Search

CN-122021603-A - Detection report generation method, device, equipment and storage medium

CN122021603ACN 122021603 ACN122021603 ACN 122021603ACN-122021603-A

Abstract

The application belongs to the technical field of artificial intelligence, and relates to a detection report generation method, a device, equipment and a storage medium, wherein semantic fingerprints are generated by inputting data to a target; comparing with semantic fingerprints of all cases in a historical case library, screening out target cases, determining the category of target input data, generating a preliminary detection report, converting the detection processing step into a checking basis in a natural language text form, and supplementing the checking basis into the preliminary detection report to generate a final detection report. The method is applied to an insurance claim detection report generation scene, can generate a final detection report containing an examination basis, avoids the problem that the detection report only has a result item and has lower detection confidence caused by no detection reason, improves the detection confidence, and meanwhile, is convenient for assisting an examination auditor and an applicant to examine and understand due to the examination basis in a natural language text form.

Inventors

  • QU XIAOYANG

Assignees

  • 平安科技（深圳）有限公司

Dates

Publication Date
20260512
Application Date
20260113

Claims (10)

  1. 1. A method for generating a detection report, comprising the steps of: Carrying out multi-mode feature extraction on target input data and generating semantic fingerprints; Comparing the semantic fingerprints with the semantic fingerprints of all cases in the historical case library, and screening out target cases; Determining the category of the target input data according to the classification category of the target case; Taking the category of the target input data as a detection result, and generating a preliminary detection report; the processing steps of multi-modal feature extraction, semantic fingerprint generation, semantic fingerprint comparison, target case screening and category determination are carried out, and a natural language text form examination basis is output through preset natural language model conversion; and supplementing the examination basis into the preliminary detection report to generate a final detection report.
  2. 2. The method for generating a detection report according to claim 1, wherein the step of performing multi-modal feature extraction on the target input data and generating a semantic fingerprint specifically includes: uniformly encoding target input data by adopting a multi-mode uniform encoding framework, and extracting multi-mode characteristics, wherein the multi-mode characteristics have uniform characteristic encoding formats; Mapping the multi-modal features to a preset multi-dimensional semantic representation space, and generating the semantic fingerprint.
  3. 3. The inspection report generation method of claim 1, wherein prior to performing the step of comparing the semantic fingerprints with the semantic fingerprints of all cases in a historical case base to screen out target cases, the method further comprises: acquiring all cases in the historical case library; Inputting all cases in the historical case library into a target classification engine; Classifying all cases in the historical case library according to the latest deployed case classification rule in the target classification engine; Obtaining a compliance case and a non-compliance case under the current latest case classification rule; Respectively extracting multi-mode features of all the compliance cases to generate semantic fingerprints corresponding to each compliance case respectively; The semantic fingerprints corresponding to each compliance case are compared and extracted, and the common semantic fingerprints of all the compliance cases are extracted and used as first common semantic fingerprints; carrying out multi-mode feature extraction on all the non-compliance cases to generate semantic fingerprints corresponding to each non-compliance case respectively; And comparing and extracting semantic fingerprints corresponding to each non-compliance case respectively, and extracting common semantic fingerprints of all the non-compliance cases to serve as second common semantic fingerprints.
  4. 4. The inspection report generating method of claim 3, wherein prior to performing the step of entering all cases in the historical case library into a target classification engine, the method further comprises: Monitoring a case classification rule influence file in real time; when the case classification rule influences that the file content changes, the changed file content is identified; detecting all case classification rules related to the changed file content, and updating the case classification rules according to the changed file content to obtain the current latest case classification rules; And updating and deploying the current latest case classification rule into the target classification engine.
  5. 5. The method for generating a test report according to claim 3, wherein the step of comparing the semantic fingerprint with the semantic fingerprints of all cases in the historical case library to screen out the target case comprises: Comparing the semantic fingerprint with a first common semantic fingerprint and comparing the semantic fingerprint with a second common semantic fingerprint; if the semantic fingerprints are identified to fully contain the first common semantic fingerprint, the target input data belongs to the compliance input data; Comparing the semantic fingerprints with the semantic fingerprints corresponding to each compliance case respectively, calculating fingerprint similarity, and obtaining the compliance case when the fingerprint similarity is the maximum value as the target case; if the semantic fingerprints are identified to fully contain the second common semantic fingerprint, the target input data belongs to non-compliant input data; and comparing the semantic fingerprint with the semantic fingerprint corresponding to each non-compliance case, calculating the similarity of the fingerprints, and obtaining the non-compliance case when the similarity of the fingerprints is the maximum value as the target case.
  6. 6. The method for generating a detection report according to claim 1, wherein the steps of processing the multimodal feature extraction, the semantic fingerprint generation, the semantic fingerprint comparison, the target case screening and the category determination, through the conversion of a preset natural language model, output the examination basis of the natural language text form, specifically comprise: Obtaining a processing result of each processing step according to the processing steps of multi-mode feature extraction, semantic fingerprint generation, semantic fingerprint comparison, target case screening and category determination; Acquiring prompt words preset for each processing step and connection words preset for continuous processing steps; Filling the prompting words, the connecting words and the processing results of each processing step into a text filling template provided by the natural language model as filling contents; outputting the examination basis of the natural language text form formed by the prompt word, the connection word, the processing result of each processing step and the text filling template.
  7. 7. The inspection report generating method as set forth in claim 6, wherein the step of filling the prompt word, the connective word, and the processing result of each processing step together as filling contents into a text filling template provided by the natural language model specifically includes: obtaining text filling templates corresponding to each processing step respectively through screening; Filling the prompting words, the connecting words and the processing results of each processing step into corresponding text filling templates; And (3) connecting the text filling templates corresponding to each processing step after filling in series according to the sequence of the processing steps to generate an integrated text summary content.
  8. 8. A detection report generating apparatus, comprising: The semantic fingerprint generation module is used for carrying out multi-mode feature extraction on target input data and generating semantic fingerprints; the target case screening module is used for comparing the semantic fingerprints with the semantic fingerprints of all cases in the historical case library and screening out target cases; the data category determining module is used for determining the category of the target input data according to the classification category of the target case; the preliminary detection report generation module is used for generating a preliminary detection report by taking the category of the target input data as a detection result; the interpretability conversion module is used for carrying out processing steps of multi-modal feature extraction, semantic fingerprint generation, semantic fingerprint comparison, target case screening and category determination, and outputting a natural language text form examination basis through preset natural language model conversion; and the final detection report generation module is used for supplementing the examination basis into the preliminary detection report and generating a final detection report.
  9. 9. A computer device comprising a memory and a processor, the memory having stored therein computer readable instructions which when executed by the processor implement the steps of the detection report generation 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 detection report generation method of any of claims 1 to 7.

