Intelligent interrogation assisting method and device for smuggling case, electronic equipment and storage medium

Abstract

The invention provides a smuggling case intelligent interrogation assistance method and device, electronic equipment and a storage medium, the method is applied to the field of natural language processing, and the method comprises the following steps: carrying out wrongly written character detection and semantic error correction on a record text to obtain an error-corrected reference record text; performing semantic analysis on the reference record text to extract key elements, and generating structured information; performing matching analysis on the structured information and the interrogation strategy label system, and outputting an information missing label list; calling an inquiry propelling strategy from a strategy template library according to the information missing label list; performing logic consistency detection on the structured information to obtain a contradictory point list and a questioning suggestion in the structured information; and constructing a standardized case information table. According to the method, record element extraction, language error correction, contradiction detection and relation graph construction can be automatically completed, interrogation personnel are effectively assisted in discovering logic vulnerabilities, asking key details and generating standard documents, and the labor burden is greatly relieved.

Description

Technical Field

[0001] This application relates to the field of natural language processing, and more specifically, to a method, apparatus, electronic device, and storage medium for intelligent interrogation assistance in smuggling cases. Background Technology

[0002] The transcript mainly records the statements of the respondent and is an important basis for case description and characterization. However, in the interrogation of smuggling cases, the current reliance on manual work to complete tasks such as transcript organization, information extraction, strategy formulation, and contradiction verification cannot automatically extract target information or perform structured analysis of the transcript content according to the target information. It can only assist the questioner in manually recording with fixed prompts. Due to the inability to automatically extract target information, the credibility of the transcript is low.

[0003] In related technologies, AI-assisted tools are used to assist the interrogation process. However, the accuracy and reliability of AI technology in interrogation scenarios are still insufficient. Speech recognition and emotion judgment are easily affected by environmental and individual differences, which seriously affects the actual combat effect. Summary of the Invention

[0004] This application provides a method, device, electronic device, and storage medium for intelligent interrogation assistance in smuggling cases. The method can automatically complete the extraction of transcript elements, language error correction, contradiction detection, and relationship graph construction, effectively assisting interrogators in discovering logical loopholes, pursuing key details, generating standardized documents, and significantly reducing the workload of human intervention.

[0005] Based on prior knowledge of smuggling cases, the input transcript text is subjected to typo detection and semantic correction to obtain the corrected reference transcript text and a list of error locations. This list of error locations is used to verify the corrected reference transcript text. The reference transcript text is semantically parsed to extract key elements, and structured information is generated based on these key elements; the structured information includes at least one of the following: the time, place, personnel, goods, amount, case summary, and summary of the interrogation content of the smuggling case. The structured information is matched and analyzed with the preset interrogation strategy labeling system to output a list of missing information labels; based on the list of missing information labels, the corresponding questioning advancement strategy is called from the strategy template library; the interrogation strategy labeling system includes three categories of labels: subjective intent, personnel status, and smuggling quantity; Based on the contradiction rule base and graph neural network hybrid reasoning mechanism, the logical consistency of the structured information is detected, and a list of contradiction points and corresponding follow-up questions are obtained. The contradiction rule base includes smuggling time conflict rules, item quantity conflict rules, item price and tax rate matching rules, and transportation route mismatch rules. By integrating the questioning approach and the follow-up questioning suggestions, an integrated interrogation guidance plan is generated; Based on the ontology and relationship extraction model in the field of smuggling, the key elements in this structured information are identified to construct an entity relationship graph and a standardized case information table. Receive the updated transcript text generated after the interrogation is conducted according to the interrogation guidance plan, and re-execute the steps of typo detection, semantic error correction, semantic parsing and structured information generation on the updated transcript text to obtain the updated structured information; Using the analytic hierarchy process (AHP), a multi-dimensional weighted evaluation is conducted on the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency test results, outputting a transcript quality score and targeted improvement suggestions.

[0006] The above-described solution enables deep logical consistency detection of interrogation transcripts through a hybrid reasoning mechanism combining a contradiction rule base and graph neural networks, automatically generating targeted follow-up questions. This function transforms the traditional contradiction discovery process, which relies on the interrogator's personal experience and on-the-spot judgment, into a systematic, automated, and continuously optimized intelligent auxiliary process. Specifically, it integrates the unique logical rules of smuggling cases and uses graph neural network-based associative reasoning to automatically compare statements made by the same person in multiple instances, or between different involved parties, regarding key elements such as time, location, goods, and amounts. It accurately identifies hidden logical loopholes, contradictions, or illogical points and, based on the identified contradiction type and context, intelligently matches and generates specific, actionable follow-up questions from the strategy base. This effectively prevents the omission of important contradictions due to information overload, fatigue, or lack of experience, greatly improving the targeting of each follow-up question and the control of the interrogation pace, thus accelerating the clarification of the core facts of the case.

[0007] In some possible implementations, the prior knowledge includes the corresponding legal provisions of the smuggling case and a pre-defined misspelling database in the smuggling field. Based on this prior knowledge of the smuggling case, the input transcript text is subjected to misspelling detection and semantic correction to obtain a corrected reference transcript text and a list of error locations, including: According to the corresponding legal provisions and the character pairing standards in the preset error word database, based on the context association neural network, characters whose frequency of occurrence in the transcript text is lower than the first preset threshold are identified as erroneous content, and the position of the erroneous content is identified as the error position list. According to the corresponding legal provisions and the preset error word database, the sequence-to-sequence model is used to correct the error content, resulting in a corrected reference transcript text.

[0008] In some possible implementations, the reference transcript text is semantically parsed to extract key elements, and structured information is generated based on these key elements, including: Based on a pre-defined dictionary in the field of smuggling, semantic analysis and key element extraction were performed on the reference transcript text to obtain the key elements of the smuggling case. The key elements include at least one of the following: the time, place, personnel, goods, and amount of money involved in the smuggling case. If the degree of matching between the key element and the standard element is greater than or equal to the second preset threshold, a case summary of the smuggling case is generated based on the key element. A query summary is generated based on a text processing algorithm and the preset dictionary. The number of words in the query summary is less than or equal to a third preset threshold. The key elements, the case summary, and the summary of the inquiry content are structured to obtain the structured information.

