CN-122021870-A - Method, device, electronic equipment and storage medium for assisting intelligent interrogation of smuggled cases

CN122021870ACN 122021870 ACN122021870 ACN 122021870ACN-122021870-A

Abstract

The application provides a method, a device, electronic equipment and a storage medium for assisting intelligent interrogation of a smuggling case, which are applied to the field of natural language processing, wherein the method comprises the steps of performing wrongly written word detection and semantic correction on a written text to obtain a corrected reference written text; the method comprises the steps of carrying out semantic analysis on a reference stroke text to extract key elements and generate structured information, carrying out matching analysis on the structured information and a interrogation strategy label system to output an information missing label list, calling an inquiry propulsion strategy from a strategy template library according to the information missing label list, carrying out logic consistency detection on the structured information to obtain a contradiction point list and an inquiry suggestion in the structured information, and constructing a standardized case information table. The method can automatically complete extraction of the elements of the stroke records, language correction, contradiction detection and relation map construction, effectively assist the interrogation personnel to find logic loopholes, inquire key details, generate a standard document and greatly reduce manual burden.

Inventors

Yue Guocan
CHEN YUN
CHEN JINHUA
SU ZAITIAN
CAI LILI
HUANG ZHUO
LIN ZHIJUN
YANG TIANQI

Assignees

厦门市美亚柏科信息安全研究所有限公司

Dates

Publication Date: 20260512
Application Date: 20251216

Claims (10)

