Film and television supervision analysis method and system based on data mining and storage medium

Abstract

The invention relates to the technical field of data processing, and discloses a film and television supervision analysis method and system based on data mining and a storage medium. The method comprises the following steps: collecting film and television data in a panoramic manner, and carrying out cross-media transcoding processing to obtain a multi-dimensional data set; performing video semantic extraction, audio emotion recognition and text tendency analysis on the data set to generate a time sequence mark list; scoring the key elements according to social influence, propagation depth and audience acceptability to form a dynamic score table; analyzing the propagation path based on the score table, and constructing an influence map; sorting supervision points according to the atlas, and making an intelligent supervision scheme; the execution deviation is analyzed through effect feedback, and the supervision parameters are optimized. According to the method, multi-dimensional content feature extraction, social influence evaluation and propagation path analysis of the film and television works are realized in a mass multimedia content environment, a precise and differentiated intelligent supervision scheme is formed, and the method has self-optimization and adjustment capabilities at the same time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a film and television supervision analysis method and system based on data mining and a storage medium. BACKGROUND

[0002] With the rapid development and popularization of Internet video platforms, the number of various types of film and television works has shown explosive growth, and the content types have become increasingly diversified. Traditional film and television content supervision mainly relies on manual auditing methods, through professional auditors to watch, screen and classify film and television works one by one, supplemented by simple technical means such as keyword filtering and blacklist management. This supervision mode can meet the basic needs when the content scale is relatively limited in the early Internet, and countries and regions have also established rating systems and content review standards to guide the distribution and dissemination of film and television content. Some regions have also developed automated content recognition technologies, such as rule-based sensitive content screening systems and simple image recognition and voice analysis tools, to improve auditing efficiency.

[0003] However, with the rise of short videos, network live broadcasts and user-generated content (UGC), traditional supervision methods face many shortcomings and challenges. Manual auditing is inefficient and cannot meet the real-time review needs of massive content, resulting in widespread influence of harmful content before it is discovered. Secondly, simple keyword filtering and rule matching mechanisms are easily evaded by content creators, making it difficult to identify harmful content expressed implicitly and integrated across media. Thirdly, traditional supervision methods lack the ability to analyze content dissemination paths and social impact, making it difficult to accurately assess the actual harm of content and prioritize processing order. In addition, existing technologies also have problems such as rigid supervision standards, coarse-grained rating, lack of personalized measures, and difficulty in adapting to the differentiated needs of different audience groups. Most importantly, traditional supervision methods lack self-learning and optimization mechanisms, making it difficult to adapt to changing content forms and dissemination patterns, and the balance between supervision effectiveness and execution cost is difficult to grasp. SUMMARY

[0004] The present application provides a film and television supervision analysis method and system based on data mining and a storage medium, which is used to realize multi-dimensional content feature extraction, social impact assessment and dissemination path analysis of film and television works in a massive multimedia content environment, and accordingly form a precise and differentiated intelligent supervision scheme, while having the ability of self-optimization and adjustment.

[0005] In a first aspect, the application provides a data mining-based film and television supervision analysis method, which comprises: panoramic data collection on film and television works of a video platform, hierarchical standardized arrangement of video, audio and text information through cross-media transcoding processing to obtain a multi-dimensional film and television content dataset; scene semantic extraction of video frames, multi-level emotion recognition of audio and cultural value trend analysis of text through content feature recognition based on the multi-dimensional film and television content dataset to obtain a time sequence correlation content marking list; cross scoring of key elements according to social influence intensity, transmission penetration depth and audience psychological acceptance through hierarchical scoring based on the time sequence correlation content marking list to obtain a dynamic weight content scoring table; network transmission path analysis of film and television works through correlation tracking based on the dynamic weight content scoring table to obtain a stereoscopic content influence map; priority ranking of supervision points through decision rules based on the stereoscopic content influence map to obtain an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individualized grading requirements; deviation correction analysis of the execution effect of supervision measures through effect feedback based on the intelligent supervision scheme to obtain a self-optimized supervision parameter set.