[0009] In some possible implementations, the structured information is matched and analyzed against a preset interrogation strategy labeling system to output a list of missing information labels, including: Based on the pre-trained language model and domain classifier, we analyze whether the structured information completely matches the interrogation strategy labeling system; If the structured information does not fully match the interrogation strategy labeling system, output a list of missing labels. Before invoking the appropriate query advancement strategy from the strategy template library, the following is also included: Retrieve the questioning angles, questioning methods, key follow-up questions, and example questions corresponding to the list of missing information tags from the strategy template library; Determine the sentence similarity between the example questions; If the sentence similarity is less than the fourth preset threshold, an inquiry advancement strategy is generated based on the questioning angle, the questioning method, the key follow-up questions, and the example question.

[0010] In some possible implementations, the structured information is logically consistent using a hybrid reasoning mechanism based on a contradiction rule base and graph neural networks to obtain a list of contradictions in the structured information and corresponding follow-up questions, including: The confidence conflict detection of this structured information is performed based on a contradiction rule base and a graph neural network hybrid reasoning mechanism to determine the confidence of each piece of information in the structured information. Information with a confidence level less than the fifth preset threshold in this structured information is identified as a list of contradictions; Based on the list of contradictions and the contextual semantics in the structured information, the follow-up questions for the list of contradictions are determined.

[0011] In some possible implementations, the smuggling domain ontology includes personnel roles, the attributes of the means of transport, and customs district codes. Based on the smuggling domain ontology and relation extraction model, key elements in the structured information are identified to construct an entity relationship graph and a standardized case information table, including: Based on the personnel role, the transportation vehicle attribute, the customs area code, and the RE-DEVO model, the relationships between key elements in the structured information are extracted; The relationship and the structured information are standardized to obtain a standardized case information table.

[0012] In some possible implementations, the analytic hierarchy process (AHP) is employed to perform a multi-dimensional weighted evaluation of the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency detection results, outputting a transcript quality score and targeted improvement suggestions, including: Obtain any missing evidence related to this smuggling case; The missing evidence, the structured information, and the standardized case information table were identified to reveal logical contradictions and procedural irregularities between the structured information and the missing evidence. The Analytic Hierarchy Process (AHP) was used to conduct a weighted comprehensive evaluation of the logical contradiction and the procedural irregularities to obtain the transcript quality rating. Improvement suggestions are generated based on the quality rating of the transcript and a pre-set suggestion library.

[0013] Secondly, a device for intelligent interrogation assistance in smuggling cases is provided, the device comprising: Based on prior knowledge of smuggling cases, the input transcript text is subjected to typo detection and semantic correction to obtain the corrected reference transcript text and a list of error locations. This list of error locations is used to verify the corrected reference transcript text. The reference transcript text is semantically parsed to extract key elements, and structured information is generated based on these key elements; the structured information includes at least one of the following: the time, place, personnel, goods, amount, case summary, and summary of the interrogation content of the smuggling case. The structured information is matched and analyzed with the preset interrogation strategy labeling system to output a list of missing information labels; based on the list of missing information labels, the corresponding questioning advancement strategy is called from the strategy template library; the interrogation strategy labeling system includes three categories of labels: subjective intent, personnel status, and smuggling quantity; Based on the contradiction rule base and graph neural network hybrid reasoning mechanism, the logical consistency of the structured information is detected, and a list of contradiction points and corresponding follow-up questions are obtained. The contradiction rule base includes smuggling time conflict rules, item quantity conflict rules, item price and tax rate matching rules, and transportation route mismatch rules. By integrating the questioning approach and the follow-up questioning suggestions, an integrated interrogation guidance plan is generated; Based on the ontology and relationship extraction model in the field of smuggling, the key elements in this structured information are identified to construct an entity relationship graph and a standardized case information table. Receive the updated transcript text generated after the interrogation is conducted according to the interrogation guidance plan, and re-execute the steps of typo detection, semantic error correction, semantic parsing and structured information generation on the updated transcript text to obtain the updated structured information; Using the analytic hierarchy process (AHP), a multi-dimensional weighted evaluation is conducted on the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency test results, outputting a transcript quality score and targeted improvement suggestions.

[0014] Thirdly, an electronic device is provided, including a memory and a processor. The memory is used to store executable program code, and the processor is used to call and run the executable program code from the memory, causing the electronic device to perform the methods described above for intelligent interrogation assistance in smuggling cases.

[0015] Fourthly, a computer program product is provided, comprising: computer program code, which, when run on a computer, causes the computer to execute the methods described above for intelligent interrogation assistance in smuggling cases.

[0016] Fifthly, a computer-readable storage medium is provided that stores computer program code, which, when executed on a computer, causes the computer to perform the methods described above for intelligent interrogation assistance in smuggling cases. Attached Figure Description

[0017] Figure 1 This is a schematic diagram illustrating the implementation environment of a method for intelligent interrogation assistance in smuggling cases provided in an embodiment of this application; Figure 2 This is a schematic flowchart illustrating a method for intelligent interrogation assistance in smuggling cases provided in an embodiment of this application; Figure 3 This is a schematic flowchart illustrating another intelligent interrogation assistance method for smuggling cases provided in the embodiments of this application; Figure 4This is a schematic diagram of the structure of an intelligent interrogation assistance device for smuggling cases provided in an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] The technical solutions in this application will be clearly and thoroughly described below with reference to the accompanying drawings. In the description of the embodiments of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in the text is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Furthermore, in the description of the embodiments of this application, "multiple" refers to two or more than two.

[0019] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0020] In the following description of the embodiments of this application, it is used as... Figure 1 Taking an example, the implementation environment of the embodiments of this application will be introduced.

[0021] For example, such as Figure 1 As shown, the implementation environment includes server 110 and client 120.

[0022] Server 110 is a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, and big data and artificial intelligence platforms. Server 110 includes a network communication unit, processor, and memory, etc. Specifically, server 110 is used to process the transcript text collected by client 120 for interrogation assistance.

[0023] Client 120 can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc. Client 120 is a physical device with information collection and basic image processing capabilities. Specifically, client 120 can collect text information or image information from transcribed text. Client 120 and server 110 are connected directly or indirectly via wired or wireless communication.

[0024] With the continuous evolution of new cybercrime methods, the crime situation is becoming increasingly complex. Achieving breakthroughs in interrogation within the golden timeframe of investigation has become a significant challenge for law enforcement agencies. Interrogation, as a core component of case investigation, is not only the primary means of ascertaining the facts and obtaining crucial clues, but it also comes with intense workload and complex investigative difficulties. However, modern criminals generally possess strong counter-investigation awareness and the ability to resist interrogation. Their legal knowledge has significantly improved, and they are very familiar with judicial procedures. This is especially true among repeat offenders, habitual offenders, and high-IQ criminals, who are adept at disrupting the interrogation process.