1. A method for intelligent interrogation assistance of smuggled cases, the method comprising: based on priori knowledge of the smuggling cases, performing wrongly written word detection and semantic error correction on the input written text to obtain an error corrected reference written text and an error position list, wherein the error position list is used for verifying the error corrected reference written text; Carrying out semantic analysis on the reference stroke text to extract key elements and generating structural information based on the key elements, wherein the structural information comprises at least one of time, place, personnel, articles, amount, case abstract and inquiry content abstract of the occurrence of a smuggling case; Carrying out matching analysis on the structured information and a preset interrogation strategy tag system, and outputting an information missing tag list; according to the information missing tag list, a corresponding inquiry propulsion strategy is called from a strategy template library, wherein the interrogation strategy tag system comprises three types of tags of subjective purpose, personnel status and smuggling ration; performing logic consistency detection on the structured information based on a conflict rule base and a graph neural network hybrid reasoning mechanism to obtain a conflict point list and corresponding overtime suggestions in the structured information, wherein the conflict rule base comprises a smuggling time conflict rule, an article quantity conflict rule, an article price tax rate matching rule and a transportation path non-compliance rule; Fusing the inquiry propulsion strategy and the inquiry suggestion to generate an integrated inquiry guiding scheme; Based on the smuggling field ontology and the relation extraction model, carrying out relation recognition on key elements in the structural information, and constructing an entity relation map and a standardized case information table; Receiving an updated written text generated after the interrogation is conducted according to the interrogation guiding scheme, and re-executing the steps of wrongly written word detection, semantic error correction, semantic analysis and structured information generation on the updated written text to obtain updated structured information; and carrying out multidimensional weighted evaluation on the updated structured information, the standardized case information table, the entity relation map and the logic consistency detection result by adopting an analytic hierarchy process, and outputting a stroke quality grading grade and a targeted improvement suggestion.
2. The method according to claim 1, wherein the prior knowledge includes a corresponding legal provision of the smuggling case and a preset word library of the smuggling field, the error word detection and the semantic correction are performed on the input written text based on the prior knowledge of the smuggling case, and the error-corrected reference written text and the error location list are obtained, including: According to the corresponding legal provision and the character collocation standard in the preset word library, determining characters with the occurrence frequency of character pairs in the written text lower than a first preset threshold value as error contents based on a context-associated neural network, and determining the positions of the error contents as an error position list; And correcting the error content by adopting a sequence-to-sequence model according to the corresponding legal provision and the preset error word library to obtain corrected reference stroke text.
3. The method of claim 1, wherein extracting key elements by semantic parsing the reference transcript text and generating structured information based on the key elements comprises: Based on a preset dictionary in the smuggling field, carrying out semantic analysis and key element extraction on the reference stroke text to obtain key elements of the smuggling case, wherein the key elements comprise at least one of time, place, personnel, articles and amount of the smuggling case; Generating a case abstract of the smuggling case based on the key element under the condition that the matching degree between the key element and the standard element is larger than or equal to a second preset threshold value; generating an inquiry content abstract based on a text processing algorithm and the preset dictionary, wherein the word number of the inquiry content abstract is smaller than or equal to a third preset threshold value; and carrying out structuring treatment on the key elements, the case abstract and the inquiry content abstract to obtain the structuring information.
4. The method of claim 1, wherein the matching the structured information with a preset interrogation policy tag system, outputting a missing information tag list, comprises: analyzing whether the structured information is completely matched with the interrogation strategy label system based on a pre-training language model and a domain classifier; Outputting an information missing tag list under the condition that the structured information is not completely matched with the interrogation strategy tag system; before the corresponding query propulsion strategy is invoked from the strategy template library, the method further comprises the following steps: Acquiring a question angle, a question mode, a key inquiry and an example question corresponding to the information missing tag list from the strategy template library; determining sentence similarity between the example questions; And generating an inquiry pushing strategy based on the question angle, the question mode, the key inquiry and the example problem under the condition that the sentence similarity is smaller than a fourth preset threshold.
5. The method of claim 1, wherein the performing logic consistency detection on the structured information based on the contradictory rule base and the graph neural network hybrid inference mechanism to obtain a contradictory point list and a corresponding inquiry suggestion in the structured information comprises: Confidence conflict detection is carried out on the structured information based on a contradictory rule base and a graph neural network mixed reasoning mechanism so as to determine the confidence of each information in the structured information; determining information with the confidence coefficient smaller than a fifth preset threshold value in the structured information as a contradictory point list; And determining the overtaking problem of the contradiction point list based on the context semantics in the contradiction point list and the structural information.
6. The method of claim 1, wherein the smuggling domain ontology includes a personnel role, the transport attribute, and a region of interest code, wherein the performing relationship recognition on key elements in the structured information based on the smuggling domain ontology and a relationship extraction model, and constructing an entity relationship map and a standardized case information table includes: Extracting association relations among key elements in the structured information based on the personnel roles, the transport means attributes, the region-of-interest codes and an RE-DEVO model, and constructing an entity relation map; And carrying out standardization processing on the association relation and the structured information to obtain a standardized case information table.
7. The method of claim 1, wherein the performing multi-dimensional weighted evaluation on the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency detection result by using a hierarchical analysis method, and outputting a quality score level of a transcript and a targeted improvement suggestion comprises: Acquiring missing evidence corresponding to the smuggling case; identifying the missing evidence, the structured information and the standardized case information table to obtain logic contradiction between the structured information and the missing evidence and program non-standard information; performing weighted comprehensive evaluation on the logic contradiction and the program non-standard information by adopting an analytic hierarchy process to obtain a stroke quality grading grade; And generating an improvement suggestion based on the score grade of the quality of the strokes and a preset suggestion library.
8. An apparatus for intelligent interrogation assistance of smuggled cases, the apparatus comprising: the error correction module is used for carrying out wrongly written word detection and semantic error correction on the input written text based on priori knowledge of the smuggling case to obtain an error corrected reference written text and an error position list, wherein the error position list is used for verifying the error corrected reference written text; The extraction and generation module is used for carrying out semantic analysis on the reference stroke text to extract key elements and generating structural information based on the key elements, wherein the structural information comprises at least one of time, place, personnel, articles, amount, case abstract and inquiry content abstract of the occurrence of a smuggling case; The analysis and calling module is used for carrying out matching analysis on the structured information and a preset interrogation strategy label system and outputting an information missing label list; according to the information missing tag list, a corresponding inquiry propulsion strategy is called from a strategy template library, wherein the interrogation strategy tag system comprises three types of tags of subjective purpose, personnel status and smuggling ration; The logic detection module is used for carrying out logic consistency detection on the structured information based on a conflict rule base and a graph neural network mixed reasoning mechanism to obtain a conflict point list and a corresponding overtime suggestion in the structured information, wherein the conflict rule base comprises a smuggling time conflict rule, an article quantity conflict rule, an article price tax rate matching rule and a transportation path non-compliance rule; The fusion and generation module is used for fusing the inquiry propulsion strategy and the inquiry suggestion to generate an integrated inquiry guiding scheme; The recognition and construction module is used for carrying out relationship recognition on key elements in the structural information based on the smuggling field ontology and the relationship extraction model, and constructing an entity relationship map and a standardized case information table; The receiving and generating module is used for receiving an updated written text generated after the interrogation is conducted according to the interrogation guiding scheme, and re-executing the steps of wrongly written word detection, semantic error correction, semantic analysis and structured information generation on the updated written text to obtain updated structured information; And the evaluation module is used for carrying out multidimensional weighted evaluation on the updated structured information, the standardized case information table, the entity relation graph and the logic consistency detection result by adopting an analytic hierarchy process and outputting a stroke quality scoring grade and a targeted improvement suggestion.
9. An electronic device, the electronic device comprising: a memory for storing executable program code; a processor for calling and running the executable program code from the memory, causing the electronic device to perform the method of any one of claims 1 to 7.
10. A computer readable storage medium, characterized in that the computer readable storage medium stores executable program code, which when executed, implements the method of any of claims 1 to 7.