[0006] In a second aspect, the application provides a data mining-based film and television supervision analysis system, which comprises: a collection module configured to collect panoramic data on film and television works of a video platform, and to arrange video, audio and text information in layers through cross-media transcoding processing to obtain a multi-dimensional film and television content dataset; an extraction module configured to extract scene semantic frames of video, multi-level emotion of audio and cultural value trend of text through content feature recognition based on the multi-dimensional film and television content dataset to obtain a time sequence correlation content marking list; a scoring module configured to cross score key elements according to social influence intensity, transmission penetration depth and audience psychological acceptance through hierarchical scoring based on the time sequence correlation content marking list to obtain a dynamic weight content scoring table; an analysis module configured to analyze network transmission path of film and television works through correlation tracking based on the dynamic weight content scoring table to obtain a stereoscopic content influence map; a ranking module configured to rank supervision points according to priority through decision rules based on the stereoscopic content influence map to obtain an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individualized grading requirements; an analysis module configured to analyze the execution effect of supervision measures through effect feedback based on the intelligent supervision scheme to obtain a self-optimized supervision parameter set.

[0007] The third aspect of the present application provides a computer device, comprising: a memory and at least one processor, the memory has instructions stored therein; the at least one processor invokes the instructions in the memory to enable the computer device to perform the data mining-based film and television supervision analysis method described above.

[0008] The fourth aspect of the present application provides a computer-readable storage medium having instructions stored therein, which, when executed on a computer, causes the computer to perform the data mining-based film and television supervision analysis method described above.

[0009] In the technical scheme provided in the present application, the layered standardization of video, audio and text information is realized through panoramic data acquisition and cross-media transcoding processing, which overcomes the problems of data acquisition fragmentation and format inconsistency in traditional film and television supervision, establishes a multi-dimensional film and television content data set, and provides a high-quality data basis for subsequent analysis; through content feature recognition technology, scene semantic extraction is performed on video frames, multi-level emotion recognition is performed on audio, and cultural value tendency analysis is performed on text, which breaks through the limitations of single media analysis, realizes the semantic association of cross-media content, forms a time sequence associated content marking list, and enhances the identification ability of implicit harmful content; based on the cross scoring of key elements according to social influence intensity, transmission penetration depth and audience psychological acceptance, a multi-dimensional evaluation system is innovatively introduced, a dynamic weight content scoring table is generated, and the supervision evaluation is more comprehensive and objective; through the network transmission path analysis of the film and television works by the correlation tracking technology, the spatio-temporal dynamic characteristics of content transmission are revealed, and a stereoscopic content influence map is constructed, which provides a decision basis for accurate supervision; based on the decision rules, the priority of the supervision points is sorted, an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individualized grading requirements is formed, and the optimization allocation and differential processing of supervision resources are realized; through the deviation correction analysis of the execution effect of the supervision measures by the effect feedback mechanism, a self-optimizing supervision parameter set is generated, so that the system has the ability of self-learning and continuous optimization. It is particularly worth emphasizing that the present scheme fully plays the technical contribution of artificial intelligence algorithms and models in specific functions and application fields, such as the deep learning visual recognition model applied in scene semantic extraction, which can accurately identify sensitive content in complex scenes; the sentiment computing algorithm used in multi-level emotion recognition, which can capture the emotional changes implied in the sound; the natural language processing technology introduced in cultural value tendency analysis, which can understand the deep semantics of the text; the network transmission algorithm used in transmission path analysis, which can predict the influence range and transmission trend of the content; the multi-objective optimization algorithm applied in supervision decision, which can balance the supervision effect and resource consumption; the organic combination of these algorithm characteristics makes the present scheme show significant technical advantages in the field of massive multimedia content supervision, which not only improves the accuracy and efficiency of supervision, but also reduces the labor cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 This is a schematic diagram of an embodiment of a film and television supervision analysis method based on data mining in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of a film and television supervision and analysis system based on data mining in an embodiment of the present application; Figure 3 It is a schematic block diagram of the structure of a computer device in an embodiment of the present invention. DETAILED DESCRIPTION

[0012] The embodiments of the present application provide a film and television supervision analysis method, system and storage medium based on data mining. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0013] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the film and television supervision analysis method based on data mining includes: Step S101: Perform panoramic data collection on film and television works on the video platform, and perform hierarchical and standardized organization of video, audio, and text information through cross-media transcoding processing to obtain a multi-dimensional film and television content dataset; Step S102: Based on the multi-dimensional film and television content dataset, scene semantics are extracted from video frames, multi-level emotion recognition is performed on audio, and cultural value tendency analysis is performed on text to obtain a list of time-series related content tags; Step S103, based on the time sequence association content label list, cross-score the key elements according to the social influence strength, the transmission penetration depth and the audience psychological acceptance by hierarchical scoring, to obtain a dynamic weight content score table; Step S104, according to the dynamic weight content score table, the network transmission path of the film and television works is analyzed by associated tracking to obtain a stereoscopic content influence graph; Step S105, according to the stereoscopic content influence graph, the supervision points are prioritized by decision rules to obtain an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individualized grading requirements; Step S106, according to the intelligent supervision scheme, the execution effect of the supervision measures is analyzed by effect feedback to obtain a self-optimizing supervision parameter set.

