Large model-based clue data analysis method and system

Abstract

The invention relates to the technical field of clue data analysis, in particular to a clue data analysis method and system based on a large model. The method comprises the following steps: obtaining the oral information of a viewer, objective clue information and basic information of a case, carrying out effective screening on the oral information of the viewer based on the objective clue information and the basic information of the case, and determining available oral information; obtaining suspect record information, performing authenticity screening on the suspect record information based on the available dictation information, the objective clue information and the case basic information, and determining record pointing information; determining a reasoning node information set and preliminary suspect information according to the record pointing information; and determining a case reasoning report according to the preliminary suspect information and the reasoning node information set. Effective information is quickly screened, redundant data is reduced, the case detection efficiency is improved, the generated reasoning report can provide powerful support for subsequent processing of the case, and police are helped to make more accurate decisions.

Description

Technical Field

[0001] The present application relates to the technical field of clue data analysis, and in particular to a clue data analysis method and system based on a large model. Background Art

[0002] With the rapid development of digital technology, big data and machine learning algorithms have gradually become important tools for processing massive multimodal clue data. Big data technology can efficiently store, manage and process this data. It provides strong technical support for processing massive data. The application of big data technology enables enterprises, governments and other institutions to more effectively explore and utilize data value.

[0003] However, existing case reasoning technology is difficult to cope with complex and changing data environments, and is unable to effectively and clearly sort out the existing massive clues and obtain some reasoning information that helps improve the efficiency of case reasoning. It is difficult to sort out the case reasoning process and related suspects. Summary of the invention

[0004] The present application provides a clue data analysis method and system based on a large model to solve the above problems.

[0005] In a first aspect, the present application provides a clue data analysis method based on a large model, the method comprising: obtaining oral information from witnesses, objective clue information and basic case information, and screening the validity of the oral information from witnesses based on the objective clue information and the basic case information to determine the available oral information; obtaining the suspect's transcript information, and screening the authenticity of the suspect's transcript information based on the available oral information, the objective clue information and the basic case information to determine the transcript pointing information; determining the reasoning node information set and preliminary suspect information based on the transcript pointing information; and determining the case reasoning report based on the preliminary suspect information and the reasoning node information set.

[0006] Through the above technical solution, the available oral information can be analyzed and screened to eliminate the impact of useless oral information on the case and provide a basis for verifying the authenticity of the subsequent suspect's written information. By comparing and screening out the written information that is inconsistent with the available oral information and objective clues, and by comparing the inconsistencies in the information, it is further clarified which written information can be used as directional evidence, which helps to determine the degree of suspicion of the suspect. The clues related to the suspect are intuitively displayed in the form of nodes, making the case analysis more organized. By having each node point to a different preliminary suspect, it helps the police sort out the final suspect.

[0007] Optionally, the validity screening of the eyewitness oral information based on the objective clue information and the basic information of the case is performed to determine the available oral information, including: screening the eyewitness oral information based on the objective clue information and the basic information of the case, and determining the information in the eyewitness oral information that can match the objective clue information and the basic information of the case as preliminary available clues; determining the information in the eyewitness oral information that is inconsistent with the objective clue information and the basic information of the case as unavailable clues; determining pending clue information based on the unavailable clues and the preliminary available clues; and determining the available oral information based on the preliminary available clues and the pending clue information.

[0008] Through the above technical solution, by determining the unusable clues, we can reduce the time and resources invested in invalid information, concentrate our energy and resources on those clues that may be useful, and thus optimize the entire investigation process. By screening the oral information of witnesses, we can effectively identify the effective clues that match the objective clues and the basic information of the case from the oral information. This helps to reduce the interference of invalid information on the case investigation and improve the efficiency of the investigation. Through the consistency of oral information with objective clues and basic information, the reliability of the case evidence chain is enhanced, ensuring that the screened information has a higher credibility.

[0009] Optionally, determining the available oral information based on the preliminary available clues and the pending clue information includes: determining objective image clue information and objective text clue information based on the objective clue information; extracting pending clue information features based on the pending clue information to determine a pending clue feature vector set; determining an image feature vector set based on the objective image clue information; determining a text feature vector set based on the objective text clue information; determining the pending clue matching degree of each pending clue based on the pending clue feature vector set, in combination with the image feature vector set and the text feature vector set; determining pending available clues based on the pending clue information and the matching degrees of several pending clues; determining the available oral information based on the pending available clues and the preliminary available clues.

[0010] Through the above technical solution, by classifying objective image clue information and objective text clue information, corresponding processing technology can be used according to different types of clues to maximize the efficiency of information extraction. Through accurate feature vector extraction and matching analysis, it can help identify more accurate available clues, reduce the risk of misjudgment, and improve the scientific nature of case analysis. By screening out the available information in the pending information, this matching analysis can effectively screen out useful clues, reduce the interference of irrelevant information, and improve the accuracy and efficiency of reasoning.

[0011] Optionally, determining the pending available clues according to the pending clue information and the pending clue matching degree set includes: determining a plurality of weighted comprehensive feature vectors according to each image feature vector of the image feature vector set and each text feature vector of the text feature vector set, specifically the following formula:

[0012] F c =α·F i +β·F t ;

[0013] Among them, F c is the weighted comprehensive feature vector, F i is the current image feature vector, F t is the current text feature vector, α is the preset image influence weight, and β is the preset text influence weight; according to the weighted comprehensive feature vectors and each pending clue feature vector of the pending clue feature vector set, the matching degree of the pending clues is determined, which is specifically the following formula:

[0014]

[0015] Among them, C is the matching degree of the pending clues at present, F d is the current pending clue feature vector of the pending clue feature vector set, F c is the current weighted comprehensive feature vector; according to the pending clue matching degree and the preset high matching degree threshold, if the current pending clue matching degree exceeds the preset high matching degree threshold, the current pending clue under the pending clue information is determined as a pending highly credible clue; if the current pending clue matching degree is lower than the preset high matching degree threshold but higher than the preset low matching degree threshold, the current pending clue under the pending clue information is determined as a pending fuzzy clue; according to the pending highly credible clue and the pending fuzzy clue, the pending available clue is determined.

[0016] Through the above technical solution, by calculating the matching degree between the pending clue vector and the weighted comprehensive feature vector, the correlation between these clues can be objectively and scientifically evaluated. The value of the matching degree of the pending clues can effectively reflect the similarity between the clues and provide a quantitative basis for subsequent analysis. By standardizing the matching degree of the pending clues, the results are more interpretable and convenient for setting the matching degree threshold. By setting low matching degree and high matching degree thresholds, clues that are significantly related to the case and those vague clues that may be helpful for case reasoning but are uncertain can be effectively screened out. This information screening can improve the efficiency of case handling.

