Intelligent auxiliary device and method for case interrogation

Abstract

The present invention discloses an intelligent case interrogation auxiliary device and method, S1, collecting audio data, video data and on-site manual recording text during the case interrogation process; S2, generating synchronous interrogation text data; S3, generating structured case record data; S4, real-time analysis of the collected video data, generating biometric data related to the interrogation object; S5, generating comprehensive case record data containing semantic information and emotional state; S6, inputting the comprehensive case record data into the case comparison module, optimizing the case comparison path using the whale optimization algorithm, and identifying historical case records related to the current case; S7, generating case comparison result data related to the current case; S8, generating intelligent interrogation prompt data. The present invention not only significantly improves the efficiency and accuracy of the interrogation auxiliary system, but also provides key technical support for the in-depth analysis of the case and the optimization of the interrogation strategy.

Description

Technical Field

[0001] The present invention relates to the field of interrogation assistance technology, and in particular to an intelligent assistance device and method for case interrogation. Background Art

[0002] With the development of artificial intelligence and big data technology, the application of intelligent assistance systems in the judicial field has gradually become a research hotspot. Case interrogation, as a key link in the judicial process, involves real-time processing and analysis of multimodal data such as audio, video, and manual recording. However, traditional interrogation assistance technology has significant limitations in multimodal data processing, case comparison, and logical reasoning, and it is difficult to meet the needs of modern judicial scenarios for efficiency and intelligence.

[0003] In recent years, some case management and auxiliary analysis systems have gradually been applied in judicial practice. Existing systems usually rely on keyword matching or simple logical rules to conduct preliminary retrieval and comparison of case information. However, these technologies have exposed the following defects in practical applications: On the one hand, the retrieval method based on keyword matching lacks semantic depth and cannot accurately capture the complex logical associations and contextual semantic relationships in the case; on the other hand, the optimization capabilities of existing technologies are insufficient and cannot efficiently process large-scale historical case databases, resulting in low retrieval efficiency and inaccurate results. In addition, the existing systems fail to effectively combine semantic information when processing the emotional analysis and biometric recognition of the interrogated subjects, resulting in a weak correlation between the interrogation data and historical case records, making it difficult to provide judicial personnel with in-depth and comprehensive case comparison and analysis.

[0004] In summary, the existing technology has significant deficiencies in the efficiency of multimodal data processing, the accuracy of case semantic matching, and the reasoning of logical relationships between cases. It is difficult to meet the real-time, precision, and intelligence requirements of modern interrogation scenarios. Therefore, there is an urgent need for a new technical method that can integrate multimodal data, optimize case matching paths, and deeply explore case semantics and their contextual relevance, so as to comprehensively improve the efficiency and accuracy of case interrogation assistance systems. Summary of the invention

[0005] One purpose of the present invention is to propose an intelligent case interrogation assistance device and method, which not only significantly improves the efficiency and accuracy of the interrogation assistance system, but also provides key technical support for in-depth analysis of cases and optimization of interrogation strategies.

[0006] A case interrogation intelligent assistance method according to an embodiment of the present invention includes the following steps:

[0007] S1. Collect audio and video data and on-site manual recording text during the case interrogation process;

[0008] S2, performing speech recognition processing on the collected recording data to generate synchronous interrogation text data;

[0009] S3, inputting the synchronous interrogation text data into the large language model to generate structured case transcript data;

[0010] S4, analyzing the collected video data in real time to generate biometric data related to the interrogation subject;

[0011] S5, fusing the generated biometric data with the structured case record data to generate comprehensive case record data containing semantic information and emotional state;

[0012] S6. Input the comprehensive case record data into the case comparison module, use the whale optimization algorithm to optimize the case comparison path, and identify the historical case records related to the current case;

[0013] S7. Analyze the matched historical case records, extract key clues, evidence chains and case patterns in the historical case records, and combine them with key statements in the comprehensive case transcript data to generate case comparison result data related to the current case;

[0014] S8. Generate intelligent interrogation prompt data based on comprehensive case record data, biometric data, case comparison result data and related information in historical case records.