[0025] In the context of interrogation, the interrogation model has gradually shifted from the traditional confession-based "from confession to evidence" to an evidence-based "from evidence to confession" model centered on objective evidence. The method of concluding a case solely based on confessions is no longer adequate for modern judicial requirements. Current judicial practice emphasizes "respect for evidence, thorough investigation, and not blindly trusting confessions," requiring interrogation activities to be based on sufficient preliminary investigation and solid evidence.

[0026] Currently, there is still significant room for improvement in the professionalism and sophistication of interrogation work. As a specialized skill integrating knowledge from multiple disciplines such as criminal psychology, linguistics, and behavioral analysis, interrogation currently suffers from insufficient systematic training and high-level professional guidance. In practice, it relies heavily on the "apprenticeship" system of experience transmission and individual self-accumulation, lacking a unified, standardized, and systematic methodology, which hinders the overall improvement of interrogation efficiency. Among related technologies, artificial intelligence-assisted tools are used to assist the interrogation process; however, the accuracy and reliability of AI technology in interrogation scenarios are still insufficient. Voice recognition and emotion assessment are easily affected by environmental and individual differences, seriously impacting practical results.

[0027] To address at least one of the aforementioned technical problems, this application provides a method for intelligent interrogation assistance in smuggling cases. This method can automatically extract elements from transcripts, correct language errors, detect contradictions, and construct relationship graphs, effectively assisting interrogators in discovering logical loopholes, pursuing key details, generating standardized documents, and significantly reducing the workload of human intervention.

[0028] Figure 2 This is a schematic flowchart illustrating a method for intelligent interrogation assistance in smuggling cases provided in an embodiment of this application.

[0029] For example, such as Figure 2 As shown, taking the server as the executing entity as an example, this application describes a method for intelligent interrogation assistance in smuggling cases. The method 200 includes the following steps.

[0030] Step 201: Based on the prior knowledge of smuggling cases, perform typo detection and semantic error correction on the input transcript text to obtain the corrected reference transcript text and a list of error positions, where the list of error positions is used to verify the corrected reference transcript text.

[0031] Among them, the prior knowledge includes the corresponding legal provisions of smuggling cases and a preset misspelling library in the smuggling field. The transcript text is the detailed information of smuggling cases recorded by interrogators during the case interrogation process.

[0032] It should be understood that the interrogation process of smuggling cases involves a large amount of professional field knowledge, and the transcript text obtained from the interrogation is likely to contain typos. Therefore, it is necessary to perform typo detection and semantic error correction on the transcript text based on the prior knowledge in the smuggling field. The list of error positions is used to verify the corrected reference transcript text. In practical applications, during the process of the server performing typo detection and semantic error correction on the transcript text, error correction mistakes may occur. Therefore, to ensure the integrity and rigor of the transcript text, it is necessary to verify again whether the corrected text information is incorrect according to the error positions.

[0033] In a possible implementation manner, according to the character collocation standards in the corresponding legal provisions and the preset misspelling library, based on a context-related neural network, characters in the transcript text whose character pair occurrence frequency is lower than the first preset threshold are determined as error contents, and the positions of these error contents are determined as the list of error positions; according to the corresponding legal provisions and the preset misspelling library, a sequence-to-sequence model is used to correct the error contents to obtain the corrected reference transcript text.

[0034] Among them, the corresponding legal provisions are the legal provisions related to smuggling cases. The preset misspelling library is a special misspelling library in the smuggling field. For example, in the misspelling library, "海關" corresponds to "海关", "绕关" corresponds to "绕官", etc. The context-related neural network is a combined model of CNN and Bi-GRAM.

[0035] The first preset threshold is a threshold automatically determined by the server, and the embodiments of this application do not limit this.

[0036] In some embodiments, based on the combined model of CNN and Bi-GRAM, characters in the transcript text whose character pair occurrence frequency is lower than the first preset threshold are determined as error contents.

[0037] For example, Bi-GRAM calculates the probability of each word in the transcript text co-occurring with its previous and next words. If the probability of consecutive character pairs in the transcript text is low, then the character pair is determined as an error content.

[0038] In some embodiments, the error contents are subjected to standardized mapping according to the corresponding legal provisions and the preset misspelling library to obtain the corrected reference transcript text.

[0039] For example, if the incorrect content is "bonded warehouse", the corrected reference transcript text will be "bonded warehouse".

[0040] In this implementation, since human error is always possible when writing transcripts, the corrected reference transcript is corrected based on the corresponding legal provisions and a pre-set error database. This results in a higher accuracy rate of the corrected reference transcript than 0.92, which reduces the impact of erroneous transcripts on the trial of smuggling cases.

[0041] Step 202: Semantically analyze the reference transcript text to extract key elements, and generate structured information based on the key elements; wherein, the structured information includes at least one of the following: time, place, personnel, goods, amount, case summary, and summary of interrogation content of the smuggling case. The key elements include at least one of the following: the time, place, people, goods, and amount involved in the smuggling case.

[0042] It should be understood that in practical applications, the corrected reference transcript may contain a large amount of irrelevant information. Therefore, it is necessary to perform semantic analysis on the reference transcript to extract the key elements in the reference transcript.

[0043] In one possible implementation, based on a pre-defined dictionary in the field of smuggling, semantic analysis and key element extraction are performed on the reference transcript text to obtain key elements of the smuggling case. These key elements include at least one of the following: the time, place, personnel, goods, and amount involved in the smuggling case. If the degree of matching between the key elements and standard elements is greater than or equal to a second pre-defined threshold, a case summary of the smuggling case is generated based on the key elements. An interrogation content summary is generated based on a text processing algorithm and the pre-defined dictionary, and the number of words in the interrogation content summary is less than or equal to a third pre-defined threshold. The key elements, the case summary, and the interrogation content summary are then subjected to structured processing to obtain the structured information.

[0044] The pre-defined dictionary for smuggling cases is used to name key elements extracted from written records. For example, if a smuggling case includes information such as "Zhang San," "Li Si," and "Wang Wu," then based on the pre-defined dictionary, these individuals can be identified as personnel, and the key element of "personnel" can be extracted. Standard elements include the time, place, personnel, goods, and amount involved in the smuggling case. The case summary represents the currently necessary information for the smuggling case. The interrogation summary identifies four categories of information: the nature of the goods, the smuggling method, the regional characteristics, and the summary of statements. The pre-defined dictionary is a collection of frequently occurring, domain-specific words and phrases found in smuggling cases.

