An ai-based mediation record legal fact keyword extraction system and method

Abstract

The application is specifically a mediation record legal fact keyword extraction system and method based on AI, which comprises the following parts: a data collection module; an analysis and processing module: the video data corresponding to the high-frequency words are analyzed to obtain visual values, and the voice data are analyzed to obtain phonetic values; the visual values and phonetic values corresponding to the high-frequency words are comprehensively analyzed to obtain evaluation values; a correction module.In the application, the video, voice and electronic record data in the mediation record are comprehensively analyzed, which can more comprehensively capture case information; through this multi-modal data fusion mode, information omission and inaccuracy caused by a single data source are avoided, thereby improving the accuracy and comprehensiveness of keyword extraction, and helping legal professionals better understand the overall situation and dispute focus of the case.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and legal document processing technology, and in particular to an AI-based system and method for extracting legal fact keywords from mediation records. Background Technology

[0002] In legal mediation, accurately extracting key legal facts from mediation records is crucial for understanding the full picture of a case, grasping the focus of the dispute, and applying relevant legal provisions.

[0003] Traditional keyword extraction methods mainly rely on manual review of mediation records for identification, which not only consumes a lot of manpower and time, but is also easily affected by subjective factors, resulting in a lack of objectivity and consistency in the extraction results.

[0004] With the development of artificial intelligence technology, some keyword extraction systems based on text analysis have emerged. However, most of these systems only focus on the text content of the transcripts and ignore the unstructured data of the parties' facial expressions, body movements, and voices in the video during the mediation process, which contain rich emotional and attitudinal information.

[0005] The lack of comprehensive consideration of this multimodal information greatly reduces the accuracy and comprehensiveness of keyword extraction, making it difficult to accurately reflect the true situation of the case and the true intentions of the parties involved.

[0006] Furthermore, existing systems often lack in-depth optimization for the legal field, failing to efficiently associate extracted keywords with relevant laws, regulations, judicial interpretations, and typical cases, thus failing to adequately meet the actual needs of legal practice. Summary of the Invention

[0007] The purpose of this invention is to provide an AI-based system and method for extracting legal fact keywords from mediation records in order to solve the above-mentioned problems.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] An AI-based system for extracting legal fact keywords from mediation records includes the following components:

[0010] Data collection module: Extracts mediation record data from cases, including video data, voice data, and electronic mediation record data during the mediation process;

[0011] Feature extraction module: Obtain high-frequency words from the electronic mediation transcript and extract the video and audio data corresponding to each high-frequency word;

[0012] Analysis and processing module: Analyzes video data corresponding to high-frequency words to obtain visualization values, and analyzes audio data to obtain speech conversion values; after comprehensively analyzing the visualization values ​​and speech conversion values ​​corresponding to high-frequency words, an evaluation value is obtained;

[0013] Correction Module: After sorting the evaluation values ​​in descending order, the module presents them to legal professionals for review of high-frequency words. Based on the review results, the high-frequency words are reordered and recorded as keywords. Based on the keyword extraction results, relevant laws, regulations, judicial interpretations, and typical cases are quickly located.

[0014] Preferably, it also includes an AI preprocessing module:

[0015] The collected electronic mediation transcript data was converted into a unified code, redundant line breaks and consecutive spaces were removed, the text was formatted into paragraphs and denoised, and stop words were filtered according to a legal stop word list.

[0016] Preferably, the acquisition of high-frequency words in the electronic mediation record specifically includes the following parts:

[0017] Count the total number of words in the electronic mediation record and record it as the total number of words;

[0018] Obtain the number of each word and the number of its corresponding similar words; sum the number of each word and the number of its corresponding similar words to get the total number of words; where similar words include synonyms, near-synonyms, and words in different positions.

[0019] The ratio of the total number of words to the total vocabulary is recorded as the single word frequency; a single word frequency threshold is preset, and single word frequencies greater than the single word frequency threshold are recorded as high-frequency words, and all high-frequency words are obtained in this way.