[0015] Optionally, S1 includes the following specific contents:

[0016] S11. Obtaining audio data generated during the interrogation process through audio acquisition equipment , using speech enhancement technology to remove recorded data background noise and interference sound to generate the final recording data ;

[0017] S12, obtaining video data V(x, y, t) during the interrogation process through a video acquisition device, where x and y are spatial coordinates in the video frame, and t is time, and smoothing multiple frames using motion compensation technology to generate video data V'(x, y, t) with optimized frame rate;

[0018] S13. Obtaining on-site manual recording text through auxiliary input devices .

[0019] Optionally, S2 includes the following specific contents:

[0020] S21, record the data Input the speech recognition model based on the end-to-end speech processing algorithm to Converted into preliminary text data ;

[0021] S22, the preliminary text data Manually record text on site Alignment is performed by timestamp, by calculating the temporal matching degree of the text paragraphs Generate aligned text data :

[0022] ;

[0023] Where n is the number of paragraphs in the text data. Represents the timestamp matching function for each piece of text;

[0024] S23, splicing the aligned text data, removing redundancy and formatting, and generating synchronous interrogation text data .

[0025] Optionally, S3 includes the following specific contents:

[0026] S31, will synchronize the interrogation text data Input the large language model, which performs semantic analysis on the synchronous interrogation text data based on natural language processing technology to extract case-related semantic features ;

[0027] S32. Using large language models to analyze semantic features Perform logical reasoning, identify the logical relationships between statements in the synchronous interrogation text data, and construct a logical association matrix :

[0028] ;

[0029] Among them, i, j represent the index of semantic units;

[0030] S33, based on the logical association matrix and semantic features , organize the case description content through the context optimization model to generate structured case record data , including case description , Time clues , key figures and the orderly arrangement of statement logic.

[0031] Optionally, S4 includes the following specific contents:

[0032] S41, input the video data V'(x, y, t) into the face recognition model to extract the facial features in the video data ;

[0033] S42, using the face recognition model to compare the face features with the known identity data in the historical database Match the identity information of the interrogation object :

[0034] ;

[0035] in, is the similarity calculation function;

[0036] S43, perform facial expression analysis on the facial features in the video data V'(x, y, t) to extract the facial expression data of the interrogated subject , including emotion categories and emotional intensity ;

[0037] S44, detecting expression changes in the video data V'(x, y, t) to generate micro-expression data ,The micro-expression data includes the timestamp of the expression and the ,corresponding micro-expression features;

[0038] S45. Generate biometric data of the interrogation subject by integrating identity information, expression data and micro-expression data obtained by face recognition .

[0039] Optionally, S5 includes the following specific contents:

[0040] S51, based on structured case record data Extract the key statement node set in the case description , for each semantic unit in the structured case record data Calculate the node importance and include the semantic unit in the key statement node set when the following conditions are met:

[0041] ;

[0042] in, It is a node importance calculation function used to measure the semantic unit Description of the case , Time clues With key figures The importance of and is the weighting coefficient, is a function that measures the relevance of semantic units to case descriptions and time clues. is a function that measures the weight of key character features in a semantic unit. is the threshold, m' is the total number of semantic units;

[0043] S52. Biometric data of the interrogation subject A collection of nodes with key statements Input multimodal feature association model, for each key statement node Calculate the correlation between biometric data and , the relevance is calculated by time-weighted multidimensional similarity integral:

[0044] ;

[0045] in, and Key statement node The corresponding time range, is the emotional information in biometric data, is the micro-expression feature in the biometric data, and is the weighting coefficient, is the temporal correlation function between emotional information and node semantic features, is the temporal correlation function between micro-expression features and node semantic features;

[0046] S53. From the key statement node set Select the most relevant key statement node set Map biometric data to corresponding nodes, mark the biometric information in structured case transcript data and generate comprehensive case transcript data :

[0047] ;

[0048] ;

[0049] in, is a comprehensive scoring function used to convert the relevance With structured case record data Midpoint The semantic match of Combined, As a fusion function, by considering the degree of fit between the biometric data and the case description, the key statement nodes that best reflect the relationship between the interrogation subject's biometrics and the case logic are selected to generate comprehensive case transcript data containing semantic information and emotional state. .