[0017] Optionally, based on the available oral information, the objective clue information and the basic information of the case, the suspect's written record information is screened for authenticity to determine the written record pointing information, including: determining the suspect's behavior information within the preset timeline based on the suspect's written record information and the preset timeline; determining a number of discrete time window information based on the pending highly credible clues; determining a number of continuous time window information based on the number of discrete time window information; determining a number of witnessed suspect behavior information in the current continuous time window based on the number of continuous time window information; analyzing the witnessed suspect behavior information and the suspect's behavior information in the current continuous time window to determine witnessed behavior difference information; analyzing the pending fuzzy clues to determine partial process reconstruction information; analyzing a number of partial process reconstruction information and the behavior information of a number of suspects to determine objective behavior difference information; determining the written record lie information based on the witnessed behavior difference information and the objective behavior difference information; determining the written record pointing information based on the number of determined written record lie information.

[0018] Through the above technical solution, by analyzing the suspect's transcript information and the preset timeline, the suspect's behavior information within a specific time period can be extracted. This provides basic data for subsequent comparison and analysis. Based on the pending fuzzy clues and pending highly credible clues of the witnesses, the suspect's transcript is analyzed to determine whether the suspect is lying. By analyzing the discrete time window, the suspect's behavior information under the continuous time window can be obtained, which provides a basis for subsequent comparison, and can provide dynamic process analysis and comparison under the continuous time window, which can make the comparison results more convincing and reliable.

[0019] Optionally, the analyzing the pending fuzzy clues to determine the partial process reconstruction information includes: analyzing the pending fuzzy clues to determine a number of objective discrete information; determining a number of discrete information feature vectors based on the number of objective discrete information; and determining the discrete information correlation degree based on the number of discrete information feature vectors, which is specifically the following formula:

[0020] Among them, S(Xi,Xj) is the discrete information correlation between the i-th discrete information feature vector and the j-th discrete information feature vector, Xi is the i-th discrete information feature vector, and Xj is the j-th discrete information feature vector; according to the plurality of objective discrete information, the plurality of discrete information correlations and the preset correlation threshold, a plurality of discrete information correlation chains are determined; according to the plurality of discrete information correlation chains, the partial process reconstruction information is determined.

[0021] Through the above technical solution, the correlation between feature vectors is calculated by designing a formula, and combined with the preset correlation threshold, highly correlated information is identified, thereby screening out useful information and ignoring redundant or irrelevant information, thereby improving the effectiveness of case reasoning. By calculating and comparing the correlation between each objective discrete information, a number of logical chains with multiple interrelated objective discrete information are formed to help construct the context and context of the case, provide more structured data support for subsequent analysis, and provide a basis for the subsequent police to judge the suspect's motives to improve data analysis efficiency, thereby increasing the speed of reasoning. By restoring part of the process, more intuitive clues can be provided for subsequent analysis. Extracting effective clues from fuzzy clues can make the processing logic more rigorous and logically reasonable.

[0022] Optionally, determining the inference node information set and preliminary suspect information based on the transcript pointing information includes: determining the current suspect of the transcript pointing information based on the transcript pointing information; determining the witnessed behavior difference information and the objective behavior difference information as the pointing information nodes of the current suspect; determining the inference node information set based on several of the pointing information nodes; and determining several preliminary suspects based on several of the current suspects.

[0023] Through the above technical solution, each transcript pointing information is analyzed one by one and organized into pointing information nodes, which can effectively associate clues with specific suspects, reduce ambiguity, and ensure that each clue can point to a clear target. Converting the transcript pointing information into reasoning nodes to form a structured information system helps to clarify the context of the case. By nodeizing the information, it can ensure that each clue and conclusion has a basis, which enhances the credibility and effectiveness of the investigation results. It provides a content basis for the generation of subsequent case reasoning reports. Identifying preliminary suspects can screen out suspects who are really suspected of committing crimes, narrow the scope of suspects, and improve the efficiency of subsequent police handling.

[0024] Optionally, determining the case reasoning report based on the preliminary suspect information and the reasoning node information set includes: analyzing the witnessed behavior difference information and the objective behavior difference information to determine the amount of difference information and the duration of the difference information; analyzing the amount of difference information and the duration of the difference information to determine the difference weight of each difference information and the suspicion value of the preliminary suspect; determining the case reasoning report based on the difference weight, the suspicion value, the reasoning node information set and the preliminary suspect.

[0025] Through the above technical solution, a quantifiable suspicion value is formed by analyzing the difference information and duration, and a systematic method is used to evaluate suspects instead of relying solely on subjective judgment. This quantitative evaluation can increase the objectivity and reliability of the reasoning results. By generating a case reasoning report for police reference, the readability and comprehensibility of information are enhanced. Automated data processing and analysis reduces the time of manual analysis and improves the efficiency of case reasoning. The display of the suspicion value can provide the police with a preliminary reference for the suspect and provide a reference for the final determination of the suspect.

[0026] Optionally, the analyzing the amount of the difference information and the duration of the difference information to determine the difference weight of each difference information and the suspicion value of the preliminary suspect is specifically the following formula:

[0027]

[0028] Among them, S m is the suspicion value of the mth preliminary suspect, α is the preset time influence weight, k is the preset difference weight influence coefficient, n is the amount of difference information of the mth preliminary suspect, T mi is the duration of the difference information of the i-th difference information of the m-th preliminary suspect, is the difference weight of each difference information of the mth preliminary suspect.

[0029] Through the above technical solution, the various difference information of each suspect can be quantified by designing a formula. This quantification helps to objectively evaluate the suspiciousness of each suspect. The formula can effectively capture the nonlinear characteristics of the effect of duration on difference information. This means that difference information with a longer duration will contribute more significantly to the suspicion value, thereby enhancing the sensitivity of the model. By traversing each difference information and performing a linear weighted summation, the comprehensive suspicion value of each preliminary suspect can be finally obtained. This aggregation method can integrate the contribution of various types of difference information, thereby forming a quantitative indicator for the comprehensive evaluation of suspects, making the results more convincing.

[0030] In a second aspect, the present application provides a clue data analysis system based on a large model, the system comprising:

[0031] An available information determination module is used to obtain oral information from witnesses, objective clue information and basic case information, and based on the objective clue information and the basic case information, perform validity screening on the oral information from witnesses to determine available oral information;

[0032] A transcript pointing determination module is used to obtain the suspect's transcript information, screen the suspect's transcript information for authenticity based on the available oral information, the objective clue information and the basic case information, and determine the transcript pointing information;

[0033] A suspect determination module, used to determine the inference node information set and preliminary suspect information based on the available oral information and the transcript pointing information;

[0034] A report generation module is used to determine a case reasoning report based on the preliminary suspect information and the reasoning node information set. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0036] Figure 1 A schematic diagram of an application scenario provided for an embodiment of the present application;

[0037] Figure 2 A flowchart of a clue data analysis method based on a large model provided in one embodiment of the present application;

[0038] Figure 3 A schematic diagram of the structure of a clue data analysis system based on a large model provided in one embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0040] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally means that the associated objects before and after are in an "or" relationship.

[0041] The embodiments of the present application are further described in detail below in conjunction with the drawings in the specification.

[0042] Existing case reasoning technology is difficult to cope with the complex and changing data environment, effectively and clearly sort out the existing massive clues, and obtain some clue information that helps improve the efficiency of case reasoning. It is difficult to sort out the case reasoning process and related suspects.