[0020] Preferably, the video data corresponding to high-frequency words is analyzed, specifically including the following parts:

[0021] Obtain video data of the parties involved corresponding to high-frequency words, and extract keyframes from them;

[0022] Mark the center of the person's head as the origin, obtain the trajectory between the initial position and the final position of the person's head within the time period corresponding to the high-frequency words, connect the trajectory formed by the initial position and the final position with a straight line, calculate the length of the straight line, and record it as the amplitude value.

[0023] Extract the facial image of the person from the keyframe, obtain the vertical distance between the upper and lower eyelids of the person, as well as the maximum and minimum values ​​of the vertical distance between the upper and lower eyelids of the person, and calculate the difference between the maximum and minimum values ​​of the vertical distance between the upper and lower eyelids of the person to obtain the range of change.

[0024] The vertical distance between the upper and lower eyelids of the subject within the time period corresponding to the high-frequency words is obtained, and the initial value of the vertical distance is determined from it; the time corresponding to the change of the vertical distance between the upper and lower eyelids from the initial value to the maximum value is obtained, and this time is recorded as the difference time. A difference time threshold is preset, and the difference time threshold is divided by the difference time to obtain the difference ratio.

[0025] The amplitude value, variation range value, and anomaly ratio are comprehensively analyzed to obtain the visualization value, and then the visualization value of each high-frequency word is obtained in turn.

[0026] Preferably, the step of obtaining a visualized value by comprehensively analyzing the amplitude value, the range of change value, and the anomaly ratio specifically includes the following parts:

[0027] After normalizing the amplitude value, range value, and anomaly ratio, the amplitude value and range value are used as two legs of a right triangle, and the remaining leg is connected to form a complete right triangle. The anomaly ratio is used as the height of the right triangle to construct a triangular pyramid model. The volume of the triangular pyramid model is calculated and the volume is used as a visualization value.

[0028] Preferably, the acquisition of the speech value specifically includes the following parts:

[0029] Obtain the speech data of the person corresponding to the high-frequency words, and obtain the pitch, speech rate and amplitude of the high-frequency words;

[0030] Obtain the pitch variation range corresponding to high-frequency words, extract the highest and lowest pitch values ​​from them, and calculate the pitch variation value by calculating the difference between the highest and lowest pitch values.

[0031] Locate the highest and lowest pitch values ​​in the speech data and the corresponding time points; calculate the difference between the time points corresponding to the highest and lowest pitch values ​​and take the absolute value to obtain the time difference;

[0032] The pitch coefficient is obtained by weighting the pitch change value and the time difference value.

[0033] Determine the number of characters in high-frequency words and obtain the pause time between each character in the high-frequency words; preset a pause time threshold, calculate the difference between each pause time and the pause time threshold, and take the absolute value to obtain the pause difference;

[0034] The average pause is obtained by summing the individual pause differences and dividing by the number of pause differences.

[0035] Extract the minimum and maximum pause times from each pause time, and calculate the difference between the minimum and maximum pause times to obtain the pause range value;

[0036] The speech rate coefficient is obtained by weighting the average pause value and the pause range value.

[0037] The amplitude coefficient is obtained by analyzing the speech amplitude corresponding to high-frequency words;

[0038] The speech value is obtained by comprehensively processing the pitch coefficient, speech rate coefficient, and amplitude coefficient.

[0039] Preferably, the process for obtaining the amplitude coefficient is as follows:

[0040] Obtain the amplitude of the speech signal within the time period corresponding to the high-frequency words, and preset the amplitude fluctuation range of the speech signal; record the amplitude that is not within the amplitude fluctuation range of the speech signal as abnormal amplitude, count the number of abnormal amplitudes within the time period corresponding to the high-frequency words, and divide it by the number of speech signal amplitudes to obtain the abnormal amplitude ratio;

[0041] Obtain the duration of each abnormal amplitude and sum the durations of each abnormal amplitude to obtain the total abnormal time;

[0042] After normalizing the abnormal vibration ratio and the total abnormal vibration time, they are used as two legs of a right triangle. The remaining leg is then connected to form a complete right triangle. The area of ​​the right triangle is calculated and recorded as the amplitude coefficient.