[0050] Optionally, S6 includes the following specific contents:

[0051] S61. Comprehensive case record data Transformed into a set of case semantic feature vectors ,in represents the pth semantic feature vector in the comprehensive case record data, p is the total number of semantic feature vectors in the comprehensive case record data, and at the same time, the case record H in the historical case database is converted into a set of historical case semantic feature vectors ,in represents the semantic feature vector of historical cases, q is the total number of historical case records;

[0052] S62. Case semantic feature vector set based on comprehensive case record data and the historical case semantic feature vector set of the historical case database , construct contextual semantic relationship matrix , the context semantic relationship matrix is ​​used to represent the logical relationship between case semantic feature vectors:

[0053] ;

[0054] S63. Use the whale optimization algorithm to optimize the case comparison path and define the optimization objective function is the weighted sum of the semantic similarity scores between cases and the contextual logical relationship:

[0055] ;

[0056] in, The elements of the weight matrix represent the semantic feature vector and The matching weight of is the similarity between semantic feature vectors, calculated by cosine similarity;

[0057] The whale optimization algorithm performs the following steps on the weight matrix: To optimize:

[0058] Initialize the weight matrix population W, the population size is N, each weight matrix for ;

[0059] For each weight matrix Calculate the objective function value ;

[0060] Update the weight matrix through the prey encirclement mechanism:

[0061] ;

[0062] in, is the optimal weight matrix in the current population, , , and is a random number, and a gradually decreases with the number of iterations;

[0063] The weight matrix is ​​further optimized through the spiral position update mechanism:

[0064] ;

[0065] Where b is the spiral shape control parameter, is a random number;

[0066] Keep the weight matrix with the highest objective function value as the optimal weight matrix ;

[0067] S64, based on the optimized weight matrix and contextual semantic relationship matrix Select the top k historical case record sets with the highest sum of similarities with the case semantic feature vector set from the historical case semantic feature vector set :

[0068] ;

[0069] in, is the element of the optimized weight matrix;

[0070] A collection of selected historical case records Include cases that have high relevance to the current case in terms of semantic content and contextual logic.

[0071] Optionally, S7 includes the following specific contents:

[0072] S71. Semantic feature vectors for historical cases Perform feature decomposition and extract the corresponding clue feature vector set , the key clue set Includes descriptions of events, timelines, and information about key characters related to historical cases;

[0073] S72, generate an evidence chain for the clue feature vector set c extracted from the historical case record set, and construct an evidence chain E by establishing a causal relationship between the features:

[0074] ;

[0075] in, Clue Features and The causal strength function is is the causal strength threshold;

[0076] S73. Map each association relationship in the evidence chain E to the comprehensive case record data The key statement node set , calculate the semantic consistency of comprehensive case record data and evidence chain , and pattern matching of key statements in comprehensive case transcript data to identify association patterns between current cases and historical cases :

[0077] ;

[0078] in, is the semantic consistency threshold, Represents the key statement node of the comprehensive case record data Characteristics of clues related to historical cases and Matching mode;

[0079] S74. Based on the matched association pattern and evidence chain E, generating case comparison result data related to the current case ,The case comparison result data includes the matched historical case record set, the key clue set C of the historical cases, the causal relationship of the chain of evidence E, and the matching pattern of the ,comprehensive case transcript data of the current case and the corresponding semantic consistency score.

[0080] A case interrogation intelligent assistance device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the case interrogation intelligent assistance method.

[0081] The beneficial effects of the present invention are:

[0082] (1) The present invention integrates speech recognition, computer vision and natural language processing technologies to uniformly process the audio, video and manually recorded texts collected during the interrogation process. Through the semantic analysis and logical reasoning functions of the large language model, structured case records are generated. In addition, the key statement nodes of the case records are deeply annotated in combination with biometric data (expression, emotion and micro-expression features), thus achieving real-time fusion of multimodal data, greatly improving data processing efficiency, ensuring the integrity and logic of information, and providing accurate and efficient data support for the interrogation process.

[0083] (2) The present invention adopts the whale optimization algorithm to dynamically optimize the case matching path, and establishes semantic associations and logical relationships between cases through a comprehensive analysis of the semantic features of case records and the semantic features of historical case databases. At the same time, it can effectively overcome the shortcomings of traditional keyword matching-based retrieval methods in terms of semantic understanding depth and efficiency. In a large-scale case database, it can quickly and accurately match historical case records that are deeply related to the current case, thereby providing strong technical support for case analysis.

