A Visual Evidence Method and System

Abstract

This invention discloses a visual evidence method and system, relating to the field of visualization technology. The key technical points include the following steps: collecting transcript text and associated text and background text; processing the transcript text to obtain a transcript text feature set and an associated paragraph set; segmenting the associated text into paragraphs based on semantic features to obtain an associated text feature set; segmenting the transcript text into paragraphs based on background text to obtain a background text feature set; comparing and analyzing the associated text feature set and the associated paragraph set to obtain an intent association result set; determining whether the transcript text feature set and the background text feature set are thematically related, and then processing and analyzing them to obtain a first intent factor set and a second intent factor set respectively; the effect is that by comparing and analyzing different text feature sets and paragraph sets, potential evidence associations can be discovered.

Description

Technical Field

[0001] This invention relates to the field of visualization technology, and more specifically, to a visualization method and system. Background Technology

[0002] In relevant fields, written records serve as crucial information carriers, making accurate identification of intent and effective presentation of evidence paramount. Current technologies for processing records and related information have limitations. Traditional methods of presenting evidence are often simplistic, consisting mainly of static text listings or simple evidence compilations, lacking intuitiveness and coherence. The presentation of large amounts of text and scattered evidence during the examination process makes it difficult to quickly grasp key information and the logical relationships between pieces of evidence. Furthermore, the lack of visualization tools hinders the deep integration and presentation of recorded content and evidentiary materials, failing to clearly present the full picture of the case and the chain of evidence, thus hindering comprehension of the facts. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a visual evidence method and system.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] A visual evidence method, comprising the following steps:

[0006] Collect transcript text as well as associated text and background text related to the transcript text;

[0007] The transcript text is processed to obtain a transcript text feature set and a set of associated paragraphs. The associated text is segmented into paragraphs based on semantic features to obtain an associated text feature set. The transcript text is segmented into paragraphs based on background text to obtain a background text feature set.

[0008] The intention association result set is obtained by comparing and analyzing the associated text feature set and the associated paragraph set;

[0009] After processing and analyzing the feature set of the transcript text and the feature set of the background text to determine whether they are related to the topic, the first intention factor set and the second intention factor set are obtained respectively.

[0010] After inputting the first intent factor set and the second intent factor set into the intent recognition prediction model, the intent recognition prediction result is obtained. The intent recognition prediction result is combined with the intent association result set to identify the intent of the transcript text and then display the visual evidence content.

[0011] Preferably, after determining whether the feature set of the transcript text and the feature set of the background text are related to the topic, the first intention factor set and the second intention factor set are obtained respectively. The specific steps include:

[0012] When the feature set of the transcript text and the feature set of the background text are subject-related, the feature set of the transcript text and the feature set of the background text are compared and analyzed to obtain the first intention factor set;

[0013] When the feature set of the transcript text and the feature set of the background text are subject-irrelevant, the feature set of the related text and the feature set of the background text are compared and analyzed to obtain the second intention factor set.

[0014] Preferably, the transcript text is processed to obtain a transcript text feature set and a set of related paragraphs. The related text is then segmented into paragraphs based on semantic features to obtain a set of related text features. The transcript text is then segmented into paragraphs based on background text to obtain a set of background text features. Specifically, this includes the following steps:

[0015] Extract the semantic distribution features of the transcript text, and then divide the transcript text into several paragraphs based on the semantic distribution features to obtain the transcript paragraph set;

[0016] The transcript paragraph set and the background text are divided into several paragraphs according to their own semantic features to obtain the related paragraph set and the background paragraph set;

[0017] After determining the tags of the transcript paragraph set and the associated text based on the semantic importance, the transcript text feature set and the associated text feature set are obtained respectively;

[0018] After determining the core intent paragraphs based on their semantic importance, the background text feature set is obtained.

[0019] Preferably, the intention association result set is obtained by comparing and analyzing the associated text feature set and the associated paragraph set, specifically including the following steps:

[0020] A dataset of semantic keyword frequencies for each paragraph in a statistically related text feature set;

[0021] The first semantic overlap frequency dataset is obtained by comparing the semantic keyword overlap frequency of paragraphs in the associated paragraph set and the associated text feature set.

[0022] The semantic proportion dataset is obtained by calculating the proportion values ​​of the first semantic overlap frequency dataset and the semantic keyword frequency dataset.

[0023] When there are values ​​in the semantic proportion dataset that are greater than the preset semantic threshold, the difference exceeding the preset semantic threshold is calculated to obtain the semantic excess dataset.

[0024] After extracting the first intent information from the associated paragraph set, a scene context dataset is set up.

[0025] The intent association result set is obtained by inputting the first intent information, the scene context dataset, the associated text feature set, and the semantic excess dataset into the intent association value prediction model.