[0043] Based on this, the present application provides a clue data analysis method and system based on a large model, which obtains the oral information of witnesses, objective clues and basic case information provided by the police, uses objective clues and basic case information to screen the oral information of witnesses, determines which information is valid, obtains available oral information, and then combines objective clues and available oral information to screen the suspect's transcript information, obtains information that lies and conceals in the suspect's transcript, and then determines a number of pieces of information about each suspect that can point to the suspect's suspicion, and generates a case reasoning report in the form of a node graph for the police's reference with this information and the corresponding suspect. Through a systematic method, effective information can be quickly screened, redundant data can be reduced, and the efficiency of case detection can be improved. Multiple data sources and clues can be combined to ensure the authenticity and reliability of information and reduce misjudgments caused by inconsistent information.

[0044] Figure 1 A schematic diagram of an application scenario provided for this application. In the process of reasoning a criminal case, the method provided by this application is applied to verify and screen the suspect's transcript information based on the basic information of the case, objective clue information and oral information of witnesses, and finally reason and generate a case reasoning report for police reference.

[0045] Specifically, the method provided in this application is applied to any server, and the oral information of witnesses, objective clues, basic case information and suspect written information provided by the police are obtained through the server. The oral information of witnesses is screened based on the objective clues and basic case information, and the new clues obtained after screening are combined with the objective clues to verify the suspect written information, and the preliminary suspect and reasoning node information set is obtained, and finally a case reasoning report is generated for the police reference. Through the integration and effective screening of multi-source information, the information quality in the case investigation process is improved, and the authenticity and validity of the data are ensured. The generated reasoning report can provide strong support for the subsequent handling of the case and help law enforcement personnel make more accurate decisions.

[0046] For specific implementation methods, please refer to the following embodiments.

[0047] Figure 2 This is a flowchart of a clue data analysis method based on a large model provided in an embodiment of the present application. The method of this embodiment can be applied to the server in the above scenario. Figure 2 As shown, the method includes:

[0048] S201. Obtain oral information from witnesses, objective clue information, and basic case information. Based on the objective clue information and basic case information, screen the effectiveness of the oral information from witnesses to determine available oral information.

[0049] Witness oral information may be information related to the case that is orally stated by witnesses related to the case.

[0050] Objective clues can be directly observed, measured or verified during case investigations. These clues do not rely on personal subjective opinions or memories, and are usually extracted through scientific methods and technical means.

[0051] Basic case information can be basic data and information related to the case, including case event overview and occurrence information.

[0052] The available oral information may be highly credible information from the oral information of eyewitnesses that is helpful for case reasoning.

[0053] Specifically, in the process of reasoning a criminal case, due to various reasons, such as confusion in the witness's memory, etc., the witness's oral information may not be completely credible, and thus all the witness's oral information cannot be used as clues for case reasoning, otherwise it will have a negative impact on the case reasoning. Therefore, the witness's oral information, objective clue information and basic case information provided by the police are obtained, and the witness's oral information is screened with reference to the objective clue information and basic case information, and the information that can be verified by the objective clue information and basic case information and the supplementary information that can help with case reasoning are used as available oral information.

[0054] S202: Obtain the suspect's written information, and screen the suspect's written information for authenticity based on available oral information, objective clue information, and basic case information to determine the information pointed to by the written information.

[0055] Suspect record information can be all the information recorded when the police interrogate the suspect.

[0056] The information pointed to by the transcript can be the information pointing to the suspect extracted after in-depth analysis of the suspect's confession.

[0057] Specifically, in the process of reasoning of criminal cases, when questioning suspects, some suspects may give untrue answers such as concealment and deception to mislead the reasoning of the case or to clear themselves of suspicion. It is necessary to screen the suspect's written information and determine the untrue information in it to further judge the suspect's degree of suspicion. Therefore, by obtaining the suspect's written information provided by the police, combined with the available oral information, objective clue information and basic case information for verification, NLP (Natural Language Processing) is used to determine whether there is any content in the suspect's written information that is inconsistent with the available oral information, objective clue information and basic case information. If so, it means that the suspect has lied and deliberately concealed the information. These contradictory content information will be used as written information to provide support for the suspect's reasoning and judgment.

[0058] S203. Determine the reasoning node information set and preliminary suspect information based on the transcript pointing information.

[0059] The inference node information set may be a number of node information pointing to the suspect's lying behavior determined on the basis of the transcript pointing information.

[0060] Preliminary suspect information can be information about a suspected criminal suspect derived based on existing clues.

[0061] Specifically, in the process of reasoning a criminal case, the clues obtained must have clear directionality and point to clear suspects. The clue information that can point to the suspect can be intuitively displayed in the form of reasoning nodes. Through these nodes, the suspect's behavior and motive can be easily traced. Therefore, each transcript pointing information is taken as a node, and then a reasoning node information set is obtained. According to the content of the transcript pointing information, several preliminary suspects are determined, and each node points to a suspect.

[0062] S204: Determine a case reasoning report based on the preliminary suspect information and the reasoning node information set.

[0063] The case reasoning report may be a report of the reasoning results of the case that is finally generated for the reference of the police.

[0064] Specifically, after the reasoning of a criminal case is completed, the generated reasoning result report can provide a more intuitive reference for the police and improve the efficiency of solving the case. Therefore, the reasoning node information set is used as the first-level node, and the preliminary suspect information is used as the second-level node. The first-level node points to the corresponding second-level node, and then an orderly arranged and pointed case reasoning report is obtained. The case reasoning report is output in the form of a node diagram through a human-computer interaction device using data visualization technology for police reference.

[0065] Through the method provided in this embodiment, the available oral information is analyzed and screened, which can eliminate the impact of useless oral information on the case and provide a basis for verifying the authenticity of the subsequent suspect's written information. By comparing and screening out the written information that is inconsistent with the available oral information and objective clues, and by comparing the inconsistencies of the information, it is further clarified which written information can be used as directional evidence, which helps to determine the degree of suspicion of the suspect. The clues related to the suspect are intuitively displayed in the form of nodes, making the case analysis more organized. By having each node point to a different preliminary suspect, it helps the police to sort out the final suspect.

[0066] In some embodiments, based on objective clue information and basic case information, the oral information of witnesses is screened, and the information in the oral information of witnesses that can match the objective clue information and the basic case information is determined as preliminary available clues; the information in the oral information of witnesses that is inconsistent with the objective clue information and the basic case information is determined as unusable clues; based on the unusable clues and the preliminary available clues, the pending clue information is determined; based on the preliminary available clues and the pending clue information, the available oral information is determined.

[0067] Preliminary available clues can be highly reliable information initially confirmed from the oral information of eyewitnesses.

[0068] Unusable clues may be information in the witness's oral statement that has no reference value.

[0069] The pending clue information can be supplementary clue information in the witness's oral information that does not obviously contradict the existing clues.