[0084] (3) The present invention conducts in-depth analysis on the matched historical case records, extracts key clues, and uses logical relationships to establish a chain of evidence. It further combines the key statement nodes of the comprehensive record of the current case to generate a case model. It can not only quickly locate the core evidence in the historical case, but also reveal the complex causal relationship and logical chain between cases, providing all-round reference support for the trial of the current case. Compared with the existing technology, the chain of evidence and pattern extraction method of the present invention are more logical and accurate, avoiding the problems of information omission and one-sided analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0086] Figure 1 A flow chart of an intelligent case interrogation assistance device and method proposed by the present invention;

[0087] Figure 2 This is a schematic diagram of the flow of a whale optimization algorithm for optimizing case comparison paths in an intelligent case interrogation auxiliary device and method proposed by the present invention. DETAILED DESCRIPTION

[0088] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0089] refer to Figure 1-Figure 2 , a case interrogation intelligent assistance method, comprising the following steps:

[0090] S1. Collect audio and video data and on-site manual recording text during the case interrogation process;

[0091] S2, performing speech recognition processing on the collected recording data to generate synchronous interrogation text data;

[0092] S3, inputting the synchronous interrogation text data into the large language model to generate structured case transcript data;

[0093] S4, analyzing the collected video data in real time to generate biometric data related to the interrogation subject;

[0094] S5, fusing the generated biometric data with the structured case record data to generate comprehensive case record data containing semantic information and emotional state;

[0095] S6. Input the comprehensive case record data into the case comparison module, use the whale optimization algorithm to optimize the case comparison path, and identify the historical case records related to the current case;

[0096] S7. Analyze the matched historical case records, extract key clues, evidence chains and case patterns in the historical case records, and combine them with key statements in the comprehensive case transcript data to generate case comparison result data related to the current case;

[0097] S8. Generate intelligent interrogation prompt data based on comprehensive case record data, biometric data, case comparison result data and related information in historical case records.

[0098] In this implementation, S1 includes the following specific contents:

[0099] S11. Obtaining audio data generated during the interrogation process through audio acquisition equipment , using speech enhancement technology to remove recorded data background noise and interference sound to generate the final recording data ;

[0100] S12, obtaining video data V(x, y, t) during the interrogation process through a video acquisition device, where x and y are spatial coordinates in the video frame, and t is time, and smoothing multiple frames using motion compensation technology to generate video data V'(x, y, t) with optimized frame rate;

[0101] S13. Obtaining on-site manual recording text through auxiliary input devices .

[0102] In this implementation, S2 includes the following specific contents:

[0103] S21, record the data Input the speech recognition model based on the end-to-end speech processing algorithm to Converted into preliminary text data ;

[0104] S22, the preliminary text data Manually record text on site Alignment is performed by timestamp, by calculating the temporal matching degree of the text paragraphs Generate aligned text data :

[0105] ;

[0106] Where n is the number of paragraphs in the text data. Represents the timestamp matching function for each piece of text;

[0107] S23, splicing the aligned text data, removing redundancy and formatting, and generating synchronous interrogation text data .

[0108] In this implementation, S3 includes the following specific contents:

[0109] S31, will synchronize the interrogation text data Input the large language model, which performs semantic analysis on the synchronous interrogation text data based on natural language processing technology to extract case-related semantic features ;

[0110] S32. Using large language models to analyze semantic features Perform logical reasoning, identify the logical relationships between statements in the synchronous interrogation text data, and construct a logical association matrix :

[0111] ;

[0112] Among them, i, j represent the index of semantic units;

[0113] S33, based on the logical association matrix and semantic features , organize the case description content through the context optimization model to generate structured case record data , including case description , Time clues , key figures and the orderly arrangement of statement logic.