[0045] The second preset threshold is a threshold automatically determined by the server, and this embodiment of the application does not limit this. For example, the second preset threshold is 0.85. The third preset threshold is a threshold automatically determined by the server, and this embodiment of the application does not limit this. For example, the third preset threshold is 120 characters.

[0046] In some embodiments, the degree of matching between the extracted key features and the standard features is determined based on ROUGE-L (Recall-Oriented Understudy for GistingEvaluation - Longest Common Subsequence) to be greater than or equal to a second preset threshold.

[0047] It should be understood that in practical applications, smuggling personnel may not cooperate in entering written records. Therefore, the key elements in the records may not be complete, meaning the degree of matching between the key elements and the standard elements is relatively low. If the degree of matching between the key elements and the standard elements is low, the case summary generated based on the key elements will also be inaccurate. Therefore, it is necessary to determine whether the degree of matching between the key elements and the standard elements is greater than or equal to a second preset threshold. For example, if the key elements include the location, personnel, goods, and amount involved in the smuggling case, then the degree of matching between this key element and the standard elements is determined to be high. In this case, a case summary of the smuggling case is generated based on the key elements.

[0048] In some embodiments, a method combining TextRank with a domain dictionary is used to generate a query content summary.

[0049] TextRank is an unsupervised text extraction and summarization algorithm.

[0050] It should be understood that in practical applications, during the processing of transcript text, the server needs to be able to directly parse and calculate information, and therefore needs to generate structured information based on this key element.

[0051] In this implementation, key elements, case summaries, and interrogation summaries of smuggling cases are generated based on a pre-defined dictionary in the field of smuggling and reference transcripts. Since the pre-defined dictionary is a collection of frequently occurring, domain-specific words and phrases in smuggling cases, it is possible to obtain highly relevant and accurate case summaries and interrogation summaries based on this dictionary and reference transcripts, facilitating subsequent evaluation of the reference transcripts.

[0052] Step 203: Match and analyze the structured information with the preset interrogation strategy labeling system to output a list of missing information labels; based on the list of missing information labels, call the corresponding interrogation advancement strategy from the strategy template library; the interrogation strategy labeling system includes three categories of labels: subjective intent, personnel status, and smuggling quantity.

[0053] The interrogation strategy labeling system includes three categories of labels: subjective intent, personnel status, and smuggling quantity. The missing information label list stores information from the interrogation strategy labeling system that does not match the structured information. The strategy template library stores multiple interrogation advancement strategies, which are used to inquire about information from the interrogation strategy labeling system that does not match the structured information. For example, the structured information may not include the smuggling quantity from the interrogation strategy labeling system; that is, the information from the interrogation strategy labeling system that does not match the structured information is the smuggling quantity. The smuggling quantity includes the name, quantity, frequency, value, tax rate, and amount of tax evaded of the smuggled goods.

[0054] It should be understood that in practical applications, the interrogation strategy labeling system is the information dimension that case handlers need to clarify during the interrogation of smuggling cases. However, since the structured information may be incomplete, it is necessary to match and analyze the structured information with the preset interrogation strategy labeling system to determine the information missing in the structured information compared to the interrogation strategy labeling system, that is, to determine the list of missing labels for the output information.

[0055] In one possible implementation, the structured information is analyzed based on a pre-trained language model and a domain classifier to determine whether it fully matches the interrogation strategy labeling system; if the structured information does not fully match the interrogation strategy labeling system, a list of missing information labels is output.

[0056] The pre-trained language model is the BERT model.

[0057] In some embodiments, when the structured information completely matches the interrogation strategy labeling system, the structured information is logically consistent based on a contradiction rule base and a graph neural network hybrid reasoning mechanism to obtain a list of contradictions in the structured information and corresponding follow-up questions.

[0058] In this implementation, since the structured information may be incomplete, the structured information is matched and analyzed with the preset interrogation strategy label system to obtain a list of missing information labels. This allows the missing information in the structured information to be supplemented based on the list of missing information labels, thereby making the interrogation information in smuggling cases more complete.

[0059] Optionally, before executing the call to the corresponding query advancement strategy from the strategy template library, the following is also included: In one possible implementation, the questioning angle, questioning method, key follow-up questions, and example questions corresponding to the missing information label list are obtained from the strategy template library; the sentence similarity between the example questions is determined; and if the sentence similarity is less than a fourth preset threshold, an inquiry advancement strategy is generated based on the questioning angle, the questioning method, the key follow-up questions, and the example questions.

[0060] It should be understood that, when missing information is identified in the structured information, the corresponding questioning angles, questioning methods, key follow-up questions, and example questions are obtained from the strategy template based on the list of missing information labels to complete the missing information in the structured information. In practical applications, during the interrogation process, in order to reduce procedural complexity and avoid repetition, the sentence similarity between example questions needs to be controlled below a fourth preset threshold.

[0061] The fourth preset threshold is a threshold automatically determined by the server. This application embodiment does not limit this threshold. For example, the fourth preset threshold can be 0.7.

[0062] In some embodiments, if the sentence similarity is less than a fourth preset threshold, the questioning angle, questioning method, key follow-up questions, and example questions corresponding to the missing information label list are retrieved again from the strategy template library.

[0063] In this implementation, since there is missing information in the structured information, in order to complete the missing information in the structured information, the corresponding questioning angle, questioning method, key follow-up questions and example questions are obtained from the strategy template. That is, the questioning advancement strategy is generated based on the questioning angle, questioning method, key follow-up questions and example questions, so as to improve the structured information and make the interrogation process more accurate.

[0064] Step 204: Based on the contradiction rule base and graph neural network hybrid reasoning mechanism, the logical consistency of the structured information is checked to obtain a list of contradictions in the structured information and corresponding follow-up questions. The contradiction rule base includes smuggling time conflict rules, item quantity conflict rules, item price and tax rate matching rules, and transportation route mismatch rules.

[0065] In one possible implementation, the structured information is subjected to confidence conflict detection based on a contradiction rule base and a graph neural network hybrid reasoning mechanism to determine the confidence level of each piece of information in the structured information; information in the structured information with a confidence level less than a fifth preset threshold is identified as a list of contradiction points; and follow-up questions for the list of contradiction points are determined based on the list of contradiction points and the contextual semantics in the structured information.

[0066] In some embodiments, the graph neural layer based on the graph neural network hybrid reasoning mechanism determines the contradictions among the various pieces of information in the structured information.