[0026] Preferably, when the feature set of the transcript text and the feature set of the background text are topic-related, the feature set of the transcript text and the feature set of the background text are compared and analyzed to obtain the first intention factor set, which specifically includes the following steps:

[0027] When the feature set of the transcript text and the feature set of the background text are topic-related, the feature set of the transcript text and the feature set of the background text are processed to obtain the second semantic overlap frequency dataset and semantic sentiment value;

[0028] The overlapping topic dataset is obtained by combining the statistical record text feature set with the background text feature set for each paragraph with the corresponding overlapping topics.

[0029] The second intent information of the background text feature set is extracted. The second semantic overlap frequency dataset, semantic sentiment value, overlapping topic dataset, second intent information, scene context dataset, intent association result set and transcript text feature set are combined to form the first intent factor set.

[0030] Preferably, when the feature set of the transcript text and the feature set of the background text are topic-related, the feature set of the transcript text and the feature set of the background text are processed to obtain a second semantic overlap frequency dataset and semantic sentiment value, specifically including the following steps:

[0031] When the feature set of the transcript text and the feature set of the background text are subject-related, the second semantic overlap frequency dataset is obtained by comparing and averaging the semantic keyword overlap frequency of each paragraph in the feature set of the transcript text and the background text.

[0032] Extract the semantic sentiment values ​​of the corresponding paragraphs from the second semantic overlap frequency dataset.

[0033] Preferably, when the feature set of the transcript text and the feature set of the background text are subject-irrelevant, the feature set of the related text is compared and analyzed with the feature set of the background text to obtain the second intention factor set, which specifically includes the following steps:

[0034] When the feature set of the transcript text and the feature set of the background text are subject-irrelevant, the subject information dataset is obtained by statistically analyzing the themes of the corresponding paragraphs in the feature set of the transcript text and the feature set of the background text.

[0035] The second semantic overlap frequency dataset, semantic sentiment tendency value, overlapping topic dataset, second intent information, scene context dataset, transcript text feature set, intent association result set and topic information dataset are combined to form the second intent factor set.

[0036] A visual evidence system, comprising:

[0037] Data Acquisition Module: Acquires transcript text, as well as associated text and background text.

[0038] Segmentation module: Processes the transcript text to obtain a transcript text feature set and a set of associated paragraphs; performs paragraph segmentation on the associated text based on semantic features to obtain a set of associated text features; and performs paragraph segmentation on the transcript text based on background text to obtain a set of background text features.

[0039] The first analysis module compares and analyzes the associated text feature set with the associated paragraph set to obtain the intention association result set;

[0040] The second analysis module: After processing and analyzing the feature set of the transcript text and the feature set of the background text under the condition of whether the topic is related, the first intention factor set and the second intention factor set are obtained respectively.

[0041] Display module: After inputting the first intent factor set and the second intent factor set into the intent recognition prediction model, the intent recognition prediction result is obtained. The intent recognition prediction result is combined with the intent association result set to identify the intent of the transcript text and then the visualized evidence content is displayed.

[0042] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a visual verification method.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] This invention mines the semantic connections between related text feature sets and related paragraph sets through operations such as statistical analysis of semantic keyword frequencies and calculation of semantic overlap frequencies. In cases of drunk driving, it accurately determines the degree of consistency between the party's statement and related texts such as laws, regulations, and relevant cases. For example, it precisely measures the consistency between the party's understanding and expression of key concepts such as "drunk driving" and "alcohol content standard," providing strong evidence for accurately determining whether the party knowingly violated the law and reducing misjudgments caused by semantic ambiguity. When generating the first and second intent factor sets, it integrates multi-dimensional information such as semantic overlap frequency, semantic sentiment value, overlapping topic data, intent information, and scene context.

[0045] From extracting semantic distribution features from transcript texts and segmenting paragraphs to comparing and analyzing related texts and paragraphs, a series of operations have been performed, achieving automation and standardization of text processing. It can quickly complete the screening, classification, and feature extraction of large amounts of text information. For example, when processing transcripts of multiple drunk driving cases, it can rapidly generate text feature sets and paragraph sets, significantly shortening the initial text processing time and improving case handling efficiency.

[0046] By tagging text according to its semantic importance, key information can be quickly located in transcripts, related texts, and background texts. In drunk driving cases, core information such as "amount of alcohol consumed," "driving time," and "blood alcohol test result" can be quickly identified, avoiding blind searching through large amounts of text. This allows investigators to focus their efforts on analyzing key information and accelerates the case analysis process.

[0047] This technical solution comprehensively covers transcripts, related texts (laws and regulations, similar cases, etc.), and background texts (law enforcement records, environmental information, etc.), examining cases from multiple perspectives. By comparing and analyzing different text feature sets and paragraph sets, potential evidence connections can be discovered. Attached Figure Description

[0048] Figure 1 This invention provides a schematic diagram illustrating the steps of a visual verification method.

[0049] Figure 2 This invention provides a schematic diagram of the modules of a visual evidence system.

[0050] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0051] 610. Processor; 620. Communication interface; 630. Memory; 640. Communication bus. Detailed Implementation

[0052] Reference Figures 1 to 3 .