Description

Detection report generation method, device, equipment and storage medium Technical Field The application relates to the technical field of artificial intelligence, and relates to a detection report generation method, device, equipment and storage medium, which are applied to a scene of detection report generation of a result of claim auditing. Background In the intelligent financial wind control scene, for example, tasks such as insurance claim auditing have extremely high requirements on the interpretability of the model, and particularly, most claim applicant does not have professional legal matters or insurance knowledge when vehicle insurance claim auditing and medical insurance claim auditing are carried out, and the claim applicant also often carries regulations or business regulations when introducing whether the claim applicant is in compliance, so that the claim applicant has difficulty in understanding. In recent years, although some auxiliary claim auditing models are presented, the claim auditing models generally only output final judging results, for example, the claim application does not pass, the claim application does not meet regulations, and the like, no more valuable contents can be provided for the principal, and the service detection result with only result items and no relevant explanation reason can cause the problem that the customer doubts the detection result and even believes the detection result. Disclosure of Invention The embodiment of the application aims to provide a detection report generation method, a device, equipment and a storage medium, which are used for solving the technical problem of low detection confidence caused by service detection results with only result items and no related explanation reason. In a first aspect, an embodiment of the present application provides a method for generating a detection report, which adopts the following technical scheme: A detection report generation method, comprising the steps of: Carrying out multi-mode feature extraction on target input data and generating semantic fingerprints; Comparing the semantic fingerprints with the semantic fingerprints of all cases in the historical case library, and screening out target cases; Determining the category of the target input data according to the classification category of the target case; Taking the category of the target input data as a detection result, and generating a preliminary detection report; the processing steps of multi-modal feature extraction, semantic fingerprint generation, semantic fingerprint comparison, target case screening and category determination are carried out, and a natural language text form examination basis is output through preset natural language model conversion; and supplementing the examination basis into the preliminary detection report to generate a final detection report. In a second aspect, an embodiment of the present application further provides a detection report generating device, which adopts the following technical scheme: A detection report generating apparatus comprising: The semantic fingerprint generation module is used for carrying out multi-mode feature extraction on target input data and generating semantic fingerprints; the target case screening module is used for comparing the semantic fingerprints with the semantic fingerprints of all cases in the historical case library and screening out target cases; the data category determining module is used for determining the category of the target input data according to the classification category of the target case; the preliminary detection report generation module is used for generating a preliminary detection report by taking the category of the target input data as a detection result; the interpretability conversion module is used for carrying out processing steps of multi-modal feature extraction, semantic fingerprint generation, semantic fingerprint comparison, target case screening and category determination, and outputting a natural language text form examination basis through preset natural language model conversion; and the final detection report generation module is used for supplementing the examination basis into the preliminary detection report and generating a final detection report. In a third aspect, an embodiment of the present application further provides a computer device, which adopts the following technical scheme: a computer device comprising a memory having stored therein computer readable instructions which when executed by a processor implement the steps of the detection report generation method described above. In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, which adopts the following technical solutions: A computer readable storage medium having stored thereon computer readable instructions which when executed by a processor implement the steps of the detection report generation method as describe