[0067] For example, if a smuggler states in the first day's record that he was overseas at 8:00 AM on the 23rd and in China at 12:00 PM, but states in the second day's record that he was overseas at 3:00 PM on the 23rd, then there is a contradiction in the record. In other words, the confidence level of this structured information is less than the fifth preset threshold.

[0068] In some embodiments, the confidence level of the structured information is determined based on the degree of matching between the individual pieces of information in the structured information.

[0069] For example, structured information includes the time, location, personnel, goods, amount involved in the smuggling case, a case summary, and a summary of the interrogation. If the time, location, personnel, and goods of the smuggling case match perfectly, but the amount does not match, the confidence level of the structured information is determined to be at level one. If the time, location, and personnel of the smuggling case match perfectly, but the goods and amount do not match, the confidence level of the structured information is determined to be at level two, which is lower than level one. If the time and location of the smuggling case match perfectly, but the amount, personnel, and goods do not match, the confidence level of the structured information is determined to be at level three, which is lower than level two.

[0070] In this implementation, a confidence conflict detection is performed on the structured information based on a contradiction rule base and a graph neural network hybrid reasoning mechanism; information in the structured information with a confidence level less than a fifth preset threshold is identified as a list of contradiction points; and follow-up questions for the list of contradiction points are determined based on the list of contradiction points and the contextual semantics in the structured information. In other words, the contradiction point list in the structured information can be quickly identified based on the contradiction rule base and graph neural network hybrid reasoning mechanism, thereby quickly determining the logical relationship of information in the transcript text of smuggling cases, so as to quickly determine the overall details of smuggling cases and provide interrogators with a good information structure.

[0071] Step 205: Integrate the questioning advancement strategy with the follow-up questioning suggestions to generate an integrated interrogation guidance plan.

[0072] It should be understood that in practical applications, after determining the questioning approach and follow-up questioning suggestions, in order to improve the structured information, it is necessary to integrate the questioning approach and follow-up questioning suggestions to obtain an interrogation guidance plan that can quickly crack the details of smuggling cases.

[0073] Step 206: Based on the ontology and relationship extraction model in the smuggling field, identify the relationships of key elements in the structured information, and construct an entity relationship graph and a standardized case information table.

[0074] The smuggling ontology includes personnel roles, vehicle attributes, and customs district codes. A relation extraction model is used to extract the relationships between key elements in the structured information. This relation extraction model is a RE-DEVO model. An entity relationship graph is used to display the relationships between key elements in the structured information. The case information table is an information table that completes the structured information.

[0075] In one possible implementation, the relationships between key elements in the structured information are analyzed based on the personnel role, the transportation vehicle attribute, the customs district code, and the RE-DEVO model, and an entity relationship graph is constructed; the relationships and the structured information are then standardized to obtain a standardized case information table.

[0076] It should be understood that in smuggling cases, different individuals play different roles and perform different tasks. For example, person A may hire person B to hide goods. Therefore, there are relationships between the key elements in the structured information.

[0077] In some embodiments, the entity relationship graph conforms to Mermaid syntax. The standardized case information table includes information such as people, time, location, and items.

[0078] Personnel information is shown in Table 1 below.

[0079]

[0080] For example, as shown in Table 1 above, the standardized case information represents the names of individuals involved in smuggling cases, their roles in the cases, contact information, identity information, and associated persons. Furthermore, relationships exist between these individuals, allowing for the determination of each individual's detailed information within the smuggling case based on this personnel information table.

[0081] The time information is shown in Table 2 below.

[0082]

[0083] For example, as shown in Table 2 above, the standardized case information represents the events in which smuggling cases occurred, as well as the time when the events occurred. Table 2 also includes notes on the process of each event to facilitate understanding of the location information in the corresponding standardized case information table.

[0084] Location information is shown in Table 3 below.

[0085]

[0086] For example, as shown in Table 3 above, the standardized case information table shows the events that occurred at each location in a smuggling case, so that interrogators can determine the events that occurred throughout the smuggling case based on the location information.

[0087] The item information is shown in Table 4 below.

[0088]

[0089] For example, as shown in Table 4 above, the standardized case information table shows the items in smuggling cases. Since different countries have different standards for the quantity of smuggled items, after extracting the item information in a smuggling case, the criminal responsibility of the smugglers can be determined based on the item information, so as to obtain the interrogation results.

[0090] In this implementation, the relationships between key elements in the structured information are analyzed based on the personnel role, the transportation vehicle attribute, the customs district code, and the RE-DEVO model, and an entity relationship graph is constructed. The relationships and the structured information are then standardized to obtain a standardized case information table. That is, the standardized case information table includes detailed information in smuggling cases. The standardized information table is generated based on the structured information to facilitate observation and understanding, thereby simplifying the interrogation process for interrogators.

[0091] Step 207: Receive the updated transcript text generated after the interrogation is conducted according to the interrogation guidance scheme, and re-execute the steps of typo detection, semantic error correction, semantic parsing and structured information generation on the updated transcript text to obtain the updated structured information.

[0092] The updated structured information includes at least one of the following: the time, place, people, goods, amount involved in the smuggling case, case summary, and summary of the interrogation.

[0093] It should be understood that this interrogation guidance program is intended to supplement information not included in the structured information, in order to assist interrogators in thoroughly investigating the smuggling case.

[0094] It is understandable that the specific implementation of re-performing the typo detection, semantic correction, semantic parsing and structured information generation steps on the updated transcript text to obtain the updated structured information can be found in the relevant descriptions in steps 201-202 above, and will not be repeated here.

[0095] Step 208: Using the analytic hierarchy process (AHP), a multi-dimensional weighted evaluation is performed on the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency test results, and the transcript quality score and targeted improvement suggestions are output.

[0096] In one possible implementation, missing evidence corresponding to the smuggling case is obtained; the missing evidence, the structured information, and the standardized case information table are identified to obtain logical contradictions and procedural irregularities between the structured information and the missing evidence; the hierarchical analysis method is used to perform a weighted comprehensive evaluation of the logical contradictions and procedural irregularities to obtain a record quality score; and improvement suggestions are generated based on the record quality score and a preset suggestion library.

[0097] Among these, outstanding evidence includes customs declarations and shipping documents. In practice, the review of smuggling cases focuses not only on the written record but also on its authenticity; therefore, evidence is needed to prove the authenticity of the written record. Information regarding procedural irregularities refers to the procedural details handled by various personnel in the smuggling case.