[0070] Specifically, in the process of criminal case reasoning, after obtaining oral information from several witnesses, due to the high subjectivity of the information, the oral information from the witnesses is not completely credible, and some of it is obviously contrary to the existing objective clues of the case. Therefore, the objective clue information is analyzed, and key information such as physical evidence and third-party evidence is extracted to obtain the objective clue feature set; the basic case information obtained is analyzed, and basic information such as the time, location, course of events, and persons involved are extracted to obtain the basic information feature set; the oral information from the witnesses is analyzed, pre-processed, and irrelevant information is removed, and the key information in the oral information is extracted using NLP. Based on objective clue information and basic case information, a text similarity algorithm, such as the Jaccard similarity algorithm, is used to determine the similarity between the witness' oral information and the basic case information and the objective clue information, and a minimum similarity threshold and a maximum similarity threshold are set. Oral information below the minimum similarity threshold is regarded as an unusable clue, and oral information above the maximum similarity threshold is regarded as a preliminary available clue. Oral information with a similarity between the maximum and minimum thresholds is determined as pending clue information, and the preliminary available clues and pending clue information are determined as available oral information.

[0071] Through the method provided by this embodiment, by determining unusable clues, it is possible to reduce the time and resource investment in invalid information, concentrate energy and resources on those clues that may be useful, and thus optimize the entire investigation process. By screening the oral information of witnesses, it is possible to effectively identify effective clues that match objective clues and basic case information from the oral information. This helps to reduce the interference of invalid information on case investigations and improve investigation efficiency. Through the consistency of oral information with objective clues and basic information, the reliability of the case evidence chain is enhanced, ensuring that the screened information has a higher degree of credibility.

[0072] In some embodiments, objective image clue information and objective text clue information are determined based on objective clue information; based on pending clue information, features of pending clue information are extracted to determine a pending clue feature vector set; based on objective image clue information, an image feature vector set is determined; based on objective text clue information, a text feature vector set is determined; based on the pending clue feature vector set, in combination with the image feature vector set and the text feature vector set, a pending clue matching degree of each pending clue is determined; based on the pending clue information and the matching degrees of several pending clues, pending available clues are determined; based on the pending available clues and preliminary available clues, available oral information is determined.

[0073] The objective image clue information may be clue information presented through images in the objective clue information.

[0074] The objective text clue information may be clue information presented in text in the objective clue information.

[0075] The pending clue feature vector set may be a feature vector set that quantifies and represents the pending clue information. The pending clue feature vector may be a numerical representation of various attributes of the pending clue, such as information content, nature, source, time, etc.

[0076] The image feature vector set may be a feature vector set extracted from objective image clue information. The image feature vector usually includes a numerical representation of important features in the image (such as edges, color distribution, texture, shape, etc.).

[0077] The text feature vector set may be a collection of feature vectors extracted from objective text clue information. The text feature vector may be a numerical representation of key information of the text content, such as vectors corresponding to word frequency, word meaning, contextual relationship, and the like.

[0078] The pending available clues may be clues in the pending clue information that may be helpful for the reasoning of the current case.

[0079] Specifically, in the process of criminal case reasoning, the content of the pending clue information still contains some unknown clues that are useful for case reasoning, and there are also clues that are not helpful for the case. It is necessary to screen them to obtain clues that are useful for case reasoning. Therefore, the objective clue information is classified to obtain objective image clue information and objective text clue information. According to the pending clue information, the feature vector of the pending clue is extracted through the sentence embedding model in NLP to obtain the pending clue feature vector set. According to the objective image clue information, the objective image clue features are extracted through the CNN (Convolutional Neural Network) technology to obtain the image feature vector set. According to the objective text clue information, the objective text feature vector is extracted through the sentence embedding model in NLP to obtain the text feature vector set. Based on the above three vector feature sets, the matching degree between the vectors in the pending clue feature vector set and the vectors in the text feature vector set and the image feature vector set is determined by mathematical analysis. By comparing the matching degree with the preset matching degree threshold, it is determined whether the current pending clue is available, and the pending available clue is obtained, and then the pending available clue and the preliminary available clue are determined as available oral information.

[0080] Through the method provided by this embodiment, by classifying objective image clue information and objective text clue information, corresponding processing techniques can be used according to different types of clues to maximize the efficiency of information extraction. Through accurate feature vector extraction and matching analysis, it can help identify more accurate available clues, reduce the risk of misjudgment, and improve the scientific nature of case analysis. By screening out the available information in the pending information, this matching analysis can effectively screen out useful clues, reduce the interference of irrelevant information, and improve the accuracy and efficiency of reasoning.

[0081] In some embodiments, a plurality of weighted comprehensive feature vectors are determined based on each image feature vector of the image feature vector set and each text feature vector of the text feature vector set, specifically the following formula (1):

[0082] F c =α·F i +β·F t (1)

[0083] Among them, F c is the current weighted comprehensive feature vector, F i is the current image feature vector, F t is the current text feature vector, α is the preset image influence weight, and β is the preset text influence weight; according to a number of weighted comprehensive feature vectors and each pending clue feature vector of a number of pending clue feature vector sets, a number of pending clue matching degrees are determined, specifically as follows:

[0084]

[0085] Among them, C is the matching degree of the current pending clue, F d is the current pending clue feature vector of the pending clue feature vector set, F c is the current weighted comprehensive feature vector; according to the pending clue matching degree and the preset high matching degree threshold, if the current pending clue matching degree exceeds the preset high matching degree threshold, the current pending clue under the pending clue information is determined as a pending highly credible clue; if the current pending clue matching degree is lower than the preset high matching degree threshold but higher than the preset low matching degree threshold, the current pending clue under the pending clue information is determined as a pending fuzzy clue; according to the pending highly credible clue and the pending fuzzy clue, the pending available clue is determined.

[0086] The weighted comprehensive feature vector may be a vector obtained by weighting the image feature vector and the text feature vector and used to represent the comprehensive features of the objective clues.

[0087] The preset image influence weight may be the influence weight of the preset image feature vector on the weighted comprehensive feature vector, and may be set according to specific actual clue conditions.

[0088] The preset text influence weight may be the influence weight of the preset text feature vector on the weighted comprehensive feature vector, and may be set according to the specific actual clue situation.

[0089] The pending clue matching degree may be a numerical value used to indicate the matching degree between the pending clue information and the weighted comprehensive feature vector.

[0090] The pending highly credible clues can be clues in the pending clue information and clues that have a high matching degree with the existing clues and can be used as part of the pending available clues.

[0091] A pending fuzzy clue may be a clue in the pending clue information that is difficult to directly determine whether it is helpful for the current case reasoning.

[0092] The preset high matching degree threshold may be a preset threshold for judging whether the current pending clue has a high matching degree with the existing clue. Exceeding the threshold means that the current pending clue has a high credibility.

[0093] The preset low matching degree threshold may be a preset threshold for determining whether the current pending clue is related to the existing clue. If the threshold is exceeded, it means that the current pending clue is related, and if the threshold is lower than the threshold, it means that the current pending clue is not related to the case reasoning.