[0114] In this implementation, S4 includes the following specific contents:

[0115] S41, input the video data V'(x, y, t) into the face recognition model to extract the facial features in the video data ;

[0116] S42, using the face recognition model to compare the face features with the known identity data in the historical database Match the identity information of the interrogation object :

[0117] ;

[0118] in, is the similarity calculation function;

[0119] S43, perform facial expression analysis on the facial features in the video data V'(x, y, t) to extract the facial expression data of the interrogated subject , including emotion categories and emotional intensity ;

[0120] S44, detecting expression changes in the video data V'(x, y, t) to generate micro-expression data ,The micro-expression data includes the timestamp of the expression and the ,corresponding micro-expression features;

[0121] S45. Generate biometric data of the interrogation subject by integrating identity information, expression data and micro-expression data obtained by face recognition .

[0122] In this implementation, S5 includes the following specific contents:

[0123] S51, based on structured case record data Extract the key statement node set in the case description , for each semantic unit in the structured case record data Calculate the node importance and include the semantic unit in the key statement node set when the following conditions are met:

[0124] ;

[0125] in, It is a node importance calculation function used to measure the semantic unit Description of the case , Time clues With key figures The importance of and is the weighting coefficient, is a function that measures the relevance of semantic units to case descriptions and time clues. is a function that measures the weight of key character features in a semantic unit. is the threshold, m' is the total number of semantic units;

[0126] S52. Biometric data of the interrogation subject A collection of nodes with key statements Input multimodal feature association model, for each key statement node Calculate the correlation between biometric data and , the relevance is calculated by time-weighted multidimensional similarity integral:

[0127] ;

[0128] in, and Key statement node The corresponding time range, is the emotional information in biometric data, is the micro-expression feature in the biometric data, and is the weighting coefficient, is the temporal correlation function between emotional information and node semantic features, is the temporal correlation function between micro-expression features and node semantic features;

[0129] S53. From the key statement node set Select the most relevant key statement node set Map biometric data to corresponding nodes, mark the biometric information in structured case transcript data and generate comprehensive case transcript data :

[0130] ;

[0131] ;

[0132] in, is a comprehensive scoring function used to convert the relevance With structured case record data Midpoint The semantic match of Combined, As a fusion function, by considering the degree of fit between the biometric data and the case description, the key statement nodes that best reflect the relationship between the interrogation subject's biometrics and the case logic are selected to generate comprehensive case transcript data containing semantic information and emotional state. .

[0133] In this implementation, S6 includes the following specific contents:

[0134] S61. Comprehensive case record data Transformed into a set of case semantic feature vectors ,in represents the pth semantic feature vector in the comprehensive case record data, p is the total number of semantic feature vectors in the comprehensive case record data, and at the same time, the case record H in the historical case database is converted into a set of historical case semantic feature vectors ,in represents the semantic feature vector of historical cases, q is the total number of historical case records;

[0135] S62. Case semantic feature vector set based on comprehensive case record data and the historical case semantic feature vector set of the historical case database , construct contextual semantic relationship matrix , the context semantic relationship matrix is ​​used to represent the logical relationship between case semantic feature vectors:

[0136] ;

[0137] S63. Use the whale optimization algorithm to optimize the case comparison path and define the optimization objective function is the weighted sum of the semantic similarity scores between cases and the contextual logical relationship:

[0138] ;

[0139] in, The elements of the weight matrix represent the semantic feature vector and The matching weight of is the similarity between semantic feature vectors, calculated by cosine similarity;

[0140] The whale optimization algorithm performs the following steps on the weight matrix: To optimize:

[0141] Initialize the weight matrix population W, the population size is N, each weight matrix for ;

[0142] For each weight matrix Calculate the objective function value ;

[0143] Update the weight matrix through the prey encirclement mechanism:

[0144] ;

[0145] in, is the optimal weight matrix in the current population, , , and is a random number, and a gradually decreases with the number of iterations;

[0146] The weight matrix is ​​further optimized through the spiral position update mechanism:

[0147] ;

[0148] Where b is the spiral shape control parameter, is a random number;

[0149] Keep the weight matrix with the highest objective function value as the optimal weight matrix ;

[0150] S64, based on the optimized weight matrix and contextual semantic relationship matrix Select the top k historical case record sets with the highest sum of similarities with the case semantic feature vector set from the historical case semantic feature vector set :

[0151] ;

[0152] in, is the element of the optimized weight matrix;

[0153] A collection of selected historical case records Include cases that have high relevance to the current case in terms of semantic content and contextual logic.