[0098] It should be understood that in practical applications, if the procedures handled by personnel are not standardized, it may lead to invalid evidence in the case, affecting the review and trial of the case. Therefore, it is necessary to identify information on procedural irregularities.

[0099] In some embodiments, the Analytic Hierarchy Process (AHP) is used to perform a weighted comprehensive evaluation of the logical contradiction and the non-standard information in the procedure, resulting in a comment text.

[0100] The number of characters in the comment text is less than the sixth preset threshold, which is a threshold automatically determined by the server. This application embodiment does not limit this threshold. For example, the sixth preset threshold is 200 characters.

[0101] In some embodiments, the evaluation level is divided into three levels: excellent, satisfactory, and requires supplementation. If the record quality rating is "requires supplementation," then it is necessary to determine the information that needs to be supplemented and obtain improvement suggestions corresponding to the supplementary information from a preset suggestion library.

[0102] In this implementation method, logical contradictions and procedural irregularities in smuggling cases are weighted and comprehensively evaluated to obtain a record quality rating. Based on this record quality rating and a preset suggestion library, improvement suggestions are generated. That is, different comment texts and improvement suggestions are determined according to different record quality ratings, so that smuggling cases can be re-examined according to the improvement suggestions, thereby improving the quality and efficiency of smuggling case review.

[0103] This application provides a method for intelligent interrogation assistance in smuggling cases. This method utilizes a hybrid reasoning mechanism combining a contradiction rule base and a graph neural network to achieve deep logical consistency detection of interrogation transcripts and automatically generate targeted follow-up questions. This function transforms the traditional contradiction discovery process, which relies on the interrogator's personal experience and on-the-spot judgment, into a systematic, automated, and continuously optimized intelligent assistance process. Specifically, it integrates the unique logical rules of smuggling cases and uses graph neural network-based associative reasoning to automatically compare statements made by the same person in multiple statements or between different involved parties regarding key elements such as time, location, goods, and amount. It accurately identifies hidden logical loopholes, contradictions, or illogical points and, based on the identified contradiction type and context, intelligently matches and generates specific, actionable follow-up questions from a strategy base. This effectively prevents the omission of important contradictions due to information overload, fatigue, or lack of experience, greatly improving the targeting of each follow-up question and the control of the interrogation pace, thus accelerating the clarification of the core facts of the case.

[0104] Figure 3 This is a schematic flowchart illustrating a method for intelligent interrogation assistance in smuggling cases provided in an embodiment of this application.

[0105] For example, such as Figure 3 As shown, taking the server as the executing entity as an example, this application describes a method for intelligent interrogation assistance in smuggling cases. The method 300 includes the following steps.

[0106] Step 301: Obtain the input transcript text.

[0107] In some embodiments, the input transcript text is obtained based on the client.

[0108] Step 302: Based on legal provisions, a pre-set error word library in the field of smuggling, a strategy template library, a contradiction rule library, and the transcript text, obtain a list of contradictions and corresponding follow-up questions in the structured information.

[0109] It is understood that the specific implementation of step 302 can be found in the relevant descriptions of steps 201-204 above.

[0110] Step 303: Perform multi-source information fusion.

[0111] It is understood that the specific implementation of step 303 can be found in the relevant description of step 205 above.

[0112] Step 304: Construct an entity relationship graph.

[0113] It is understood that the specific implementation of step 304 can be found in the relevant description of step 206 above.

[0114] Step 305: Output the evaluation results.

[0115] It is understood that the specific implementation of step 305 can be found in the relevant descriptions of steps 207-208 above.

[0116] Figure 4 This is a schematic diagram of the structure of an intelligent interrogation assistance device for smuggling cases provided in an embodiment of this application.

[0117] For example, the device 400 includes: The error correction module 401 is used to perform typo detection and semantic error correction on the input transcript text based on prior knowledge of smuggling cases, and obtain the corrected reference transcript text and the error location list. The error location list is used to verify the corrected reference transcript text. The extraction and generation module 402 is used to perform semantic parsing on the reference transcript text to extract key elements and generate structured information based on the key elements; wherein, the structured information includes at least one of the following: time, place, personnel, goods, amount, case summary, and summary of interrogation content of the smuggling case. The analysis and invocation module 403 is used to match and analyze the structured information with the preset interrogation strategy label system and output a list of missing information labels; based on the list of missing information labels, the corresponding interrogation advancement strategy is invoked from the strategy template library; the interrogation strategy label system includes three categories of labels: subjective intent, personnel status, and smuggling quantity; The logic detection module 404 is used to perform logical consistency detection on the structured information based on the contradiction rule base and graph neural network hybrid reasoning mechanism, and obtain a list of contradiction points in the structured information and corresponding follow-up questions. The contradiction rule base includes smuggling time conflict rules, item quantity conflict rules, item price tax rate matching rules, and transportation route mismatch rules. The fusion and generation module 405 is used to fuse the questioning advancement strategy and the follow-up questioning suggestions to generate an integrated interrogation guidance plan; The identification and construction module 406 is used to identify the relationships of key elements in the structured information based on the ontology and relationship extraction model in the smuggling field, and to construct an entity relationship graph and a standardized case information table. The receiving and generating module 407 is used to receive the updated transcript text generated after the interrogation is conducted according to the interrogation guidance scheme, and to re-execute the steps of typo detection, semantic error correction, semantic parsing and structured information generation on the updated transcript text to obtain the updated structured information. The evaluation module 408 is used to perform a multi-dimensional weighted evaluation of the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency detection results using the analytic hierarchy process, and outputs the transcript quality score and targeted improvement suggestions.

[0118] In one possible implementation, the device 400 includes: The error correction module 401 is specifically used to determine characters whose frequency of occurrence of character pairs in the transcript text is lower than a first preset threshold as erroneous content, based on the corresponding legal provisions and the character pairing standards in the preset error word library, and to determine the position of the erroneous content as an error position list, according to the context association neural network. The error correction module 401 is specifically used to correct the error content according to the corresponding legal provisions and the preset error word library, using a sequence-to-sequence model, to obtain the corrected reference transcript text.

[0119] In one possible implementation, the device 400 includes: The extraction and generation module 402 is specifically used to perform semantic analysis and key element extraction on the reference transcript text based on a preset dictionary in the field of smuggling, to obtain the key elements of the smuggling case. The key elements include at least one of the time, place, personnel, goods, and amount of money involved in the smuggling case. The extraction and generation module 402 is specifically used to generate a case summary of the smuggling case based on the key element when the degree of matching between the key element and the standard element is greater than or equal to a second preset threshold. The extraction and generation module 402 is specifically used to generate a query content summary based on the text processing algorithm and the preset dictionary, wherein the number of words in the query content summary is less than or equal to a third preset threshold. The extraction and generation module 402 is specifically used to perform structured processing on the key elements, the case summary, and the inquiry content summary to obtain the structured information.