Description

Method, device, electronic equipment and storage medium for assisting intelligent interrogation of smuggled cases Technical Field The present application relates to the field of natural language processing, and more particularly, to a method, an apparatus, an electronic device, and a storage medium for intelligent interrogation assistance of a smuggling case. Background The record is mainly a statement of a record answer person and is an important basis for case description and qualitative, however, in the prior art of examination and interrogation of a smuggling case, the record arrangement, information extraction, strategy formulation, contradiction check and other works are manually finished, the target information cannot be automatically extracted, the content of the record cannot be structurally analyzed according to the target information, only a questioner can be assisted by the provided fixed prompt to manually record, and the reliability of the record is low because the target information cannot be automatically extracted. In the related art, the auxiliary tool with artificial intelligence assists the interrogation process, however, the accuracy and reliability of the artificial intelligence technology in the interrogation scene are insufficient, and the voice recognition, emotion judgment and the like are easily interfered by the environment and individual difference, so that the actual combat effect is seriously affected. Disclosure of Invention The application provides a method, a device, electronic equipment and a storage medium for assisting intelligent interrogation of a smuggling case, which can automatically complete extraction of a stroke record element, language correction, contradiction detection and relation map construction, effectively assist an interrogation person to find a logic loophole, inquire key details and generate a standard document, and greatly reduce the labor burden. Based on priori knowledge of the smuggling cases, performing wrongly written word detection and semantic error correction on the input written text to obtain an error corrected reference written text and an error position list, wherein the error position list is used for verifying the error corrected reference written text; Carrying out semantic analysis on the reference stroke text to extract key elements and generating structural information based on the key elements, wherein the structural information comprises at least one of time, place, personnel, articles, amount, case abstract and inquiry content abstract of the occurrence of a smuggling case; Carrying out matching analysis on the structured information and a preset interrogation strategy tag system, and outputting an information missing tag list; according to the information missing tag list, a corresponding inquiry propulsion strategy is called from a strategy template library, wherein the interrogation strategy tag system comprises three types of tags of subjective purpose, personnel status and smuggling ration; Carrying out logic consistency detection on the structured information based on a conflict rule base and a graph neural network mixed reasoning mechanism to obtain a conflict point list and corresponding inquiry suggestions in the structured information, wherein the conflict rule base comprises a smuggling time conflict rule, an article quantity conflict rule, an article price tax rate matching rule and a transportation path non-compliance rule; Fusing the inquiry propulsion strategy and the inquiry suggestion to generate an integrated inquiry guiding scheme; Based on the smuggling field ontology and the relation extraction model, carrying out relation recognition on key elements in the structured information, and constructing an entity relation map and a standardized case information table; Receiving an updated written text generated after the interrogation is conducted according to the interrogation guiding scheme, and re-executing the steps of wrongly written word detection, semantic error correction, semantic analysis and structured information generation on the updated written text to obtain updated structured information; and carrying out multidimensional weighted evaluation on the updated structured information, the standardized case information table, the entity relation map and the logical consistency detection result by adopting an analytic hierarchy process, and outputting a stroke quality grading grade and a targeted improvement suggestion. Through the scheme, the deep logic consistency detection of the interrogation record information can be realized through the mixed reasoning mechanism of the contradictory rule base and the graphic neural network, and the targeted inquiry suggestion is automatically generated. The function converts the contradiction discovery process of the traditional personal experience and the on-site judgment of the dependent interrogation personnel into an intelligent auxiliary flow which is systematic,