[0154] In this implementation, S7 includes the following specific contents:

[0155] S71. Semantic feature vectors for historical cases Perform feature decomposition and extract the corresponding clue feature vector set , the key clue set Includes descriptions of events, timelines, and information about key characters related to historical cases;

[0156] S72, generate an evidence chain for the clue feature vector set c extracted from the historical case record set, and construct an evidence chain E by establishing a causal relationship between the features:

[0157] ;

[0158] in, Clue Features and The causal strength function is is the causal strength threshold;

[0159] S73. Map each association relationship in the evidence chain E to the comprehensive case record data The key statement node set , calculate the semantic consistency of comprehensive case record data and evidence chain , and pattern matching of key statements in comprehensive case transcript data to identify association patterns between current cases and historical cases :

[0160] ;

[0161] in, is the semantic consistency threshold, Represents the key statement node of the comprehensive case record data Characteristics of clues from historical cases and Matching mode;

[0162] S74. Based on the matched association pattern and evidence chain E, generating case comparison result data related to the current case ,The case comparison result data includes the matched historical case record set, the key clue set C of the historical cases, the causal relationship of the chain of evidence E, and the matching pattern of the ,comprehensive case transcript data of the current case and the corresponding semantic consistency score.

[0163] A case interrogation intelligent assistance device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes a case interrogation intelligent assistance method.

[0164] The present invention integrates speech recognition, computer vision and natural language processing technologies to uniformly process the audio, video and manually recorded texts collected during the interrogation process, generates structured case records through the semantic analysis and logical reasoning functions of the large language model, and deeply annotates the key statement nodes of the case records in combination with biometric data (expression, emotion and micro-expression features), realizing the real-time fusion of multimodal data, greatly improving data processing efficiency, ensuring the integrity and logic of information, and providing accurate and efficient data support for the interrogation process.

[0165] The present invention adopts the whale optimization algorithm to dynamically optimize the case matching path, and establishes semantic associations and logical relationships between cases through a comprehensive analysis of the semantic features of case records and the semantic features of historical case databases. At the same time, it can effectively overcome the shortcomings of traditional keyword matching-based retrieval methods in terms of semantic understanding depth and efficiency, and can quickly and accurately match historical case records that are deeply related to the current case in a large-scale case database, thereby providing strong technical support for case analysis.

[0166] The present invention conducts in-depth analysis on the matched historical case records, extracts key clues, and uses logical relationships to establish a chain of evidence. It further combines the key statement nodes of the current case comprehensive record to generate a case model. It can not only quickly locate the core evidence in historical cases, but also reveal the complex causal relationships and logical chains between cases, providing all-round reference support for the trial of the current case. Compared with the existing technology, the chain of evidence and pattern extraction method of the present invention are more logical and accurate, avoiding the problems of information omission and one-sided analysis.

[0167] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A case interrogation intelligent assistance method, characterized in that: The steps include: S1. Collect audio and video data and on-site manual recording text during the case interrogation process; S2, performing speech recognition processing on the collected recording data to generate synchronous interrogation text data; S3, inputting the synchronous interrogation text data into the large language model to generate structured case transcript data; S4, analyzing the collected video data in real time to generate biometric data related to the interrogation subject; S5, fusing the generated biometric data with the structured case record data to generate comprehensive case record data containing semantic information and emotional state; S6. Input the comprehensive case record data into the case comparison module, use the whale optimization algorithm to optimize the case comparison path, and identify the historical case records related to the current case; S7. Analyze the matched historical case records, extract key clues, evidence chains and case patterns in the historical case records, and combine them with key statements in the comprehensive case transcript data to generate case comparison result data related to the current case; S8, generating intelligent interrogation prompt data based on comprehensive case record data, biometric data, case comparison result data and related information in historical case records; The S2 includes the following specific contents: S21, recording data D' a (t) Input the speech recognition model based on the end-to-end speech processing algorithm and convert the recorded data D′ a (t) Converted into preliminary text data T a ; S22, the preliminary text data T a Manually record text on site r Alignment is performed by timestamp, and the time matching degree M(T a ,T r ) Generate aligned text data T sync : Where n is the number of paragraphs in the text data, and δ represents the timestamp matching function for each paragraph of text; S23, splicing the aligned text data, removing redundancy and formatting, and generating synchronous interrogation text data T final ; The S3 includes the following specific contents: S31, synchronize the interrogation text data T final Input the large language model, which performs semantic analysis on the synchronous interrogation text data based on natural language processing technology and extracts case-related semantic features F semantic ; S32, using a large language model to semantically represent the feature F semantic Perform logical reasoning, identify the logical relationship between the statements in the synchronous interrogation text data, and construct the logical association matrix L matrix : Among them, i, j represent the index of semantic units; S33, based on the logical association matrix L matrix and semantic features F semantic , organize the case description content through the context optimization model to generate structured case record data T structured , including case description P desc , time clue T timeline 、Key Person K keypoints and the orderly arrangement of statement logic.