[0120] In one possible implementation, the device 400 includes: The analysis and invocation module 403 is specifically used to analyze whether the structured information is completely matched with the interrogation strategy labeling system based on the pre-trained language model and the domain classifier. The analysis and invocation module 403 is specifically used to output a list of missing information tags when the structured information does not fully match the interrogation strategy tag system. In one possible implementation, the device 400 includes: The analysis and invocation module 403 is also used to obtain from the strategy template library the questioning angle, questioning method, key follow-up questions and example questions corresponding to the list of missing information tags; The analysis and invocation module 403 is also used to determine the sentence similarity between the example questions; The analysis and invocation module 403 is also used to generate an inquiry advancement strategy based on the questioning angle, the questioning method, the key follow-up questions, and the example questions when the sentence similarity is less than the fourth preset threshold.

[0121] In one possible implementation, the device 400 includes: The logic detection module 404 is specifically used to perform confidence conflict detection on the structured information based on the contradiction rule base and graph neural network hybrid reasoning mechanism, so as to determine the confidence of each piece of information in the structured information; The logic detection module 404 is specifically used to identify information in the structured information with a confidence level less than the fifth preset threshold as a list of contradiction points; The logic detection module 404 is specifically used to determine the follow-up questions for the list of contradictions based on the list of contradictions and the contextual semantics in the structured information.

[0122] In one possible implementation, the device 400 includes: The identification and construction module 406 is specifically used to extract the relationship between key elements in the structured information based on the personnel role, the transportation vehicle attribute, the customs area code, and the RE-DEVO model, and to construct an entity relationship graph. The identification and construction module 406 is specifically used to standardize the relationship and the structured information to obtain a standardized case information table.

[0123] In one possible implementation, the device 400 includes: Evaluation module 408 is specifically used to obtain any missing evidence corresponding to the smuggling case; The evaluation module 408 is specifically used to identify the omitted evidence, the structured information, and the standardized case information table, and to obtain the logical contradictions and procedural irregularities between the structured information and the omitted evidence. The evaluation module 408 is specifically used to perform a weighted comprehensive evaluation of the logical contradiction and the non-standard information of the procedure using the analytic hierarchy process, and to obtain the quality score of the transcript. The evaluation module 408 is specifically used to generate improvement suggestions based on the quality rating of the transcript and the preset suggestion library.

[0124] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0125] For example, such as Figure 5As shown, the electronic device 500 includes a memory 501 and a processor 502. The memory 501 stores executable program code 503, and the processor 502 is used to call and execute the executable program code 503 to perform a method for intelligent interrogation assistance in smuggling cases.

[0126] Furthermore, embodiments of this application also protect an apparatus that may include a memory and a processor, wherein the memory stores executable program code, and the processor is used to call and execute the executable program code to perform a method for intelligent interrogation assistance in smuggling cases provided in embodiments of this application.

[0127] This embodiment can divide the device into functional modules based on the above method example. For example, each module can correspond to a separate function, or two or more functions can be integrated into one processing module. The integrated module can be implemented in hardware. It should be noted that the module division in this embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0128] It should be understood that the device provided in this embodiment is used to execute the above-described method for intelligent interrogation assistance in smuggling cases, and therefore can achieve the same effect as the above-described implementation method.

[0129] When using integrated units, the device may include a processing module and a storage module. When applied to an electronic device, the processing module can be used to control and manage the operation of the electronic device. The storage module can be used to support the execution of relevant program code by the electronic device.

[0130] The processing module may be a processor or a controller, which can implement or execute various exemplary logic blocks, modules, and circuits shown in conjunction with the disclosure of this application. The processor may also be a combination of functions that implement computing capabilities, such as a combination of one or more microprocessors, a combination of digital signal processing (DSP) and microprocessors, etc., and the storage module may be a memory.

[0131] In addition, the device provided in the embodiments of this application may specifically be a chip, component or module. The chip may include a connected processor and a memory. The memory is used to store instructions. When the processor calls and executes the instructions, the chip can execute the intelligent interrogation assistance method for smuggling cases provided in the above embodiments.

[0132] This embodiment also provides a computer-readable storage medium storing computer program code. When the computer program code is run on a computer, the computer executes the above-described related method steps to implement the intelligent interrogation assistance method for smuggling cases provided in the above embodiment.

[0133] This embodiment also provides a computer program product that, when run on a computer, causes the computer to perform the aforementioned related steps to realize the intelligent interrogation assistance method for smuggling cases provided in the above embodiment.

[0134] In this embodiment, the device, computer-readable storage medium, computer program product, or chip are all used to execute the corresponding methods provided above. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods provided above, and will not be repeated here.

[0135] Through the above description of the embodiments, those skilled in the art will understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0136] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0137] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for intelligent interrogation assistance in smuggling cases, characterized in that, The method includes: Based on prior knowledge of smuggling cases, the input transcript text is subjected to typo detection and semantic correction to obtain the corrected reference transcript text and a list of error locations. The list of error locations is used to verify the corrected reference transcript text. The reference transcript text is semantically parsed to extract key elements, and structured information is generated based on the key elements; wherein, the structured information includes at least one of the following: time, place, personnel, goods, amount, case summary, and summary of interrogation content of the smuggling case; The structured information is matched and analyzed with a preset interrogation strategy labeling system to output a list of missing information labels; based on the list of missing information labels, the corresponding interrogation advancement strategy is called from the strategy template library; the interrogation strategy labeling system includes three categories of labels: subjective intent, personnel status, and smuggling quantity; Based on the contradiction rule base and graph neural network hybrid reasoning mechanism, the structured information is logically consistent to obtain a list of contradictions in the structured information and corresponding follow-up questions. The contradiction rule base includes smuggling time conflict rules, item quantity conflict rules, item price and tax rate matching rules, and transportation route mismatch rules. By integrating the aforementioned questioning advancement strategy and the aforementioned follow-up questioning suggestions, an integrated interrogation guidance plan is generated; Based on the ontology and relationship extraction model in the smuggling field, the key elements in the structured information are identified to construct an entity relationship graph and a standardized case information table. Receive the updated transcript text generated after the interrogation is conducted according to the interrogation guidance scheme, and re-execute the steps of misspelling detection, semantic error correction, semantic parsing and structured information generation on the updated transcript text to obtain the updated structured information; The analytic hierarchy process (AHP) is used to perform a multi-dimensional weighted evaluation of the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency detection results, and outputs a transcript quality score and targeted improvement suggestions.