[0094] Specifically, in the process of criminal case reasoning, the oral information of the witnesses will contain some new clues. However, due to the strong subjectivity of the clues, it is impossible to directly judge whether they are helpful for the current case reasoning. In order to quantify which information in the pending clue information is available, formula (1) is designed to weight each vector in the image feature vector set and the text feature vector set to obtain a number of weighted comprehensive feature vectors, and then formula (2) is designed to calculate the cosine similarity between the current pending clue vector and the current weighted comprehensive feature vector by the cosine similarity calculation method, and at the same time standardize the result of the dot product so that the value of the cosine similarity is always between [-1,1]. When the two vectors are exactly the same (the same direction), the cosine similarity is 1. When the two vectors are orthogonal (completely different directions), the cosine similarity is 0. When the two vectors are in opposite directions, the cosine similarity is -1. That is, the value of the cosine similarity, positive values ​​indicate similarity, and negative values ​​indicate oppositeness. The preset low matching threshold is set to 0.4, and the preset high matching threshold is set to 0.8. If the current cosine similarity is less than the preset low matching threshold, it is considered that the current pending clue has no effective association with the current case, has no effective help for the current case reasoning, and is not used as a reference clue. If the current cosine similarity is greater than the preset high matching threshold, it is considered that the current pending clue has a great association with the current case, can significantly help the current case reasoning, and is used as a pending highly credible clue. If the current pending clue matching degree is lower than the preset high matching threshold, but higher than the preset low matching threshold, it is considered that the current pending clue is related to the current case, but it is unclear whether it is significantly helpful for the current case reasoning, and is used as a pending fuzzy clue. Then the pending highly credible clues and pending fuzzy clues are used as pending available clues.

[0095] By the method provided in this embodiment, by calculating the matching degree between the pending clue vector and the weighted comprehensive feature vector, the correlation between these clues can be objectively and scientifically evaluated. The value of the matching degree of the pending clues can effectively reflect the similarity between the clues and provide a quantitative basis for subsequent analysis. By standardizing the matching degree of the pending clues, the results are more interpretable and convenient for setting the matching degree threshold. By setting low matching degree and high matching degree thresholds, clues that are significantly related to the case, as well as those vague clues that may be helpful but uncertain for case reasoning, can be effectively screened out. This information screening can improve the efficiency of case handling.

[0096] In some embodiments, based on the suspect's written information and a preset timeline, the suspect's behavior information within the preset timeline is determined; based on pending highly credible clues, a number of discrete time window information is determined; based on a number of discrete time window information, a number of continuous time window information is determined; based on a number of continuous time window information, a number of witnessed suspect behavior information under the current continuous time window is determined; the witnessed suspect behavior information and the suspect's behavior information in the current continuous time window are analyzed to determine witnessed behavior difference information; pending fuzzy clues are analyzed to determine partial process reconstruction information; a number of partial process reconstruction information and a number of suspects' behavior information are analyzed to determine objective behavior difference information; based on the witnessed behavior difference information and the objective behavior difference information, the written lie information is determined; based on a number of determined written lie information, the written direction information is determined.

[0097] The preset timeline may be a preset timeline related to the time of the crime, and is used to organize and analyze the sequence of events and behavioral information related to the suspect.

[0098] Discrete time window information can be a specific time period extracted from the available oral information, usually a short time segment that contains some key events.

[0099] Continuous time window information can be a longer time period integrated from several discrete time window information. These continuous time periods can reveal broader behavioral patterns or the development of events.

[0100] Witness suspect behavior information may be a description or evidence provided by a witness about a specific behavior of the suspect within the current continuous time window.

[0101] Witness behavior discrepancy information can be the difference between the witness's description and the suspect's own description of the behavior, which contains the suspect's misrepresentation or hidden information.

[0102] The partial process reconstruction information may be the reconstruction information of the event process, which may be reconstructed based on pending fuzzy clues and clues that have a chronological connection on the timeline.

[0103] Objective behavior difference information can be information about the behavior differences in the suspect's description of his own behavior and partial process reconstruction information.

[0104] The transcript lie information can be part of the suspect's transcript that contains lies, including witness behavior difference information and objective behavior difference information.

[0105] Specifically, in the process of criminal case reasoning, the suspect's written information may contain lies due to various objective reasons of the suspect, such as being the murderer or having a vested interest in the case. Therefore, according to the suspect's written information and the preset timeline, the behavior information of the suspect in the written information on the preset timeline is obtained, and the pending highly credible clues are analyzed to obtain the behavior information of several suspects in several time periods by several witnesses, and obtain several discrete time window information. Then, through the DTW (Dynamic Time Warping) technology, it is analyzed whether there is time continuity in the timeline of several discrete time window information, and several continuous time window information is obtained, and then the behavior information of the suspect in each continuous time window is obtained. The key behavior information of the suspect in each continuous time window is extracted to obtain the witnessed behavior information of the suspect, and compared with the behavior information in the suspect's written information. Through NLP and DTW technology, the witnessed behavior difference information of the suspect's behavior information in the written information and the witnessed behavior information is obtained. By analyzing the pending fuzzy clues and processing the associated fuzzy clues through the Bayesian network modeling tool, partial process reconstruction information that is closer to the actual process can be obtained. By comparing the partial process reconstruction information with the behavioral information in the suspect's written record information, objective behavior difference information is obtained through NLP and DTW technology, where the suspect's behavioral information in the written record information is different from the partial process reconstruction information. If there are witnessed behavior differences and / or objective behavior differences, the witnessed behavior difference information and / or objective behavior difference information are determined as written record lie information, and based on several written record lie information, analyze which suspects have lied, and then determine the written record pointing information, and determine which written record lie information points to which suspects.

[0106] Through the method provided in this embodiment, by analyzing the suspect's transcript information and the preset timeline, the suspect's behavior information within a specific time period can be extracted. This provides basic data for subsequent comparison and analysis. Based on the pending fuzzy clues and pending highly credible clues of the witnesses, the suspect's transcript is analyzed to determine whether the suspect has lied. By analyzing the discrete time window, the suspect's behavior information under the continuous time window can be obtained, which provides a basis for subsequent comparison, and can provide dynamic process analysis and comparison under the continuous time window, which can make the comparison result more convincing and reliable.

[0107] In some embodiments, the undetermined fuzzy clues are analyzed to determine a number of objective discrete information; a number of discrete information feature vectors are determined based on the number of objective discrete information; and the discrete information correlation is determined based on the number of discrete information feature vectors, which is specifically the following formula (3):

[0108]

[0109] Among them, S(Xi,Xj) is the discrete information correlation between the i-th discrete information feature vector and the j-th discrete information feature vector, Xi is the i-th discrete information feature vector, and Xj is the j-th discrete information feature vector; according to a number of objective discrete information, a number of discrete information correlations and a preset correlation threshold, a number of discrete information correlation chains are determined; according to the number of discrete information correlation chains, partial process reconstruction information is determined.

[0110] Objective discrete information can be several different pieces of information provided by several witnesses among ambiguous clues.

[0111] The discrete information feature vector may be a vector used to represent key features of objective discrete information.

[0112] The discrete information correlation degree can be the correlation degree between several objective discrete information.

[0113] The discrete information association chain can be a series of interconnected information sets formed based on the correlation between discrete information.

[0114] The preset correlation threshold may be a preset correlation value used to determine whether there is a correlation between objective discrete information.