2. The intelligent case interrogation assistance method according to claim 1 is characterized in that: The S1 includes the following specific contents: S11. Obtain the audio data generated during the interrogation process through audio acquisition equipment D a (t), use speech enhancement technology to remove the recording data D a (t) background noise and interference sound, generate the final recording data D′ a (t); S12, obtaining video data V(x, y, t) during the interrogation process through a video acquisition device, where x and y are spatial coordinates in the video frame, and t is time, and smoothing multiple frames using motion compensation technology to generate video data V′(x, y, t) with optimized frame rate; S13, obtaining the on-site manual recording text T through the auxiliary input device r .

3. The intelligent case interrogation assistance method according to claim 2 is characterized in that: The S4 includes the following specific contents: S41, input the video data V′(x, y, t) into the face recognition model, and extract the face features F in the video data face ; S42, using the face recognition model to compare the face features with the known identity data ID in the historical database db Match the identity information ID of the interrogation object match : Among them, sim is the similarity calculation function; S43, perform facial expression analysis on the facial features in the video data V′(x, y, t) to extract the facial expression data E of the interrogation subject emotion , including the emotion category E class and emotional intensity E intensity ; S44, detecting the expression changes in the video data V′(x, y, t) and generating micro-expression data M micro ,The micro-expression data includes the timestamp of the expression and the ,corresponding micro-expression features; S45, generate the biometric data B of the interrogation object by integrating the identity information, expression data and micro-expression data obtained by face recognition bio .

4. The intelligent case interrogation assistance method according to claim 3 is characterized in that: The S5 includes the following specific contents: S51. Based on structured case record data T structured Extract the key statement node set N in the case description nodes , for each semantic unit S in the structured case record data k Calculate the node importance and include the semantic unit in the key statement node set when the following conditions are met: Ξ(K keypoints (S k ))>θ}; Among them, Ω(S k ) is the node importance calculation function, which is used to measure the semantic unit S k In the case description P desc (S k ), time clue T timeline (S k ) and key figure T timeline (S k ), γ1 and γ2 are weighting coefficients, Ψ is a function that measures the relevance of semantic units to case descriptions and time clues, Ξ is a function that measures the weight of key character features in semantic units, θ is a threshold, and m' is the total number of semantic units; S52, the interrogation subject's biometric data B bio With key statement node set N nodes Input multimodal feature association model, for each key statement node n i Calculate the correlation between the biometric data and its corr(B bio ,n i ), the relevance is calculated by time-weighted multidimensional similarity integral: Among them, τ start (n i ) and τ end (n i ) is the key statement node n i The corresponding time range, E emotion is the facial expression data, M micro is the micro-expression feature in the biometric data, α and β are weighting coefficients, λ(E emotion ,n i ,t) is the temporal correlation function between facial expression data and node semantic features, ρ(M micro ,n i ,t) is the temporal correlation function between micro-expression features and node semantic features; S53, from the key statement node set N nodes Select the most relevant key statement node set N selected Map the biometric data to the corresponding nodes, mark the biometric information in the structured case record data and generate comprehensive case record data T comprehensive : D(B bio ,n i )=χ(corr(B bio ,n i ),η(T structured ,n i )); Among them, Δ(B bio ,n i ) is a comprehensive scoring function used to convert the correlation corr(B bio ,n i ) and structured case record data T structured Middle Node n i The semantic matching degree η(T structured ,n i ), χ is the fusion function, and by considering the degree of fit between the biometric data and the case description, the key statement nodes that best reflect the relationship between the interrogation subject’s biometrics and the case logic are selected to generate comprehensive case transcript data T containing semantic information and emotional state. comprehensive .

