Search

CN-122019706-A - Problem processing method and device based on artificial intelligence, computer equipment and medium

CN122019706ACN 122019706 ACN122019706 ACN 122019706ACN-122019706-A

Abstract

The application belongs to the technical field of artificial intelligence, and relates to a problem processing method based on artificial intelligence, which comprises the steps of analyzing input problem data to obtain triples; the method comprises the steps of carrying out interference evaluation on candidate option pairs based on triplets to screen low confidence options, deleting the low confidence options from the candidate options to obtain first options, carrying out counterfactual verification on the first options to screen error options, deleting the error options from the first options to obtain second options, carrying out multipath reasoning on each second option to obtain path reasoning results if the number of the second options is multiple, carrying out confidence evaluation on each path reasoning result to obtain multiple confidence data, carrying out numerical comparison on the confidence data to obtain target options corresponding to target confidence data with highest numerical values, and generating answer data based on the target options and outputting the answer data. The application can be applied to the problem processing scene in the field of financial science and technology, and the answer accuracy of the problem processing is effectively improved through the application.

Inventors

  • WANG JIANZONG
  • ZHANG NAN
  • QU XIAOYANG

Assignees

  • 平安科技（深圳）有限公司

Dates

Publication Date
20260512
Application Date
20260106

Claims (10)

  1. 1. The problem processing method based on artificial intelligence is characterized by comprising the following steps: Receiving question data input by a user; Analyzing the problem data based on a preset semantic decomposition model to obtain a corresponding triplet, wherein the triplet comprises a target variable, a constraint set and a candidate set, and the candidate set comprises a plurality of candidate options; Based on the triplets, performing interference evaluation on the candidate option pairs by using a preset interference evaluation network to screen out low-confidence options, and deleting the low-confidence options from the candidate options to obtain first options; Performing inverse fact verification on the first option based on a preset large language model to screen out a wrong option, and deleting the wrong option from the first option to obtain a second option; If the number of the second options is multiple, carrying out multipath reasoning processing on each second option based on a preset generation frame to obtain path reasoning results corresponding to each second option; carrying out confidence assessment on each path reasoning result to obtain a plurality of corresponding confidence data; Performing numerical comparison on all the confidence data to determine target confidence data with the highest numerical value, and acquiring target options corresponding to the target confidence data; And generating corresponding answer data based on the target options, and outputting the answer data.
  2. 2. The method for processing the problem based on the artificial intelligence according to claim 1, wherein the step of parsing the problem data based on the preset semantic decomposition model to obtain the corresponding triples specifically includes: Preprocessing the problem data to obtain corresponding processing data; carrying out semantic analysis on the processing data based on the semantic decomposition model to obtain a corresponding semantic analysis result; decomposing the problem data based on the semantic analysis result to obtain a corresponding decomposition result; And taking the decomposition result as the triplet.
  3. 3. The method for processing an artificial intelligence based question according to claim 1, wherein the step of performing interference assessment on the candidate option pairs using a preset interference assessment network based on the triplets to screen out low confidence options specifically comprises: calling a preset interference degree evaluation network; for a first designated option, calculating the semantic similarity and the logic conflict degree between the first designated option and the target variable by using the interference degree evaluation network, wherein the first designated option is any one option of all candidate options; Calculating the semantic similarity and the logic conflict degree based on a preset score calculation formula to obtain corresponding score data; judging whether the score data is smaller than a preset score threshold value or not; If yes, the first designated option is determined to be a low confidence option.
  4. 4. The artificial intelligence based question processing method of claim 1, wherein the step of performing counterfactual verification on the first option based on a preset large language model to screen out a wrong option, specifically comprises: generating a hypothesis statement corresponding to a second designated option, wherein the second designated option is any one option in all the first options; Performing fact verification on the reasoning result by using the large language model based on known facts; if the reasoning result is contradictory, calculating the conflict strength between the reasoning result and the known facts; judging whether the conflict strength is larger than a preset conflict threshold value or not; if yes, determining the second designated option as a wrong option.
  5. 5. The artificial intelligence based problem processing method of claim 4, wherein the step of calculating a collision strength between the inference result and the known fact, in particular, comprises: Extracting key rules in the known facts; Counting the degree data of the inference result violating the key rule; Acquiring weight data corresponding to the key rule; Calculating the degree data and the weight data based on a preset conflict strength calculation strategy to obtain a corresponding calculation result; the calculation result is taken as the collision strength between the reasoning result and the known fact.
  6. 6. The method for processing an artificial intelligence based question according to claim 1, wherein the step of performing confidence evaluation on each of the path inference results to obtain a corresponding plurality of confidence data comprises: Acquiring a specified path reasoning result corresponding to a third specified option, wherein the third specified option is any one option of all the second options; acquiring a preset weighted fusion strategy; carrying out weighted calculation processing on the appointed path reasoning result based on the weighted fusion strategy to obtain corresponding calculation data; And taking the calculated data as the specified confidence data of the third specified option.
  7. 7. The artificial intelligence based question processing method of claim 1, further comprising, after the step of generating corresponding answer data based on the target option and outputting the answer data: calling a preset rendering engine; Generating an inference track graph corresponding to the multipath inference processing based on the rendering engine; Acquiring a preset display mode; and carrying out display processing on the reasoning track diagram based on the display mode.
  8. 8. An artificial intelligence based problem-handling device, comprising: The receiving module is used for receiving the problem data input by the user; the analysis module is used for analyzing the problem data based on a preset semantic decomposition model to obtain a corresponding triplet, wherein the triplet comprises a target variable, a constraint set and a candidate set, and the candidate set comprises a plurality of candidate options; the first processing module is used for carrying out interference evaluation on the candidate option pairs by using a preset interference degree evaluation network based on the triplets so as to screen out low-confidence options, and deleting the low-confidence options from the candidate options to obtain first options; the second processing module is used for performing inverse fact verification on the first option based on a preset large language model to screen out a wrong option, and deleting the wrong option from the first option to obtain a second option; The reasoning module is used for carrying out multipath reasoning on each second option based on a preset generating frame if the number of the second options is multiple, so as to obtain path reasoning results corresponding to each second option; The evaluation module is used for carrying out confidence evaluation on each path reasoning result to obtain a plurality of corresponding confidence data; the third processing module is used for carrying out numerical comparison on all the confidence data to determine target confidence data with the highest numerical value, and obtaining target options corresponding to the target confidence data; and the output module is used for generating corresponding answer data based on the target options and outputting the answer data.
  9. 9. A computer device comprising a memory having stored therein computer readable instructions which when executed implement the steps of the artificial intelligence based problem handling method of any of claims 1 to 7.
  10. 10. A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor implement the steps of the artificial intelligence based problem handling method according to any of claims 1 to 7.