[0115] Specifically, in the process of reasoning in a criminal case, some of the pending fuzzy clues are relevant and helpful to the current case reasoning, while other clues are relevant but not helpful. For example, the pending fuzzy clues may be in the following situation: different witnesses have significantly different descriptions of the relevant information of the same suspect, and the information described by one witness can significantly help the case reasoning, but the information described by another witness is related to the case but not obviously helpful for the case reasoning. It is necessary to integrate and analyze the useful information to obtain information that is helpful for case reasoning. Therefore, the pending fuzzy clues are analyzed to obtain several objective discrete information provided by several witnesses. Through NLP, several discrete information feature vectors are extracted, and then the cosine similarity between each two discrete information feature vectors is calculated by designing formula (3). The correlation between the feature vectors is then reflected by the cosine similarity. Then, the preset correlation threshold is used for comparison. If the correlation between two objective discrete information exceeds the threshold, it is considered that the two information are highly correlated. Then, the correlation between each two objective discrete information is calculated and compared to obtain several sets of objective discrete information that are correlated with each other, and then several discrete information association chains are obtained. Through NLP and RF (Random Forest) algorithms, several discrete information association chains are analyzed. Generative models, such as GAN (Generative Adversarial Networks), are used to restore the scene of some processes and obtain partial process reconstruction information.

[0116] Through the method provided in this embodiment, the correlation between feature vectors is calculated by designing a formula, and combined with a preset correlation threshold, highly correlated information is identified, thereby screening out useful information and ignoring redundant or irrelevant information, thereby improving the effectiveness of case reasoning. By calculating and comparing the correlation between each objective discrete information, a number of logical chains with multiple interrelated objective discrete information are formed, which helps to construct the context and context of the case, provide more structured data support for subsequent analysis, improve data analysis efficiency, and thus increase reasoning speed. By restoring part of the process, more intuitive clues can be provided for subsequent analysis. Extracting effective clues from fuzzy clues can make the processing logic more rigorous and logically reasonable.

[0117] In some embodiments, based on the transcript pointing information, the current suspect of the current transcript pointing information is determined; the witnessed behavior difference information and the objective behavior difference information are determined as the pointing information node of the current suspect; based on several pointing information nodes, an inference node information set is determined; based on several current suspects, several preliminary suspects are determined.

[0118] The pointing information nodes may be several information nodes that clearly point to a suspect's lying behavior.

[0119] The current suspect may be a suspect who has lied in the written record information.

[0120] Specifically, in the reasoning of criminal cases, a complete reasoning logic is required, and the clues obtained must have clear directionality and point to clear suspects. The clue information that can point to the suspect can be intuitively displayed in the form of reasoning nodes. Therefore, the transcript pointing information of each suspect is analyzed to determine the suspect of the current transcript, and each information in the transcript that can point to the suspect's lying behavior is used to construct a pointing information node, thereby obtaining a reasoning node information set and determining the current suspect as a preliminary suspect.

[0121] Through the method provided in this embodiment, each transcript pointing information is analyzed one by one and organized into pointing information nodes, which can effectively associate clues with specific suspects, reduce ambiguity, and ensure that each clue can point to a clear target. Converting the transcript pointing information into reasoning nodes to form a structured information system helps to clarify the context of the case. By nodeizing the information, it can be ensured that each clue and conclusion has a basis, which enhances the credibility and effectiveness of the investigation results. It provides a content basis for the generation of subsequent case reasoning reports. Determining preliminary suspects can screen out suspects who are really suspected of committing crimes, narrow the scope of suspects, and improve the efficiency of subsequent police handling.

[0122] In some embodiments, the witnessed behavior difference information and the objective behavior difference information are analyzed to determine the amount of difference information and the duration of the difference information; the amount of difference information and the duration of the difference information are analyzed to determine the difference weight of each difference information and the suspicion value of the preliminary suspect; and the case reasoning report is determined based on the difference weight, suspicion value, reasoning node information set and the preliminary suspect.

[0123] The number of differential information can be the behavioral information in the suspect's written record and the highly convincing oral information of the witness, and the number of different pieces of information that exist under the confirmation of objective clues and basic case information.

[0124] The duration of the difference information may be the sum of the duration of each difference information of the current suspect. For example, if the suspect lies during the time period from 1:00 to 2:00, the time period is the duration of the current difference information.

[0125] The suspicion value may be a value used to evaluate the degree of suspicion of a suspect, and the larger the value, the greater the degree of suspicion of the suspect.

[0126] The difference weight may be the influence weight of the current difference information on the suspicion value.

[0127] Specifically, after the reasoning of the criminal case is completed, the different suspicion levels and related pointing information of each preliminary suspect need to be output in the form of a report for the police's reference. Therefore, the witness behavior difference information and the objective behavior difference information are analyzed, and each difference information of the current suspect's written information is summed to obtain the number of difference information of the two difference information. When analyzing the duration of each difference information, the difference information duration of the current suspect's written information is summed up. The length of the difference information duration can reflect the duration of the information concealed by the suspect. The longer the duration, the greater the impact on the suspect's suspicion value. According to the obtained number of difference information and the duration of the difference information, the difference weight of each difference information is calculated through mathematical analysis, and then the suspicion value of the current suspect is obtained. Then, the difference weight, suspicion value, reasoning node information set and preliminary suspect are combined, and through data visualization tools such as Tableau, the reasoning node information set is used as the first-level node, the preliminary suspect is used as the second-level node, the difference weight is used as the weight of the reasoning node information set pointing to the preliminary suspect, and the suspicion value is used as the attribute of the preliminary suspect, and then a case reasoning report is generated.

[0128] Through the method provided in this embodiment, a quantifiable suspicion value is formed by analyzing the difference information and duration, and a systematic method is used to evaluate suspects instead of relying solely on subjective judgment. This quantitative evaluation can increase the objectivity and reliability of the reasoning results. By generating a case reasoning report for police reference, the readability and comprehensibility of information are enhanced, and automated data processing and analysis reduce the time of manual analysis and improve case reasoning efficiency. The display of the suspicion value can provide the police with a preliminary reference for the suspect and provide a reference for the final determination of the suspect.

[0129] In some embodiments, the amount of difference information and the duration of the difference information are analyzed to determine the difference weight of each difference information and the suspicion value of the preliminary suspect, which is specifically the following formula (4):

[0130]

[0131] Among them, S m is the suspicion value of the mth preliminary suspect, α is the preset time influence weight, k is the preset difference weight influence coefficient, n is the number of difference information of the mth preliminary suspect, T mi is the duration of the difference information of the ith difference information of the mth preliminary suspect, is the difference weight of the ith difference information of the mth preliminary suspect.

[0132] The preset difference weight influence coefficient may be a preset coefficient for controlling the degree of influence of the difference weight on the suspicion value. The preset difference weight influence coefficient may be obtained by fitting and analyzing the historical difference weight and the suspicion value through a data fitting algorithm, such as a nonlinear regression fitting algorithm.

[0133] The preset time impact weight may be a preset weight for controlling the degree of influence of the duration of the difference information on the difference weight. The preset time impact weight may be obtained by fitting and analyzing the duration of the historical difference information and the difference weight through a data fitting algorithm, such as a linear regression fitting algorithm.