5. The intelligent case interrogation assistance method according to claim 4 is characterized in that: The S6 includes the following specific contents: S61, the comprehensive case record data T comprehensive Transformed into case semantic feature vector set F case1 ={f1,f2,…,f p },1≤i≤p, where f i represents the i-th semantic feature vector in the comprehensive case record data, p is the total number of semantic feature vectors in the comprehensive case record data, and at the same time, the case record H in the historical case database is converted into a set of historical case semantic feature vectors F history ={h1,h2,…,h q },1≤j≤q, where h j represents the semantic feature vector of historical cases, q is the total number of historical case records; S62. Case semantic feature vector set F based on comprehensive case record data case1 and the historical case semantic feature vector set F of the historical case database history , construct the contextual semantic relationship matrix L matrix , the context semantic relationship matrix is ​​used to represent the logical relationship between case semantic feature vectors: S63. Use the whale optimization algorithm to optimize the case comparison path, and define the optimization objective function Φ as the weighted sum of the semantic similarity scores between cases and the contextual logical relationship: Among them, w ij The elements of the weight matrix represent the semantic feature vector f i and h j The matching weight, sim(f i ,h j ) is the similarity between semantic feature vectors, calculated by cosine similarity; The whale optimization algorithm performs the following steps to optimize the weight matrix w ij To optimize: Initialize the weight matrix population W, the population size is N, each weight matrix W (n) for For each weight matrix W (n) Calculate the objective function value Φ (n) ; Update the weight matrix through the prey encirclement mechanism: W (n) (t+1)=W * (t)-A·|C·W * (t)-W (n) (t)|; Among them, W * (t) is the optimal weight matrix in the current population, A = 2a·r1-a, C = 2·r2, r1 and r2 are random numbers, and a gradually decreases with the number of iterations; The weight matrix is ​​further optimized through the spiral position update mechanism: W (n) (t+1)=e bl ·cos(2πl)·(W * (t)-W (n) (t))+W * (t); Among them, b is the spiral shape control parameter, l∈[-1,1] is a random number; Keep the weight matrix with the highest objective function value as the optimal weight matrix W * ; S64, based on the optimized weight matrix W * and contextual semantic relationship matrix L matrix Select the first k historical case record sets H with the highest sum of similarities with the case semantic feature vector set from the historical case semantic feature vector set selected : in, is the element of the optimized weight matrix; The selected historical case record set H selected Include cases that have high relevance to the current case in terms of semantic content and contextual logic.

6. The intelligent case interrogation assistance method according to claim 5 is characterized in that: The S7 includes the following specific contents: S71. Semantic feature vector h of historical cases j Perform feature decomposition and extract the corresponding clue feature vector set C j , the clue feature vector set C j Includes descriptions of events, timelines, and information about key characters related to historical cases; S72. Clue feature vector set C extracted from the historical case record set j Generate evidence chain and build evidence chain E by establishing causal relationship between features: Among them, link(c a ,c b ) is the clue feature c a and c b The causal strength function, θ e is the causal strength threshold; S73, map each association relationship in the evidence chain E to the comprehensive case record data T comprehensive The key statement node set N nodes , calculate the semantic consistency sim(n i ,c a ), and pattern matching is performed on key statements in the comprehensive case transcript data to identify the association patterns between the current case and historical cases. pattern : M pattern ={(n i ,c a ,c b )∣sim(n i ,c a )>θ m ,link(c a ,c b )>θ e }; Among them, θ m is the semantic consistency threshold, (n i ,c a ,c b ) represents the key statement node n of the comprehensive case record data i c a and c b Matching mode; S74, based on the matched association mode M pattern and evidence chain E, generate case comparison result data R related to the current case compare The case comparison result data includes the matched historical case record set and clue feature vector set C j , the causal relationship of the chain of evidence E and the matching pattern of the comprehensive case transcript data of the current case and the corresponding semantic consistency score.

7. An intelligent auxiliary device for case interrogation, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the intelligent case interrogation assistance method described in any one of claims 1-6 is implemented.

Images