[0043] Preferably, the evaluation value obtained by comprehensively analyzing the visualization and speech values ​​corresponding to high-frequency words specifically includes the following parts:

[0044] The weighting factors for the visual and speech values ​​are preset. The visual and speech values ​​are multiplied by their corresponding weighting factors and then summed to obtain the evaluation value.

[0045] A method for extracting legal fact keywords from mediation records based on AI, comprising the following parts:

[0046] Data collection: Acquiring electronic record data during the mediation process;

[0047] Data analysis: High-frequency words are extracted from the electronic transcript data, and the corresponding video and audio data are analyzed to obtain visualization and speech values, respectively.

[0048] Preliminary processing: The visual and speech values ​​are combined to obtain the evaluation value, and the corresponding high-frequency words are sorted according to the evaluation value;

[0049] Assessment and processing: The high-frequency words corresponding to the assessment values ​​are submitted to legal professionals for review. Based on the review results, the high-frequency words are reordered and recorded as keywords.

[0050] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0051] 1. This invention, by comprehensively analyzing video, audio, and electronic record data in mediation transcripts, can capture case information more comprehensively. Through this multimodal data fusion approach, it avoids information omissions and inaccuracies that may be caused by a single data source, thereby improving the accuracy and comprehensiveness of keyword extraction and helping legal professionals better understand the whole picture and the focus of the dispute.

[0052] 2. This invention can quickly locate the relevant laws, regulations, judicial interpretations, and typical cases corresponding to each keyword. This not only saves a lot of time and effort and improves work efficiency, but also ensures the accuracy and consistency of the application of law and improves the quality of legal practice. Version management of keyword extraction results makes it easy to trace and compare different versions of extraction results, which helps to continuously optimize the keyword extraction process and further improve the efficiency and quality of legal practice. Attached Figure Description

[0053] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0054] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0055] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0056] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0057] Please see Figure 1 As shown, the present invention provides a technical solution:

[0058] An AI-based system for extracting legal fact keywords from mediation records includes the following components:

[0059] Data collection module: Extract the mediation record data of the case, including video data, voice data and electronic mediation record data during the mediation process;

[0060] AI preprocessing module: Convert the collected electronic mediation record data into a unified encoding to avoid character display errors caused by encoding problems; Remove redundant line breaks and consecutive spaces, and organize the text into a single line or paragraph format for subsequent processing; And perform denoising processing. According to the stop word list in the legal field, in addition to common stop words (such as "of", "already", "is", etc.), it also includes common meaningless words in the legal field (such as "in view of", "according to", etc.), filter out stop words to reduce data noise;

[0061] Feature extraction module: Obtain the high-frequency words in the electronic mediation record, and extract the corresponding video data and voice data for each high-frequency word;

[0062] Obtain the high-frequency words in the electronic mediation record, which specifically includes the following parts:

[0063] Count the total number of all words in the electronic mediation record and record it as the total number of words;

[0064] Obtain the number of each word and the number of its corresponding similar words; After summing up the number of each word and the number of its corresponding similar words, obtain the total number of words; Among them, similar words include synonyms, near synonyms, and hypernyms / hyponyms;

[0065] Record the ratio of the total number of words to the total number of words as the single word frequency; Preset a single word frequency threshold, and record the single word frequency greater than the single word frequency threshold as a high-frequency word, and obtain all high-frequency words based on this;

[0066] Analysis and processing module: Analyze the video data corresponding to the high-frequency words to obtain a visualization value, and analyze the voice data to obtain a voice value; Comprehensively analyze the visualization value and voice value corresponding to the high-frequency words to obtain an evaluation value;

[0067] Analyze the video data corresponding to the high-frequency words, which specifically includes the following parts:

[0068] Obtain the video data of the parties corresponding to the high-frequency words, and extract key frames from it;

[0069] Mark the center of the parties' heads as the origin, obtain the trajectory between the initial position and the final position of the parties' heads during the time period corresponding to the high-frequency words, connect the trajectory formed by the initial position and the final position with a straight line, and calculate the length of this straight line, and record it as the amplitude value;

[0070] Extract the facial image of the person from the keyframe, obtain the vertical distance between the upper and lower eyelids of the person, as well as the maximum and minimum values ​​of the vertical distance between the upper and lower eyelids of the person, and calculate the difference between the maximum and minimum values ​​of the vertical distance between the upper and lower eyelids of the person to obtain the range of change.

[0071] The vertical distance between the upper and lower eyelids of the subject within the time period corresponding to the high-frequency words is obtained, and the initial value of the vertical distance is determined from it; the time corresponding to the change of the vertical distance between the upper and lower eyelids from the initial value to the maximum value is obtained, and this time is recorded as the difference time. A difference time threshold is preset, and the difference time threshold is divided by the difference time to obtain the difference ratio.

[0072] The determination of the initial value of the vertical distance between the upper and lower eyelids includes: acquiring keyframe images extracted from the video data corresponding to all high-frequency words, recording the vertical distance between the upper and lower eyelids of the person in each keyframe image, counting the number of times each identical vertical distance occurs, and dividing the number of times each identical vertical distance occurs by the total number of times in the keyframe images to obtain the occurrence ratio.

[0073] Sort the occurrence ratios in descending order of size, select the largest occurrence ratio, and record the vertical distance corresponding to the largest occurrence ratio as the initial value of the vertical distance.

[0074] After comprehensively analyzing the amplitude value, the range of change value, and the anomaly ratio, a visualization value is obtained, and then the visualization values ​​of each high-frequency word are obtained in turn.

[0075] The visualized value is obtained by comprehensively analyzing the amplitude value, the range of change value, and the ratio of anomalies, which specifically includes the following parts:

[0076] After normalizing the amplitude value, range value, and anomaly ratio, the amplitude value and range value are used as two legs of a right triangle, and the remaining leg is connected to form a complete right triangle. The anomaly ratio is used as the height of the right triangle to construct a triangular pyramid model. The volume of the triangular pyramid model is calculated and the volume is used as a visualization value.

[0077] The acquisition of speech-based values ​​includes the following parts:

[0078] Obtain the speech data of the person corresponding to the high-frequency words, and obtain the pitch, speech rate and amplitude of the high-frequency words;

[0079] Obtain the pitch variation range corresponding to high-frequency words, extract the highest and lowest pitch values ​​from them, and calculate the pitch variation value by calculating the difference between the highest and lowest pitch values.

[0080] Locate the highest and lowest pitch values ​​in the speech data and the corresponding time points; calculate the difference between the time points corresponding to the highest and lowest pitch values ​​and take the absolute value to obtain the time difference;

[0081] The pitch coefficient is obtained by weighting the pitch change value and the time difference value.

[0082] The pitch coefficient is obtained by multiplying the pitch change value and the time difference value by the corresponding weighting factor and then summing the products.

[0083] Determine the number of characters in high-frequency words and obtain the pause time between each character in the high-frequency words; preset a pause time threshold, calculate the difference between each pause time and the pause time threshold, and take the absolute value to obtain the pause difference;

[0084] The average pause is obtained by summing the individual pause differences and dividing by the number of pause differences.

[0085] Extract the minimum and maximum pause times from each pause time, and calculate the difference between the minimum and maximum pause times to obtain the pause range value;

[0086] The speech rate coefficient is obtained by weighting the average pause value and the pause range value.