[0053] Example 1 further illustrates a visualization method for evidence presented in this invention.

[0054] A visual evidence method, comprising the following steps:

[0055] Collect transcript text as well as associated text and background text related to the transcript text;

[0056] The transcript text is processed to obtain a transcript text feature set and a set of associated paragraphs. The associated text is segmented into paragraphs based on semantic features to obtain an associated text feature set. The transcript text is segmented into paragraphs based on background text to obtain a background text feature set.

[0057] The intention association result set is obtained by comparing and analyzing the associated text feature set and the associated paragraph set;

[0058] After processing and analyzing the feature set of the transcript text and the feature set of the background text to determine whether they are related to the topic, the first intention factor set and the second intention factor set are obtained respectively.

[0059] After inputting the first intent factor set and the second intent factor set into the intent recognition prediction model, the intent recognition prediction result is obtained. The intent recognition prediction result is combined with the intent association result set to identify the intent of the transcript text and then display the visual evidence content.

[0060] This application first collects transcript texts (such as statements), related texts (such as templates for similar cases, legal provisions), and background texts (such as case-related environmental information and historical processing records). For example, in a dangerous driving case, the transcript text contains a description of the incident, related texts may include relevant clauses of the Road Traffic Safety Law, and background texts involve the time and location information of the law enforcement recorder video.

[0061] The transcript text is processed to extract core semantic features (such as time, location, people, and action verbs), and related paragraph sets (such as paragraphs describing the incident and alcohol test results) are segmented based on semantic relevance. Related text is then segmented into related text feature sets (such as sentencing standards paragraphs in legal clauses) based on semantic features. Background text (such as video recordings) is parsed to extract timestamps, scene descriptions, and other information to generate background text feature sets (such as the recording time and location of law enforcement recorder videos).

[0062] The semantic comparison is performed between the set of associated text features (such as legal clauses) and the set of associated paragraphs in the transcript (such as paragraphs where the party mentions "driving under the influence of alcohol") to calculate the keyword overlap frequency (such as "alcohol content," "driving route," etc.) and output the intention association result set (such as judging the strength of the relevance between the party's statement and the legal clauses). The text is vectorized using RoBERTa or BERT models, and the semantic matching degree is calculated using cosine similarity.

[0063] Analyze whether the feature set of the written record (such as the core elements of the case) and the feature set of the background text (such as the scene recorded in the video) belong to the same theme (such as "whether the time and place of the incident are consistent"). If the theme is related (such as the intersection mentioned in the written record being consistent with the location of the video shooting), then count the semantic keyword overlap frequency (such as "Dagui intersection" and "alcohol test value 110mg / 100ml"), sentiment tendency value (such as the degree of cooperation in the parties' statements), and length of overlapping themes (such as the length of the paragraph describing the incident in common). Combined with historical intent information (such as the handling pattern of similar cases), generate the first intent factor set.

[0064] If the topics are unrelated (e.g., the time mentioned in the transcript contradicts the time in the video), then the topic distance between the two (e.g., time difference, location deviation) is calculated and combined with other semantic features to generate a second set of intent factors.

[0065] The first and second intent factor sets are input into the intent recognition prediction model fine-tuned by the GLM4-9b large model, and the intent recognition result is output by combining the intent association result set (such as "dangerous driving crime established" or "contradictory statement").

[0066] Based on the intent recognition results, associated visualization resources (such as body camera video clips, evidence photos, and legal clause texts) are invoked and displayed in a structured manner according to preset rules (such as timeline and evidence chain logic). When the party concerned recounts the events of the incident, video clips from the body camera at the corresponding time points are simultaneously displayed, along with annotations of key information such as alcohol test values ​​and driving routes.

[0067] After determining whether the feature set of the transcript text and the feature set of the background text are related to the topic, the first intention factor set and the second intention factor set are obtained through processing and analysis. The specific steps include:

[0068] When the feature set of the transcript text and the feature set of the background text are subject-related, the feature set of the transcript text and the feature set of the background text are compared and analyzed to obtain the first intention factor set;

[0069] When the feature set of the transcript text and the feature set of the background text are subject-irrelevant, the feature set of the related text and the feature set of the background text are compared and analyzed to obtain the second intention factor set.

[0070] This application first needs to obtain the feature set of the transcript text, which is a set of key information extracted from the transcript text. For example, the key semantic information related to the time, place, and behavior of the parties involved in the case can be extracted from the transcript of the statements of the parties involved in the case to form this set.

[0071] Simultaneously, a background text feature set is acquired. This background text may contain environmental information related to the case, other relevant materials, etc., from which corresponding feature information is extracted. For example, in a traffic accident case, relevant information such as weather conditions and road surveillance records constitute the background text feature set.