2. The method according to claim 1, characterized in that, The prior knowledge includes the corresponding legal provisions of the smuggling case and a preset misspelling database in the smuggling field. Based on the prior knowledge of the smuggling case, the input transcript text is subjected to misspelling detection and semantic correction to obtain the corrected reference transcript text and a list of error locations, including: According to the corresponding legal provisions and the character pairing standards in the preset error word library, based on the context association neural network, characters whose frequency of occurrence in the transcript text is lower than the first preset threshold are identified as erroneous content, and the positions of the erroneous content are identified as an error position list. According to the corresponding legal provisions and the preset error word library, the sequence-to-sequence model is used to correct the error content, and the corrected reference transcript text is obtained.

3. The method according to claim 1, characterized in that, The step of semantically parsing the reference transcript text to extract key elements and generating structured information based on the key elements includes: Based on a pre-defined dictionary in the field of smuggling, semantic analysis and key element extraction are performed on the reference transcript text to obtain the key elements of the smuggling case. The key elements include at least one of the time, place, personnel, goods, and amount of money involved in the smuggling case. If the degree of matching between the key elements and the standard elements is greater than or equal to a second preset threshold, a case summary of the smuggling case is generated based on the key elements. A query content summary is generated based on a text processing algorithm and the preset dictionary, wherein the number of characters in the query content summary is less than or equal to a third preset threshold. The key elements, the case summary, and the summary of the interrogation content are structured to obtain the structured information.

4. The method according to claim 1, characterized in that, The step involves matching and analyzing the structured information against a preset interrogation strategy tagging system to output a list of missing tags, including: The structured information is analyzed based on a pre-trained language model and a domain classifier to determine whether it fully matches the interrogation strategy labeling system. If the structured information does not fully match the interrogation strategy labeling system, output a list of missing labels. Before invoking the appropriate query advancement strategy from the strategy template library, the following is also included: Retrieve from the strategy template library the questioning angles, questioning methods, key follow-up questions, and example questions corresponding to the list of missing information tags; Determine the sentence similarity between the example questions; If the sentence similarity is less than a fourth preset threshold, an inquiry advancement strategy is generated based on the questioning angle, the questioning method, the key follow-up questions, and the example questions.

5. The method according to claim 1, characterized in that, The logical consistency detection of the structured information based on the hybrid reasoning mechanism of contradiction rule base and graph neural network is performed to obtain a list of contradictions in the structured information and corresponding follow-up questions, including: The confidence conflict detection of the structured information is performed based on a contradiction rule base and a graph neural network hybrid reasoning mechanism to determine the confidence of each piece of information in the structured information; Information with a confidence level less than a fifth preset threshold in the structured information is identified as a list of contradictions; Based on the list of contradictions and the contextual semantics in the structured information, the follow-up questions for the list of contradictions are determined.

6. The method according to claim 1, characterized in that, The smuggling domain ontology includes personnel roles, vehicle attributes, and customs district codes. Based on the smuggling domain ontology and relationship extraction model, key elements in the structured information are identified to construct an entity relationship graph and a standardized case information table, including: Based on the personnel roles, the transportation vehicle attributes, the customs area codes, and the RE-DEVO model, the relationships between key elements in the structured information are extracted and an entity relationship graph is constructed. The aforementioned relationships and structured information are standardized to obtain a standardized case information table.

7. The method according to claim 1, characterized in that, The analytic hierarchy process (AHP) is used to perform a multi-dimensional weighted evaluation of the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency detection results, outputting a transcript quality score and targeted improvement suggestions, including: Obtain any missing evidence corresponding to the smuggling case; The omitted evidence, the structured information, and the standardized case information table are identified to obtain logical contradictions and procedural irregularities between the structured information and the omitted evidence. The hierarchical analysis method is used to perform a weighted comprehensive evaluation of the logical contradictions and procedural irregularities to obtain the transcript quality rating. Improvement suggestions are generated based on the recorded quality rating and the preset suggestion library.

8. A device for intelligent interrogation assistance in smuggling cases, characterized in that, The device includes: The error correction module is used to detect typos and perform semantic correction on the input transcript text based on prior knowledge of smuggling cases, so as to obtain the corrected reference transcript text and the error location list. The error location list is used to verify the corrected reference transcript text. The extraction and generation module is used to perform semantic parsing on the reference transcript text to extract key elements, and generate structured information based on the key elements; wherein, the structured information includes at least one of the following: time, place, personnel, goods, amount, case summary, and summary of interrogation content of the smuggling case; The analysis and invocation module is used to match and analyze the structured information with a preset interrogation strategy labeling system, and output a list of missing information labels; based on the list of missing information labels, it invokes the corresponding interrogation advancement strategy from the strategy template library; the interrogation strategy labeling system includes three categories of labels: subjective intent, personnel status, and smuggling quantity; The logic detection module is used to perform logical consistency detection on the structured information based on the contradiction rule base and graph neural network hybrid reasoning mechanism, and obtain a list of contradiction points in the structured information and corresponding follow-up questions. The contradiction rule base includes smuggling time conflict rules, item quantity conflict rules, item price and tax rate matching rules, and transportation route mismatch rules. The fusion and generation module is used to fuse the questioning advancement strategy and the follow-up questioning suggestions to generate an integrated interrogation guidance scheme; The identification and construction module is used to identify the relationships of key elements in the structured information based on the ontology and relationship extraction model in the smuggling field, and to construct an entity relationship graph and a standardized case information table. The receiving and generating module is used to receive the updated transcript text generated after the interrogation is conducted according to the interrogation guidance scheme, and to re-execute the steps of misspelling detection, semantic error correction, semantic parsing and structured information generation on the updated transcript text to obtain the updated structured information. The evaluation module is used to perform a multi-dimensional weighted evaluation of the updated structured information, the standardized case information table, the entity relationship graph, and the logical consistency detection results using the analytic hierarchy process (AHP), and outputs a transcript quality score and targeted improvement suggestions.

9. An electronic device, characterized in that, The electronic device includes: Memory, used to store executable program code; A processor for calling and running the executable program code from the memory, causing the electronic device to perform the method as described in 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 that, when executed, implements the method as described in any one of claims 1 to 7.

Images