[0087] The amplitude coefficient is obtained by analyzing the speech amplitude corresponding to high-frequency words;

[0088] The process of obtaining the amplitude coefficient is as follows:

[0089] Obtain the amplitude of the speech signal within the time period corresponding to the high-frequency words, and preset the amplitude fluctuation range of the speech signal; record the amplitude that is not within the amplitude fluctuation range of the speech signal as abnormal amplitude, count the number of abnormal amplitudes within the time period corresponding to the high-frequency words, and divide it by the number of speech signal amplitudes to obtain the abnormal amplitude ratio;

[0090] Obtain the duration of each abnormal amplitude and sum the durations of each abnormal amplitude to obtain the total abnormal time;

[0091] After normalizing the abnormal vibration ratio and the total abnormal vibration time, they are used as two legs of a right triangle, and the remaining leg is connected to form a complete right triangle. The area of ​​the right triangle is calculated and recorded as the amplitude coefficient.

[0092] The speech value is obtained by comprehensively processing the pitch coefficient, speech rate coefficient, and amplitude coefficient.

[0093] The pitch coefficient, speech rate coefficient, and amplitude coefficient are preset with weighting factors. The pitch coefficient, speech rate coefficient, and amplitude coefficient are multiplied by their corresponding weighting factors and then summed to obtain the speech value.

[0094] The evaluation value is obtained by comprehensively analyzing the visualization and speech values ​​corresponding to high-frequency words, and specifically includes the following parts:

[0095] The weighting factors for the visual and speech values ​​are preset. The visual and speech values ​​are multiplied by their corresponding weighting factors and then summed to obtain the evaluation value.

[0096] The correction module sorts the evaluation values ​​in descending order and presents them to legal professionals for review of high-frequency words. Based on the review results, the high-frequency words are reordered and recorded as keywords. Based on the keyword extraction results, relevant laws, regulations, judicial interpretations, and typical cases are quickly located.

[0097] The video and audio data corresponding to each high-frequency word are presented to legal professionals, who then score the importance of the words. The importance score ranges from 0 to 10, and the importance is directly proportional to the score.

[0098] The evaluation values ​​of each high-frequency word are converted into importance scores, which are then used as the base scores for each high-frequency word. The importance scores assigned to each high-frequency word by legal professionals are calculated and summed with the base scores to obtain the final score.

[0099] Based on the final score of each high-frequency word, they are sorted in descending order according to the score, and the high-frequency words are recorded as keywords.

[0100] Output module: Outputs the finalized keywords in the form of a list, with each keyword accompanied by its corresponding final score; so that users can understand their importance. The output format can be customized according to user needs, such as CSV, JSON, TXT, etc.

[0101] The keyword extraction results are stored in a database for easy subsequent querying, statistics and analysis; version management of the keyword extraction results is implemented, recording information such as the extraction time, model and parameters used each time, which facilitates the tracking and comparison of different versions of the extraction results;

[0102] By extracting results from keywords, key legal facts in mediation records can be quickly identified, helping legal professionals better understand the overall picture and focus of the dispute. For example, in contract disputes, analyzing keywords such as "contract terms," ​​"performance status," and "liability for breach of contract" can provide a clear understanding of the core issues in the case.

[0103] By utilizing the relationships between keywords, we can analyze the legal relationships in a case, such as contractual relationships, tort relationships, and property rights relationships. This helps legal professionals to accurately apply the law and propose reasonable solutions.

[0104] Based on the keyword extraction results, relevant laws, regulations, judicial interpretations, and typical cases can be quickly located. For example, when handling tort disputes, keywords such as "tortious act," "damage consequences," and "liability for compensation" can be used to find relevant tort liability legal provisions and similar cases, providing a legal basis for handling the case.

[0105] A method for extracting legal fact keywords from mediation records based on AI, comprising the following parts:

[0106] Data collection: Acquiring electronic record data during the mediation process;

[0107] Data analysis: High-frequency words are extracted from the electronic transcript data, and the corresponding video and audio data are analyzed to obtain visualization and speech values, respectively.