[0072] The feature sets of the written record and the background text are analyzed to determine whether they are thematically related. For example, if the time and place of the incident described in the written record match the time and place of the surveillance footage in the background text, and the subjects involved are the same, it can be preliminarily determined that the subject is related; conversely, if there is a clear conflict, such as the written record stating that the incident occurred during the day, while the background surveillance shows that it occurred at night, it is determined that the subject is not related.

[0073] When the feature set of the transcript text and the feature set of the background text are determined to be topic-related, a comparative analysis is performed on these two feature sets. The comparison process may involve calculating the overlap of features, such as counting the frequency of identical keywords or the proportion of similar semantic segments. It may also involve analyzing the relationships between features, such as the interaction between the behavior of characters in the transcript and environmental factors in the background text. These comparisons yield a first intention factor set. This factor set contains a set of key factors that reflect the underlying intention of the text after comparing the two topic-related text feature sets.

[0074] If the subject matter is irrelevant, the relevant text feature set will be compared with the background text feature set. Here, the relevant text feature set may be extracted from other materials related to the case, such as relevant laws and regulations, records of similar cases, etc.

[0075] For example, analyzing the differences between the standards specified in the related text and the actual situation in the background text yields a second set of intent factors. This set of factors reflects the key elements of textual intent derived from the analysis of both the related and background texts when the topics of the transcript and the background text are unrelated.

[0076] The process involves processing the transcript text to obtain a transcript text feature set and a set of associated paragraphs. Based on semantic features, the associated text is segmented into paragraphs to obtain an associated text feature set. Similarly, based on background text, the transcript text is segmented into paragraphs to obtain a background text feature set. The specific steps include:

[0077] Extract the semantic distribution features of the transcript text, and then divide the transcript text into several paragraphs based on the semantic distribution features to obtain the transcript paragraph set;

[0078] The transcript paragraph set and the background text are divided into several paragraphs according to their own semantic features to obtain the related paragraph set and the background paragraph set;

[0079] After determining the tags of the transcript paragraph set and the associated text based on the semantic importance, the transcript text feature set and the associated text feature set are obtained respectively;

[0080] After determining the core intent paragraphs based on their semantic importance, the background text feature set is obtained.

[0081] This application uses natural language processing technology to analyze the transcripts of drunk driving cases. For example, from the party's transcript, "Last night I had dinner with friends at XX restaurant, during which I drank about 3 bottles of beer. After dinner, around 9:30 pm, I thought it was close to home, so I drove home myself, but I was stopped by the traffic police on the way," semantic distribution features such as "drinking time (during dinner last night)," "amount of alcohol consumed (3 bottles of beer)," "driving time (around 9:30 pm after dinner)," and "reason for driving (thinking it was close to home)" are extracted. These features reflect the distribution of key information in the transcript.

[0082] Based on the extracted semantic distribution features, the transcript text is divided into different paragraphs. In the above case, the "drinking situation paragraph" (describing the situation of eating and drinking at the restaurant) and the "driving decision and process paragraph" (narrating the decision to drive after the meal and the process of being stopped by the traffic police) can be segmented to obtain the transcript paragraph set.

[0083] The obtained transcript segments are further subdivided based on semantic features. In cases of drunk driving, the "drinking situation segment" can be further subdivided into related segments such as "drinking location" (XX restaurant) and "amount and type of alcohol consumed" (3 bottles of beer); the "driving decision and process segment" can be subdivided into related segments such as "driving time" (around 9:30 am), "reason for driving" (thinking it was close to home), and "circumstances of being checked" (being checked by traffic police on the road), thus forming a set of related segments.

[0084] Background text may include text information recorded by law enforcement recorders, surveillance text information of the road section where the incident occurred, etc. This background text is segmented according to its semantic features. For example, text recorded by law enforcement recorders can be segmented chronologically into "time segments of traffic police setting up checkpoints" and "time segments of the appearance and interception of the vehicle involved"; surveillance text of the road section where the incident occurred can be segmented into "trajectory segments of the vehicle involved," etc., ultimately resulting in a set of background segments.

[0085] Assess the semantic importance of each paragraph in the transcript. In drunk driving transcripts, paragraphs such as "amount and type of alcohol consumed" and "driving time" are crucial for determining whether drunk driving constitutes drunk driving and the severity of the offense, and therefore have high semantic importance. By marking these important paragraphs and key semantic information, a textual feature set is obtained, such as clearly recording key information like "the amount of alcohol consumed was 3 bottles of beer" and "the driving time was around 9:30 pm".

[0086] Related texts can be legal regulations related to drunk driving, similar case texts, etc. These related texts are marked according to their semantic importance. For example, in legal regulations concerning drunk driving, content such as "blood alcohol content standards for drunk driving" and "corresponding penalty clauses" has high semantic importance and, after being marked, forms a set of related text features.

[0087] Identify the core intent paragraphs within the background text set. In the background text recorded by the body camera, the paragraph recording the alcohol test results of the vehicle at the time of interception is considered a core intent paragraph; similarly, the paragraphs recording the vehicle's speed and route in the surveillance footage of the incident location may also be core intent paragraphs. Labeling these core intent paragraphs yields the background text feature set.