[0134] Specifically, when determining that a suspect is lying, it is necessary to quantify the degree of suspicion based on the relevant lying information. Therefore, formula (4) is used to quantify the influence of several difference information of each suspect and the duration of each difference information on the suspect's suspicion value. Based on the duration T of each difference information, mi , calculate the difference weight of each difference information of the mth preliminary suspect Because when the duration of the difference information is short, the difference weight increases slowly with the increase of the duration of the difference information, and when the duration of the difference information is long, the difference weight increases rapidly with the increase of the duration of the difference information. Therefore, the influence of duration on the difference weight expressed in the form of an exponent is nonlinear, which can amplify the influence of long-term differences, and then traverse each difference information, and obtain the suspicion value of each preliminary suspect through linear weighted summation.

[0135] Through the method provided by this embodiment, the various difference information of each suspect can be quantified by designing a formula. This quantification helps to objectively evaluate the degree of suspicion of each suspect. The formula can effectively capture the nonlinear characteristics of the impact of duration on difference information. This means that difference information with a longer duration will contribute more significantly to the suspicion value, thereby enhancing the sensitivity of the model. By traversing each difference information and performing a linear weighted summation, the comprehensive suspicion value of each preliminary suspect can be finally obtained. This aggregation method can integrate the contribution of various types of difference information, thereby forming a quantitative indicator for the comprehensive evaluation of the suspect, making the result more convincing.

[0136] Figure 3 A schematic diagram of a clue data analysis system based on a large model is provided in one embodiment of the present application, such as Figure 3 As shown, a clue data analysis system 300 based on a large model in this embodiment includes: an available information determination module 301, a transcript pointing determination module 302, a suspect determination module 303, and a report generation module 304.

[0137] The available information determination module 301 is used to obtain the oral information of the witness, the objective clue information and the basic information of the case, and to screen the effectiveness of the oral information of the witness based on the objective clue information and the basic information of the case to determine the available oral information;

[0138] The transcript pointing determination module 302 is used to obtain the suspect's transcript information, and based on the available oral information, the objective clue information and the basic case information, screen the suspect's transcript information for authenticity, and determine the transcript pointing information;

[0139] A suspect determination module 303, for determining a reasoning node information set and preliminary suspect information according to the available oral information and the transcript pointing information;

[0140] The report generation module 304 is used to determine a case reasoning report based on the preliminary suspect information and the reasoning node information set.

[0141] Optionally, the available information determination module 301 is specifically used to: screen the eyewitness oral information based on the objective clue information and the basic information of the case, and determine the information in the eyewitness oral information that can match the objective clue information and the basic information of the case as preliminary available clues; determine the information in the eyewitness oral information that is inconsistent with the objective clue information and the basic information of the case as unavailable clues; determine pending clue information based on the unavailable clues and the preliminary available clues; and determine the available oral information based on the preliminary available clues and the pending clue information.

[0142] Optionally, when the available information determination module 301 determines the available oral information based on the preliminary available clues and the pending clue information, it is specifically used to: determine objective image clue information and objective text clue information based on the objective clue information; extract pending clue information features based on the pending clue information to determine a pending clue feature vector set; determine an image feature vector set based on the objective image clue information; determine a text feature vector set based on the objective text clue information; determine the pending clue matching degree of each pending clue based on the pending clue feature vector set, in combination with the image feature vector set and the text feature vector set; determine pending available clues based on the pending clue information and the matching degrees of several pending clues; determine the available oral information based on the pending available clues and the preliminary available clues.

[0143] Optionally, when determining the pending available clues according to the pending clue information and the pending clue matching degree set, the available information determination module 301 is specifically used to: determine a number of weighted comprehensive feature vectors according to each image feature vector set and each text feature vector of the text feature vector set, which is specifically the following formula:

[0144] F c =α·F i +β·F t ;

[0145] Among them, F c is the weighted comprehensive feature vector, F i is the current image feature vector, F t is the current text feature vector, α is the preset image influence weight, and β is the preset text influence weight; according to the weighted comprehensive feature vectors and each pending clue feature vector of the pending clue feature vector set, the matching degree of the pending clues is determined, which is specifically the following formula:

[0146]

[0147] Among them, C is the matching degree of the pending clues at present, F d is the current pending clue feature vector of the pending clue feature vector set, F c is the current weighted comprehensive feature vector; according to the pending clue matching degree and the preset high matching degree threshold, if the current pending clue matching degree exceeds the preset high matching degree threshold, the current pending clue under the pending clue information is determined as a pending highly credible clue; if the current pending clue matching degree is lower than the preset high matching degree threshold but higher than the preset low matching degree threshold, the current pending clue under the pending clue information is determined as a pending fuzzy clue; according to the pending highly credible clue and the pending fuzzy clue, the pending available clue is determined.

[0148] Optionally, the transcript direction determination module 302 is specifically used to: determine the suspect's behavior information within the preset timeline based on the suspect's transcript information and the preset timeline; determine a number of discrete time window information based on the pending highly credible clues; determine a number of continuous time window information based on the number of discrete time window information; determine a number of witnessed suspect behavior information under the current continuous time window based on the number of continuous time window information; analyze the witnessed suspect behavior information and the suspect's behavior information in the current continuous time window to determine witnessed behavior difference information; analyze the pending fuzzy clues to determine partial process reconstruction information; analyze a number of partial process reconstruction information and the behavior information of a number of suspects to determine objective behavior difference information; determine transcript lie information based on the witnessed behavior difference information and the objective behavior difference information; determine the transcript direction information based on the number of determined transcript lie information.

[0149] Optionally, when analyzing the pending fuzzy clues and determining the partial process reconstruction information, the transcript pointing determination module 302 is specifically used to: analyze the pending fuzzy clues to determine a number of objective discrete information; determine a number of discrete information feature vectors based on the objective discrete information; determine the discrete information correlation degree based on the discrete information feature vectors, which is specifically the following formula:

[0150]

[0151] Among them, S(Xi,Xj) is the discrete information correlation between the i-th discrete information feature vector and the j-th discrete information feature vector, Xi is the i-th discrete information feature vector, and Xj is the j-th discrete information feature vector; according to the plurality of objective discrete information, the plurality of discrete information correlations and the preset correlation threshold, a plurality of discrete information correlation chains are determined; according to the plurality of discrete information correlation chains, the partial process reconstruction information is determined.

[0152] Optionally, the suspect identification module 303 is specifically configured to:

[0153] According to the transcript pointing information, determine the current suspect of the current transcript pointing information; determine the witnessed behavior difference information and the objective behavior difference information as the pointing information node of the current suspect; according to several of the pointing information nodes, determine the reasoning node information set; according to several of the current suspects, determine several of the preliminary suspects.

[0154] Optionally, the report generation module 304 is specifically used to: analyze the witnessed behavior difference information and the objective behavior difference information to determine the amount of difference information and the duration of the difference information; analyze the amount of difference information and the duration of the difference information to determine the difference weight of each difference information and the suspicion value of the preliminary suspect; determine the case reasoning report based on the difference weight, the suspicion value, the reasoning node information set and the preliminary suspect.