[0108] Preliminary processing: The visual and speech values ​​are combined to obtain the evaluation value, and the corresponding high-frequency words are sorted according to the evaluation value;

[0109] Assessment and Processing: The high-frequency words corresponding to the assessment values ​​are submitted to legal professionals for review. Based on the review results, the high-frequency words are reordered and recorded as keywords. Based on the keyword extraction results, relevant laws, regulations, judicial interpretations, and typical cases are quickly located.

[0110] The video and audio data corresponding to each high-frequency word are presented to legal professionals. After the legal professionals score the importance of each high-frequency word, the final score is obtained, and the high-frequency words are sorted in descending order according to the final score.

[0111] The above formulas are derived from software simulations using a large amount of data and are selected to be close to the actual values. The influence weight factors and specific coefficient values ​​in the formulas are set by those skilled in the art based on the actual situation and can be adjusted and modified in the future.

[0112] The above description of the embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-based mediation record legal fact keyword extraction system, characterized in that, The method comprises the following parts: A data collection module: extract the mediation record data of a case, including video data, voice data and electronic mediation record data in the mediation process; A feature extraction module: obtain high-frequency words in the electronic mediation record, and extract video data and voice data corresponding to each high-frequency word; An analysis and processing module: analyze the video data corresponding to the high-frequency words to obtain a visual value, and analyze the voice data to obtain a phonetic value; After comprehensive analysis of the visual value and the phonetic value corresponding to the high-frequency words, an evaluation value is obtained; Which includes: After comprehensive analysis of the amplitude value, the change range value and the difference ratio value, a visual value is obtained, and the visual values of each high-frequency word are obtained in turn, which specifically includes the following parts: After normalization processing of the amplitude value, the change range value and the difference ratio value, the amplitude value and the change range value are respectively taken as two legs of a right triangle, and the remaining leg is connected to form a complete right triangle; the difference ratio value is taken as the height of the right triangle to construct a three-prism model, the volume of the three-prism model is calculated, and the volume is taken as the visual value; The acquisition of the phonetic value specifically includes the following parts: Obtain the voice data of the parties corresponding to the high-frequency words, and obtain the pitch, speed and amplitude corresponding to the high-frequency words; Obtain the pitch change range corresponding to the high-frequency words, and extract the highest pitch value and the lowest pitch value from the pitch change range; after difference calculation of the highest pitch value and the lowest pitch value, a pitch change value is obtained; Locate the time points corresponding to the highest pitch value and the lowest pitch value from the voice data; after difference calculation of the time points corresponding to the highest pitch value and the lowest pitch value, an absolute value is taken to obtain a time difference value; After weighted calculation of the pitch change value and the time difference value, a pitch coefficient is obtained; Determine the number of characters of the high-frequency words, and obtain the pause time between each character of the high-frequency words; preset a pause time threshold value, difference calculation is performed between each pause time and the pause time threshold value, and an absolute value is taken to obtain a pause difference value; After summing up each pause difference value, divide the pause difference value by the number of pause difference values to obtain a pause average value; Extract the minimum pause time and the maximum pause time from each pause time, and difference calculation is performed between the minimum pause time and the maximum pause time to obtain a pause range value; After weighted calculation of the pause average value and the pause range value, a speed coefficient is obtained; After analyzing the voice amplitude corresponding to the high-frequency words, an amplitude coefficient is obtained; After comprehensive processing of the pitch coefficient, the speed coefficient and the amplitude coefficient, a phonetic value is obtained; A correction module: after arranging the evaluation values in descending order according to the size, the legal professional is presented to review the high-frequency words, the high-frequency words are reordered according to the review results, and the high-frequency words are recorded as key words; according to the key word extraction result, relevant laws and regulations, judicial interpretations and typical cases are quickly located.