Description

Problem processing method and device based on artificial intelligence, computer equipment and medium Technical Field The application relates to the technical field of artificial intelligence, which can be applied to the field of financial science and technology, in particular to a problem processing method, a device, computer equipment and a storage medium based on artificial intelligence. Background In the field of financial insurance, the deep application of artificial intelligence technology has become a key force for promoting industry digital transformation and improving service efficiency and quality. At present, artificial intelligence is widely permeated into a plurality of core business scenes such as risk assessment, claim settlement audit, product recommendation, intelligent customer service and the like, and remarkable benefit improvement and business innovation are brought to financial institutions and insurance enterprises. Most of the existing intelligent systems in the field of financial insurance rely on an end-to-end Large Language Model (LLM) to realize automatic question-answering and decision support functions. The model can rapidly process a large amount of text information by virtue of strong language understanding and generating capability, provides seemingly reasonable answers and decision suggestions for users, and meets the basic requirements of business to a certain extent. However, in the practical application process, the end-to-end large language model exposes various limitations when processing specific complex problems, which seriously affects the accuracy of answer data, and further restricts the further popularization and application of the answer data in the field of financial insurance. Specifically, when facing the problem of multi-choice decision or interference options, the model is extremely prone to the defects of 'cognitive interference' and 'misjudgment and reasoning'. Taking an insurance claim auditing scenario as an example, insurance clauses generally have high expertise and complexity, and situations of similar semantics but significant logic differences may exist between different clauses. When the model needs to precisely select one of a plurality of similar terms which is most suitable for the customer situation, the traditional large language model often has difficulty in accurately grasping the logic relationship among terms and is easily misled by semantic similarity, so that the option which is logically wrong but has the closest semantic meaning is selected. Such misjudgment may not only lead to unfair results of claims, causing customer disputes, but also bring unnecessary economic loss and reputation risks to insurance companies. In summary, the existing financial insurance intelligent system relying on the end-to-end large language model has the problem of lower answer accuracy when processing complex questions, and cannot meet the requirement of industry on high-precision decision support. Therefore, the development of the novel intelligent decision support system capable of effectively overcoming the problems and improving the accuracy of answer data has important practical significance and wide application prospect. Disclosure of Invention The embodiment of the application aims to provide a problem processing method, device, computer equipment and storage medium based on artificial intelligence, so as to solve the technical problem of lower answer accuracy in the existing problem processing mode depending on an end-to-end large language model. In a first aspect, there is provided an artificial intelligence based problem handling method, comprising: Receiving question data input by a user; Analyzing the problem data based on a preset semantic decomposition model to obtain a corresponding triplet, wherein the triplet comprises a target variable, a constraint set and a candidate set, and the candidate set comprises a plurality of candidate options; Based on the triplets, performing interference evaluation on the candidate option pairs by using a preset interference evaluation network to screen out low-confidence options, and deleting the low-confidence options from the candidate options to obtain first options; Performing inverse fact verification on the first option based on a preset large language model to screen out a wrong option, and deleting the wrong option from the first option to obtain a second option; If the number of the second options is multiple, carrying out multipath reasoning processing on each second option based on a preset generation frame to obtain path reasoning results corresponding to each second option; carrying out confidence assessment on each path reasoning result to obtain a plurality of corresponding confidence data; Performing numerical comparison on all the confidence data to determine target confidence data with the highest numerical value, and acquiring target options corresponding to the target confidence data; And