[0155] Optionally, when the report generating module 304 analyzes the amount of the difference information and the duration of the difference information to determine the difference weight of each difference information and the suspicion value of the preliminary suspect, the specific formula is as follows:

[0156] Among them, S m is the suspicion value of the mth preliminary suspect, α is the preset time influence weight, k is the preset difference weight influence coefficient, n is the amount of difference information of the mth preliminary suspect, T mi is the duration of the difference information of the i-th difference information of the m-th preliminary suspect, is the difference weight of each difference information of the mth preliminary suspect.

[0157] The system of this embodiment can be used to execute the method of any of the above embodiments. The implementation principles and technical effects are similar and will not be described in detail here.

Claims

1. A clue data analysis method based on a large model, characterized in that: include: Obtaining oral information from witnesses, objective clue information and basic case information, and screening the effectiveness of the oral information from witnesses based on the objective clue information and the basic case information to determine usable oral information; Obtaining the suspect's written information, and screening the authenticity of the suspect's written information based on the available oral information, the objective clue information, and the basic case information to determine the information pointed to by the written information; Determine the reasoning node information set and preliminary suspect information according to the transcript pointing information; A case reasoning report is determined based on the preliminary suspect information and the reasoning node information set.

2. The method according to claim 1, characterized in that The validity screening of the oral information of the witness based on the objective clue information and the basic information of the case to determine the usable oral information includes: Based on the objective clue information and the basic case information, the eyewitness oral information is screened, and the information in the eyewitness oral information that can match the objective clue information and the basic case information is determined as a preliminary available clue; Determine the information in the oral information of the witness that contradicts the objective clue information and the basic information of the case as unusable clues; Determining pending clue information according to the unavailable clues and the preliminary available clues; The available oral information is determined according to the preliminary available clues and the pending clue information.

3. The method according to claim 2, characterized in that The step of determining the available oral information according to the preliminary available clues and the pending clue information includes: Determining objective image clue information and objective text clue information according to the objective clue information; Extracting features of the pending clue information according to the pending clue information and determining a pending clue feature vector set; Determining an image feature vector set according to the objective image clue information; Determining a text feature vector set according to the objective text clue information; Determine the pending clue matching degree of each pending clue according to the pending clue feature vector set, in combination with the image feature vector set and the text feature vector set; Determining available pending clues according to the pending clue information and the matching degrees of the pending clues; The available oral information is determined according to the pending available clues and the preliminary available clues.

4. The method according to claim 3, characterized in that The step of determining the pending available clues according to the pending clue information and the pending clue matching degree set includes: According to each image feature vector of the image feature vector set and each character feature vector of the character feature vector set, a plurality of weighted comprehensive feature vectors are determined, which are specifically the following formulas: F c =α·F i +β·F t ; Among them, F c is the weighted comprehensive feature vector, F i is the current image feature vector, F t is the current text feature vector, α is the preset image influence weight, and β is the preset text influence weight; According to the weighted comprehensive feature vectors and each pending clue feature vector of the pending clue feature vector set, the matching degree of the pending clues is determined, which is specifically the following formula: Among them, C is the matching degree of the pending clues at present, F d is the current pending clue feature vector of the pending clue feature vector set, F c is the weighted comprehensive feature vector currently described; According to the pending clue matching degree and the preset high matching degree threshold, if the current pending clue matching degree exceeds the preset high matching degree threshold, the current pending clue under the pending clue information is determined as a pending highly credible clue; If the matching degree of the current pending clue is lower than the preset high matching degree threshold but higher than the preset low matching degree threshold, the current pending clue under the pending clue information is determined as a pending fuzzy clue; Determine pending available clues according to the pending highly credible clues and the pending fuzzy clues.

5. The method according to claim 4, characterized in that The authenticity screening of the suspect's written record information based on the available oral information, the objective clue information and the basic case information to determine the written record pointing information includes: Determining the suspect's behavior information within the preset timeline based on the suspect's written information and the preset timeline; Determine a number of discrete time window information according to the pending highly credible clues; Determine a plurality of continuous time window information according to the plurality of discrete time window information; Determining the witness suspect behavior information in a number of current continuous time windows according to the continuous time window information; Analyze the witness suspect behavior information and the suspect behavior information in the current continuous time window to determine witness behavior difference information; Analyze the pending fuzzy clues to determine partial process reconstruction information; analyzing the partial process reconstruction information and the behavior information of the suspects to determine objective behavior difference information; Determining the recorded lie information according to the witnessed behavior difference information and the objective behavior difference information; The transcript pointing information will be determined based on some of the information about lies in the transcript.

6. The method according to claim 5, characterized in that The analyzing the pending fuzzy clues to determine partial process reconstruction information includes: Analyze the pending fuzzy clues to determine a number of objective discrete information; Determining a plurality of discrete information feature vectors according to the plurality of objective discrete information; According to the discrete information feature vectors, the discrete information association degree is determined, which is specifically the following formula: Wherein, S(Xi,Xj) is the discrete information association degree between the i-th discrete information feature vector and the j-th discrete information feature vector, Xi is the i-th discrete information feature vector, and Xj is the j-th discrete information feature vector; Determining a plurality of discrete information association chains according to the plurality of objective discrete information, the plurality of discrete information association degrees and a preset association degree threshold; The partial process reconstruction information is determined according to a plurality of discrete information association chains.

7. The method according to claim 6, characterized in that Determining the reasoning node information set and preliminary suspect information based on the transcript pointing information includes: Determine the current suspect pointed to by the written record according to the written record; Determine the witnessed behavior difference information and the objective behavior difference information as the pointing information node of the current suspect; Determine the reasoning node information set according to the plurality of pointing information nodes; Based on the plurality of current suspects, a plurality of preliminary suspects are determined.

8. The method according to claim 7, characterized in that Determining a case reasoning report based on the preliminary suspect information and the reasoning node information set includes: Analyzing the witnessed behavior difference information and the objective behavior difference information to determine the amount of difference information and the duration of the difference information; Analyzing the amount of the difference information and the duration of the difference information, and determining the difference weight of each difference information and the suspicion value of the preliminary suspect; The case reasoning report is determined according to the difference weight, the suspicion value, the reasoning node information set and the preliminary suspect.

9. The method according to claim 8, characterized in that The analyzing the amount of the difference information and the duration of the difference information to determine the difference weight of each difference information and the suspicion value of the preliminary suspect is specifically the following formula: Among them, S m is the suspicion value of the mth preliminary suspect, α is the preset time influence weight, k is the preset difference weight influence coefficient, n is the amount of difference information of the mth preliminary suspect, T mi is the duration of the difference information of the i-th difference information of the m-th preliminary suspect, is the difference weight of each difference information of the mth preliminary suspect.

10. A clue data analysis system based on a large model, characterized in that: include: An available information determination module is used to obtain oral information from witnesses, objective clue information and basic case information, and based on the objective clue information and the basic case information, perform validity screening on the oral information from witnesses to determine available oral information; A transcript pointing determination module is used to obtain the suspect's transcript information, screen the suspect's transcript information for authenticity based on the available oral information, the objective clue information and the basic case information, and determine the transcript pointing information; A suspect determination module, used to determine the inference node information set and preliminary suspect information based on the available oral information and the transcript pointing information; A report generation module is used to determine a case reasoning report based on the preliminary suspect information and the reasoning node information set.

Images