[0014] It can be understood that the execution subject of the present application can be a film and television supervision analysis system based on data mining, and can also be a terminal or a server, which is not limited here. The server is taken as an example for illustration in the embodiments of the present application.

[0015] Specifically, the panoramic data acquisition technology is used to collect data of the film and television works on the video platform, and the cross-media transcoding processing technology is used to convert the heterogeneous data into a unified standard format according to different formats of video, audio and text data. Specifically, for video data, key frames are extracted and coded to standardize; for audio data, the sampling rate and bit depth are unified; for text data, the coding format is converted and the structure is standardized. These processed data are arranged in a hierarchical structure, and a multi-dimensional film and television content data set is constructed, which contains time axis aligned media content and metadata information.

[0016] According to the multi-dimensional film and television content data set, content feature recognition is performed. Scene semantic extraction technology is used for video frames to identify key elements, scene types and visual semantics in the picture; multi-level emotion recognition technology is used for audio content to analyze features such as tone, rhythm and volume change, and extract emotional tendency; cultural value tendency analysis technology is used for text content to mine the value orientation and cultural position contained therein. The three types of feature data are associated and integrated according to the time stamp to form a time sequence association content label list, which contains the multi-modal content features and labels of each time point of the film and television works.

[0017] Based on the time sequence correlation content label list, the key elements are evaluated in multiple dimensions using a hierarchical scoring technique. The social influence strength score is based on the sensitivity of the content and the social attention heat; the transmission penetration depth score is based on the transmission speed and coverage of the content; and the audience psychological acceptance score considers the psychological feelings and acceptance of the content by different groups. Through cross-scoring method, the three dimensions are weighted and calculated to generate a content score table containing dynamic weight coefficients, realizing the accurate quantitative rating of the film and television content. According to the dynamic weight content score table, the correlation tracking technology is used to analyze the network of the transmission path of the film and television works, track the transmission trajectory and influence range of the content in different platforms, regions and groups. By establishing a content correlation network, the transmission key nodes and transmission path characteristics are identified, combined with time and regional data, a three-dimensional content influence map is constructed, and the distribution and development trend of the social influence of the film and television content are intuitively displayed.

[0018] According to the three-dimensional content influence map, decision rule technology is applied to systematically analyze and prioritize the supervision points. The content compliance of high-risk supervision points is evaluated, and specific modification suggestions are generated; combined with the transmission range and influence data, hierarchical restriction measures are developed; based on the characteristics of the audience groups, personalized hierarchical requirements are generated. These content integrations form an intelligent supervision scheme, providing precise supervision decision support for different types of film and television content. According to the implementation of the intelligent supervision scheme, the effect feedback mechanism is used to collect the execution data of the supervision measures, compare the differences between the expected effect and the actual effect, and analyze the root causes of the deviation. For different deviation factors, parameter adjustment strategies are developed, and after verification, a self-optimized supervision parameter set is formed, realizing the continuous optimization and self-regulation of the supervision system.

[0019] Taking a popular TV series as an example, the video, audio and text data of the series are collected, and the multi-dimensional video content dataset is constructed by transcoding processing. Then the semantic extraction of violent scenes in video frames, multi-level emotion recognition of angry emotions in character dialogues, and cultural tendency analysis of value orientation in dialogue texts are carried out to generate a list of time-correlated content tags. Based on the list, the social impact of violent content is evaluated as high-risk level, the penetration depth of adolescent group is 76%, and the audience psychological acceptance is medium-low, thus generating a dynamic weight content score table. Through correlation tracking, it is found that the violent content spreads in a network on social media, the influence range continues to expand, and a three-dimensional content influence map is formed. The decision rule determines that the content is a high-priority supervision point, and generates precise modification suggestions, gradient restriction measures for different age groups and individualized grading requirements to form an intelligent supervision scheme. After implementation, the feedback effect shows that the exposure rate of the adolescent group has dropped to an acceptable range, but some restriction measures exist bypass phenomenon, after deviation correction analysis, the supervision parameters are adjusted to form a self-optimized supervision parameter set, and effective supervision of the video content is realized.