[0088] The intention association result set is obtained by comparing and analyzing the associated text feature set and the associated paragraph set. The specific steps include:

[0089] A dataset of semantic keyword frequencies for each paragraph in a statistically related text feature set;

[0090] The first semantic overlap frequency dataset is obtained by comparing the semantic keyword overlap frequency of paragraphs in the associated paragraph set and the associated text feature set.

[0091] The semantic proportion dataset is obtained by calculating the proportion values ​​of the first semantic overlap frequency dataset and the semantic keyword frequency dataset.

[0092] When there are values ​​in the semantic proportion dataset that are greater than the preset semantic threshold, the difference exceeding the preset semantic threshold is calculated to obtain the semantic excess dataset.

[0093] After extracting the first intent information from the associated paragraph set, a scene context dataset is set up.

[0094] The intent association result set is obtained by inputting the first intent information, the scene context dataset, the associated text feature set, and the semantic excess dataset into the intent association value prediction model.

[0095] This application uses a text feature set associated with drunk driving cases, including legal and regulatory texts related to drunk driving (such as the provisions on drunk driving in the Road Traffic Safety Law) and medical research texts on the effects of alcohol. The frequency of semantic keywords is statistically analyzed for each paragraph in these associated text feature sets. For example, in the relevant clauses of the Road Traffic Safety Law, terms such as "drunk driving," "alcohol content," "revocation of driver's license," and "criminal liability" may be semantic keywords, and their frequency in that paragraph is statistically analyzed. Assuming "drunk driving" appears 5 times and "alcohol content" appears 3 times, a semantic keyword frequency dataset is formed.

[0096] The associated paragraph set is a collection of paragraphs derived from the transcripts of drunk driving cases based on semantic features, such as "paragraphs about drinking" and "paragraphs about the driving process." These associated paragraph sets are then compared with paragraphs in the associated text feature set based on the frequency of semantic keyword overlap.

[0097] For example, the associated paragraph set describes "I drank about 3 bottles of beer, and then drove on the road." Keywords such as "beer" and "driving" are compared with keywords in the "Road Traffic Safety Law" clause paragraphs in the associated text feature set. Assuming there are 10 keywords in the associated text feature set and 5 overlapping keywords in the associated paragraph set, the overlap frequency is calculated. After comparing multiple paragraphs, the first semantic overlap frequency dataset is obtained.

[0098] The semantic overlap frequency dataset obtained in the previous step is compared with the semantic keyword frequency dataset obtained in the first step to calculate the proportion. For example, if the keyword "drunk driving" has a frequency of 5 times in a certain associated text feature set, and the overlap frequency of "drunk driving" in the compared associated paragraph set is 2 times, then the semantic proportion of the keyword "drunk driving" is 2 ÷ 5 = 0.4. This calculation is performed on all related keywords to obtain the semantic proportion dataset.

[0099] Set a semantic threshold, such as 0.3 (this can be determined based on actual needs and training with a large number of samples). Check the values ​​in the semantic proportion dataset. When there are values ​​greater than 0.3, calculate the difference exceeding 0.3. For example, the semantic proportion of the keyword "drunk driving" is 0.4, exceeding the preset threshold of 0.3, so the difference is 0.4 - 0.3 = 0.1. Calculate the difference for all keywords greater than the threshold to obtain the semantic excess dataset.

[0100] Extract primary intent information from the set of related paragraphs. For example, in cases of drunk driving, the set of related paragraphs might mention "I thought drinking this much alcohol would be fine," reflecting the individual's insufficient awareness of the risks of drunk driving. Combine this with the context of the incident, such as time (evening) and location (roads near the restaurant), to create a contextual dataset.

[0101] The initial intent information, the scene context dataset, the associated text feature set, and the semantic excess dataset are input into an intent association value prediction model (which can be trained based on machine learning algorithms, such as logistic regression, decision trees, etc.) to obtain an intent association result set. For example, it can be used to determine the degree of association between a party's statement and laws and regulations related to drunk driving, assess whether the party is aware that their behavior constitutes drunk driving, and other intent association situations.

[0102] When the feature set of the transcript text and the feature set of the background text are topic-related, the transcript text feature set and the background text feature set are compared and analyzed to obtain the first intention factor set, which specifically includes the following steps:

[0103] When the feature set of the transcript text and the feature set of the background text are topic-related, the feature set of the transcript text and the feature set of the background text are processed to obtain the second semantic overlap frequency dataset and semantic sentiment value;

[0104] The overlapping topic dataset is obtained by combining the statistical record text feature set with the background text feature set for each paragraph with the corresponding overlapping topics.

[0105] The second intent information of the background text feature set is extracted, and the second semantic overlap frequency dataset, semantic sentiment value, overlapping topic dataset, second intent information, scene context dataset, intent association result set and transcript text feature set are combined to form the first intent factor set.

