a simple refusing or unconstrained free generation mode, either the reduced text which can be directly submitted or the complete finalization is output, so that the compliance is insufficient, or the effective heuristic coaching cannot be provided for avoiding the excessive refusal of the violation, and the teaching support effect is limited. On the other hand, the prior art lacks effective verification on the submittability of the output text, can not automatically judge whether the text meets the submission requirements and carries out degradation treatment, is difficult to dynamically adjust the output granularity according to the policy requirements of different courses and the actual state of a learner, has poor flow consistency and low reproducibility, and seriously influences the quality of reference consultation service and the implementation of academic integrity specifications. Disclosure of Invention The application embodiment provides a library reference consultation question-answering method based on a generated model for an educational scene, which aims to at least solve the technical problems existing in the related art. According to a first aspect of the embodiment of the application, there is provided a library reference consultation question-answering method based on a generative model for an educational scene, comprising: The method comprises the steps of obtaining a reference consultation request text, and analyzing the request text to obtain a structural analysis result, wherein the structural analysis result comprises a task type, a delivery type, a completion clue and an integrity intention clue; constructing a learner representation based on the structured analysis result and a preset course policy parameter set; Calculating a comprehensive risk score based on the learner representation and the course policy parameter set, and determining a risk level according to the comprehensive risk score and a preset risk threshold; Determining a highest allowable output level according to the risk level, the course policy parameter set and the task completion degree in the learner profile; providing an inference input to a generative model, the inference input comprising constraints extracted from the structured parsing result, the learner representation, and the risk level, generating a candidate answer; processing the candidate answers through a submittability judgment model to obtain a submittability score; And if the submittability score is greater than or equal to a preset submittability threshold, performing a degrading rewrite operation on the candidate replies, generating compliance replies and outputting the compliance replies. The learner representation comprises an integrity risk tendency score and a task completion degree, wherein the integrity risk tendency score is calculated based on an integrity intention clue in the structural analysis result, and the task completion degree is calculated based on a completion degree clue in the structural analysis result. As an alternative scheme, the comprehensive risk score is obtained by weighting and summing the integrity risk tendency score, the complement value of the task completion degree and an approximation index, wherein the approximation index is obtained by calculating the transfer times and the amplitude of the user request type in a multi-reference consultation session. The step of determining the allowed highest output level includes determining an initial upper limit according to a preset highest output level in the course policy parameter set, and constraining the initial upper limit according to the risk level, where