2. The AI-based mediation record legal fact keyword extraction system according to claim 1, wherein, It also includes an AI preprocessing module: Convert the collected electronic mediation record data into a unified code, remove redundant line breaks and continuous spaces, arrange the text into a paragraph format, and perform denoising processing; according to the stop word table in the legal field, filter the stop words. 3.The AI-based mediation record legal fact keyword extraction system of claim 1, wherein, The high-frequency words in the electronic mediation record are obtained, specifically including the following parts: Counting the number of all words in the electronic mediation record and recording the total number of words; Obtaining the number of each word and the number of similar words corresponding to each word; summing the number of each word and the number corresponding to the similar word to obtain the total number of words; wherein the similar words include synonyms, near-synonyms, and superordinate and subordinate words; The ratio of the total number of words to the total number of words is recorded as the individual word frequency; a preset individual word frequency threshold is set, and the individual word frequency greater than the individual word frequency threshold is recorded as a high-frequency word, and all high-frequency words are obtained. 4.The AI-based mediation record legal fact keyword extraction system of claim 1, wherein, The video data corresponding to the high-frequency word is analyzed, which specifically includes the following parts: Obtain the video data corresponding to the high-frequency word of the parties, and extract the key frames therefrom; Mark the center of the head of the parties as the origin, obtain the trajectory between the initial position and the final position of the head of the parties within the time period corresponding to the high-frequency word, connect the initial position and the final position to form a straight line, and calculate the length of the straight line, which is recorded as the amplitude value; Extract the face image of the parties from the key frame, obtain the vertical distance between the upper and lower eyelids of the parties, and the maximum and minimum values of the vertical distance between the upper and lower eyelids of the parties, and the difference between the maximum and minimum values of the vertical distance between the upper and lower eyelids of the parties is obtained. The change range value is obtained by difference calculation; Obtain the vertical distance between the upper and lower eyelids of the parties within the time period corresponding to the high-frequency word, and determine the initial value of the vertical distance therefrom; obtain the time corresponding to the change of the vertical distance between the upper and lower eyelids from the initial value to the maximum value, and record the time as the difference time. A difference time threshold is preset, and the difference time threshold is divided by the difference time to obtain the difference ratio. 5.The AI-based mediation record legal fact keyword extraction system of claim 1, wherein, The amplitude coefficient is obtained as follows: Obtain the speech signal amplitude in the time period corresponding to the high-frequency word, and preset the speech signal amplitude fluctuation range; the amplitude not in the speech signal fluctuation range is recorded as an abnormal amplitude, the number of abnormal amplitudes in the time period corresponding to the high-frequency word is counted, and the abnormal amplitude ratio is obtained by dividing the number of abnormal amplitudes by the number of speech signal amplitudes; Obtain the duration of each abnormal amplitude, and accumulate the duration of each abnormal amplitude to obtain the total abnormal time; After normalization of the abnormal amplitude ratio and the total abnormal time, they are respectively taken as two right-angle sides of a right-angle triangle, and the remaining one side is connected to form a complete right-angle triangle, the area of the right-angle triangle is calculated, and recorded as the amplitude coefficient. 6.The AI-based mediation record legal fact keyword extraction system of claim 5, wherein, The evaluation value is obtained by comprehensively analyzing the visual value and the phonetic value corresponding to the high-frequency word, which specifically includes the following parts: Preset the weight factor of the visual value and the phonetic value, respectively multiply the visual value and the phonetic value by the corresponding weight factor, and sum to obtain the evaluation value.

7. An AI-based mediation record legal fact keyword extraction method, using any one of the AI-based mediation record legal fact keyword extraction systems of claims 1-6, characterized in that, It includes the following parts: Data acquisition: obtaining electronic record data in the mediation record process; Data analysis: extracting high-frequency words from electronic record data, and analyzing video data and voice data corresponding to high-frequency words to obtain visual value and phonetic value respectively; Preliminary processing: obtaining the evaluation value by comprehensively processing the visual value and the phonetic value, and sorting the corresponding high-frequency words according to the evaluation value; Evaluation processing: the high-frequency words corresponding to the evaluation values are presented to legal professionals for auditing, the high-frequency words are reordered according to the auditing results, and the high-frequency words are recorded as keywords.

Images