[0106] In drunk driving cases, the text feature set of the transcript contains key information about the person involved regarding drinking and the driving process, such as "I drank a few bottles of beer and then drove on the road." The background text feature set may come from law enforcement recorder recordings, surrounding surveillance information, etc., such as "A vehicle was found to have an abnormal driving trajectory at a specific time on a certain road section and was intercepted."

[0107] The two feature sets are compared to identify recurring semantic keywords, such as "beer," "driving," and "driving," and the frequency of their co-occurrence in the two feature sets is calculated to form a second semantic overlap frequency dataset. For example, the word "driving" is mentioned multiple times in both the transcript and the background text, and the percentage of times it overlaps with other words is calculated.

[0108] Determine the emotional attitude implied in the statements of the parties involved in the transcript. For example, if a party says, "I really didn't realize that this was drunk driving, I just thought it was okay since the road was short," natural language processing technology can be used to analyze their emotional tendency and determine that the party had a sense of luck or negligence, resulting in a semantic emotional tendency value (which can be quantified into different levels of emotional indicators, such as -0.3 indicating slight luck).

[0109] The overlapping topics are determined by analyzing each paragraph in the feature set of the transcript and the feature set of the background text. In the case of drunk driving, the description of "driving on the road after drinking" in the transcript and the "inspection of vehicles driving after drinking" recorded by the law enforcement recorder in the background text are considered overlapping topics. The overlapping topics are then statistically analyzed to identify the corresponding paragraphs in the two feature sets, forming an overlapping topic dataset. This dataset includes information such as the number and specific location of overlapping topic paragraphs.

[0110] Extract information related to the intent of the case from background text feature sets (such as body camera recordings). For example, details in body camera recordings showing the person being stopped, such as nervousness and avoiding eye contact, reflect their potential awareness that their actions were illegal, i.e., secondary intent information.

[0111] The first intention factor set is formed by integrating the previously obtained second semantic overlap frequency dataset, semantic sentiment value, overlap topic dataset, second intention information, scene context dataset (including scene information such as the time of the incident being at night and the location being a road near a restaurant), intention association result set (the result obtained through previous analysis of associated texts and paragraphs), and transcript text feature set.

[0112] When the feature set of the transcript text and the feature set of the background text are topic-related, the feature set of the transcript text and the feature set of the background text are processed to obtain the second semantic overlap frequency dataset and semantic sentiment value. The specific steps include:

[0113] When the feature set of the transcript text and the feature set of the background text are subject-related, the second semantic overlap frequency dataset is obtained by comparing and averaging the semantic keyword overlap frequency of each paragraph in the feature set of the transcript text and the background text.

[0114] Extract the semantic sentiment values ​​of the corresponding paragraphs from the second semantic overlap frequency dataset.

[0115] In drunk driving cases, the text feature set of the transcript contains the party's description of the event, such as "I had a party with friends in the evening, drank a few glasses of baijiu (Chinese liquor), and then drove home." Semantic keywords such as "evening," "baijiu," "driving," and "home" are extracted from this. The background text feature set may come from law enforcement recorder recordings, intersection surveillance information, etc., such as "A vehicle was found to have an abnormal driving trajectory on a certain road section at night, and upon inspection, the driver showed signs of drinking alcohol." Relevant semantic keywords such as "night," "driving," and "drinking alcohol" are also extracted from this.

[0116] The semantic keywords of each paragraph in the transcript text feature set and the background text feature set are compared. For example, if the transcript text mentions "evening" and the background text mentions "nighttime," they are considered to overlap due to their similar semantics; "driving" and "driving" are also semantically related and are considered to overlap as well. The frequency of these semantic keywords appearing in both feature sets is counted to calculate the overlap frequency of each keyword. The overlap frequency data of multiple keywords are collected to obtain the second semantic overlap frequency dataset. For example, the keyword "drinking" appears frequently in both the transcript text and the background text, and its overlap frequency is calculated to be 0.6 (meaning it appears simultaneously in 60% of the relevant texts).

[0117] Sentiment analysis models can be used to categorize the emotions in text into positive, negative, and neutral categories, and assign corresponding quantitative values. For example, if a person's statement reflects a lack of awareness of the risks of drunk driving, it is judged as a negative sentiment tendency and assigned a sentiment tendency value, such as -0.4 (the value range can be set according to actual conditions; here, a negative number represents negative sentiment). By making such judgments on multiple related paragraphs, multiple semantic sentiment tendency values ​​are obtained, and these values ​​reflect the emotional attitude information contained in the text.

[0118] When the feature set of the transcript text and the feature set of the background text are topically irrelevant, the feature set of the related text is compared and analyzed with the feature set of the background text to obtain the second intention factor set. The specific steps include:

[0119] When the feature set of the transcript text and the feature set of the background text are subject-irrelevant, the subject information dataset is obtained by statistically analyzing the themes of the corresponding paragraphs in the feature set of the transcript text and the feature set of the background text.

[0120] Among them, the second semantic overlap frequency dataset, semantic sentiment tendency value, overlap topic dataset, second intent information, scene context dataset, transcript text feature set, intent association result set and topic information dataset are combined to form the second intent factor set.

[0121] In cases of drunk driving, if the written record mentions that the driver "drove after drinking at XX bar," but the background text feature set (such as nearby surveillance footage) shows the incident occurred on a highway far from the bar, and the time doesn't match, it indicates that the two are thematically unrelated. In this case, it's necessary to statistically analyze the thematic differences between paragraphs in the written record and background text feature sets. For example, the record might emphasize the drinking scene at the bar and the starting point of the drive, while the background text focuses on the vehicle's movement on the highway and the location of the interception. By determining the themes of these paragraphs and identifying information such as "location differences (bar vs. highway)" and "time differences (discrepancy between record time and surveillance time)," this thematic information should be compiled into a thematic information dataset.

[0122] The second semantic overlap frequency dataset reflects the frequency of semantic keyword overlap between the feature set of the transcript text and the feature set of the background text (even if the topics are unrelated, there is still some semantic connection). For example, in drunk driving cases, keywords such as "vehicle" and "driving" may appear in both the transcript and the background text. The dataset is formed by counting their overlap frequency.

[0123] The quantitative value of sentiment tendency obtained from relevant paragraphs of the transcript. For example, if the person complains in the transcript that "I only drank a little alcohol, but the traffic police are too strict", the sentiment analysis model will show a negative sentiment tendency value of dissatisfaction and resistance, such as -0.3.

[0124] Although the transcripts and background texts are not related in topic, there may still be a small number of overlapping topics, such as both involving "driving on the road". The dataset is formed by collecting information related to these overlapping topics.

[0125] Information related to the intent of the case is extracted from background text feature sets (such as body camera recordings). For example, if a person is evasive during questioning in a body camera recording, it reflects that they may have the intention to conceal certain facts, i.e., secondary intent information.

[0126] This dataset includes environmental information at the time of the incident, such as weather conditions (whether rain affected driving) and road conditions (traffic flow on highways). It also contains key information about the drunk driving incident described by the parties involved, such as the amount of alcohol consumed and the driving route. Furthermore, it includes a dataset reflecting the degree of intentional connection between texts, obtained through previous comparative analysis of related text feature sets and related paragraph sets. Finally, it includes a dataset reflecting the thematic differences between the transcript text feature set and the background text feature set.

[0127] Example 2 further illustrates a visual verification method proposed in this invention.

[0128] A visual evidence system, comprising:

[0129] Data Acquisition Module: Acquires transcript text, as well as associated text and background text.

[0130] Segmentation module: Processes the transcript text to obtain a transcript text feature set and a set of associated paragraphs; performs paragraph segmentation on the associated text based on semantic features to obtain a set of associated text features; and performs paragraph segmentation on the transcript text based on background text to obtain a set of background text features.

[0131] The first analysis module compares and analyzes the associated text feature set with the associated paragraph set to obtain the intention association result set;

[0132] The second analysis module: After processing and analyzing the feature set of the transcript text and the feature set of the background text under the condition of whether the topic is related, the first intention factor set and the second intention factor set are obtained respectively.

[0133] Display module: After inputting the first intent factor set and the second intent factor set into the intent recognition prediction model, the intent recognition prediction result is obtained. The intent recognition prediction result is combined with the intent association result set to identify the intent of the transcript text and then the visualized evidence content is displayed.

[0134] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a visual verification method.

[0135] like Figure 3 As shown, the electronic device may include a processor 610, a communications interface 620, a memory 630, and a communication bus 640. The processor 610, communications interface 620, and memory 630 communicate with each other via the communication bus 640. The processor 610 can call logical instructions from the memory 630 to execute a visual verification method.

[0136] Furthermore, the logical instructions in the aforementioned memory 630 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0137] On the other hand, the present invention also provides a computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is able to perform a visual verification method.

[0138] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform a visual verification method.

[0139] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0140] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method of visualizing a witness, characterized by, The method comprises the following steps: Collecting a record text, and associated text and background text associated with the record text; Processing the record text to obtain a record text feature set and an associated paragraph set, segmenting the associated text according to semantic features to obtain an associated text feature set, and segmenting the record text according to the background text to obtain a background text feature set; Comparing and analyzing the associated text feature set and the associated paragraph set to obtain an intent association result set; Judging whether the record text feature set and the background text feature set are relevant to the theme, and obtaining a first intent factor set and a second intent factor set after processing and analysis under the condition that the record text feature set and the background text feature set are relevant to the theme; Inputting the first intent factor set and the second intent factor set into an intent recognition prediction model to obtain an intent recognition prediction result, and displaying visual evidence content after identifying the intent of the record text by combining the intent recognition prediction result with the intent association result set.

2. The method of claim 1, wherein, Judging whether the record text feature set and the background text feature set are relevant to the theme, and obtaining a first intent factor set and a second intent factor set after processing and analysis under the condition that the record text feature set and the background text feature set are relevant to the theme, specifically comprising the following steps: When the record text feature set and the background text feature set are relevant to the theme, comparing and analyzing the record text feature set and the background text feature set to obtain the first intent factor set; When the record text feature set and the background text feature set are not relevant to the theme, comparing and analyzing the associated text feature set and the background text feature set to obtain the second intent factor set.

3. The method of claim 2, wherein, Processing the record text to obtain a record text feature set and an associated paragraph set, segmenting the associated text according to semantic features to obtain an associated text feature set, and segmenting the record text according to the background text to obtain a background text feature set, specifically comprising the following steps: Extracting semantic distribution features of the record text, segmenting the record text into a plurality of paragraphs according to the semantic distribution features to obtain a record paragraph set; Segmenting the record paragraph set and the background text into a plurality of paragraphs according to their respective semantic features to obtain an associated paragraph set and a background paragraph set; Determining the labels of the record paragraph set and the associated text according to the semantic importance to obtain a record text feature set and an associated text feature set, respectively; Determining the labels of the core intent paragraphs of the background paragraph set according to their respective semantic importance to obtain a background text feature set.

4. The method of claim 3, wherein, Comparing and analyzing the associated text feature set and the associated paragraph set to obtain an intent association result set, specifically comprising the following steps: Statistically obtaining a semantic keyword frequency data set of each paragraph in the associated text feature set; Comparing and analyzing the associated paragraph set and the paragraphs in the associated text feature set according to the corresponding semantic keyword overlap frequency to obtain a first semantic overlap frequency data set; Calculating the proportion of the first semantic overlap frequency data set and the semantic keyword frequency data set to obtain a semantic proportion data set; When the semantic proportion data set contains a value greater than a preset semantic threshold, calculating the difference between the value and the preset semantic threshold to obtain a semantic excess data set; Extracting first intent information of the associated paragraph set to set a scene context data set; The first intention information, the scene context data set, the associated text feature set and the semantic excess data set are input into an intention association value prediction model to obtain an intention association result set.

5. The method of claim 4, wherein, When the record text feature set and the background text feature set are in a theme-related condition, the record text feature set and the background text feature set are compared and analyzed to obtain a first intention factor set, specifically including the following steps: When the record text feature set and the background text feature set are in a theme-related condition, the record text feature set and the background text feature set are processed to obtain a second semantic coincidence frequency data set and a semantic sentiment tendency value. The coincidence theme data set is obtained by counting the corresponding coincident themes of each paragraph in the record text feature set and the background text feature set. The second semantic coincidence frequency data set, the semantic sentiment tendency value, the coincidence theme data set, the second intention information, the scene context data set, the intention association result set and the record text feature set are combined to form the first intention factor set.

6. The method of claim 5, wherein, When the record text feature set and the background text feature set are in a theme-related condition, the record text feature set and the background text feature set are processed to obtain a second semantic coincidence frequency data set and a semantic sentiment tendency value, specifically including the following steps: When the record text feature set and the background text feature set are in a theme-related condition, the second semantic coincidence frequency data set is obtained by counting the comparison coincidence frequency of the corresponding semantic keywords of each paragraph in the record text feature set and the background text feature set. The semantic sentiment tendency value of the corresponding paragraph of the second semantic coincidence frequency data set is extracted.

7. A visualization method according to claim 6, wherein, When the record text feature set and the background text feature set are in a theme-unrelated condition, the associated text feature set and the background text feature set are compared and analyzed to obtain a second intention factor set, specifically including the following steps: When the record text feature set and the background text feature set are in a theme-unrelated condition, the theme information data set is obtained by counting the themes between the corresponding paragraphs in the record text feature set and the background text feature set. The second semantic coincidence frequency data set, the semantic sentiment tendency value, the coincidence theme data set, the second intention information, the scene context data set, the record text feature set, the intention association result set and the theme information data set are combined to form the second intention factor set.

8. A visual evidence system for use in a visual evidence method according to any one of claims 1 to 7, characterized in that It includes: A collection module: collecting a record text, associated text and background text associated with the record text; A segmentation module: processing the record text to obtain a record text feature set and an associated paragraph set, segmenting the associated text according to semantic features to obtain an associated text feature set, and segmenting the record text according to the background text to obtain a background text feature set; A first analysis module: comparing and analyzing the associated text feature set and the associated paragraph set to obtain an intention association result set; A second analysis module: determining whether the record text feature set and the background text feature set are in a theme-related condition, and processing and analyzing to obtain a first intention factor set and a second intention factor set, respectively; The display module: input the first set of intention factors and the second set of intention factors into the intention recognition prediction model to obtain an intention recognition prediction result, and display visual evidence content after identifying the intention of the record text according to the intention recognition prediction result and the set of intention association results.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the visual evidence method according to any one of claims 1 to 7 when executing the program.

Images