[0020] In the embodiments of the present application, the layered standardization arrangement of video, audio and text information is realized through panoramic data acquisition and cross-media transcoding processing, which overcomes the problems of data acquisition fragmentation and format inconsistency in traditional film and television supervision, establishes a multi-dimensional film and television content data set, and provides a high-quality data basis for subsequent analysis; through content feature recognition technology, scene semantic extraction is performed on video frames, multi-level emotion recognition is performed on audio, and cultural value tendency analysis is performed on text, which breaks through the limitations of single media analysis, realizes the semantic association of cross-media content, forms a time sequence associated content marking list, and enhances the identification ability of implicit harmful content; based on the grading scoring mechanism, the key elements are cross-scored according to the social influence intensity, the transmission penetration depth and the audience psychological acceptance, a multi-dimensional evaluation system is innovatively introduced, a dynamic weight content scoring table is generated, and the supervision and evaluation are more comprehensive and objective; through the association tracking technology, the reticular transmission path of the film and television works is analyzed, the spatio-temporal dynamic characteristics of content transmission are revealed, a stereoscopic content influence map is constructed, and a decision basis is provided for accurate supervision; based on the decision rule, the priority of the supervision points is sorted, an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individualized grading requirements is formed, and the optimization allocation and differential processing of supervision resources are realized; through the effect feedback mechanism, the deviation correction analysis of the implementation effect of the supervision measures is performed, a self-optimizing supervision parameter set is generated, and the system has the ability of self-learning and continuous optimization. It is particularly emphasized that the present scheme fully plays the technical contribution of artificial intelligence algorithms and models in specific functions and application fields, such as the deep learning visual recognition model applied in scene semantic extraction, which can accurately identify sensitive content in complex scenes; the sentiment computing algorithm used in multi-level emotion recognition, which can capture the emotional changes implied in the sound; the natural language processing technology introduced in cultural value tendency analysis, which can understand the deep semantics of text; the network propagation algorithm used in transmission path analysis, which can predict the influence range and transmission trend of content; the multi-objective optimization algorithm applied in supervision decision, which can balance the supervision effect and resource consumption; the organic combination of these algorithm characteristics makes the present scheme show significant technical advantages in the field of massive multimedia content supervision, which not only improves the accuracy and efficiency of supervision, but also reduces the labor cost.

[0021] In a specific embodiment, the process of performing step S101 can specifically include the following steps: (1) Deep penetration collection of film and television content of multi-source video platforms is performed through an adaptive capture network to obtain full-spectrum film and television original data; (2) Digital fingerprint recognition is performed on the full-spectrum film and television original data, and intelligent detection is performed on the video frame rate, audio sampling and text encoding to obtain a media feature map; (3) According to the media feature map, the heterogeneous media format is processed by spatio-temporal synchronization through cross-coding conversion to obtain a unified standard media stream; (4) The unified standard media stream is subjected to semantic segmentation, and the video, audio and text are precisely sliced through key frame positioning technology to obtain a semantic association slice set; (5) The semantic association slice set is subjected to hierarchical classification through multi-dimensional cross mapping, a content association network is constructed, and a structured content matrix is obtained; (6) The structured content matrix is written into a data warehouse through distributed storage, a multi-dimensional retrieval channel is established, and a multi-dimensional video content data set is obtained.

[0022] Specifically, the video content of the multi-source video platform is deeply penetrated and collected through an adaptive capture network. The adaptive capture network is a distributed crawler cluster technology that dynamically adjusts the crawling strategy according to the data structure characteristics of different video platforms, breaks through the platform anti-crawling mechanism, and realizes effective collection of encrypted content, streaming media and hidden pages. The network is composed of multiple layers of proxy nodes, and through rotating IP, simulating user behavior and parallel task allocation, it can maintain efficient collection while avoiding being blocked by the platform. During the collection process, the network will automatically adjust the request frequency according to the platform response speed, and use differentiated crawling depth for different types of content. The full-spectrum video original data obtained includes video files, audio streams, subtitle texts, comment data and metadata information, etc. The full-spectrum video original data collected is subjected to digital fingerprint identification processing, and unique data features are extracted from video, audio and text content through feature extraction algorithms. Video frame rate detection calculates the frame rate of the original video by analyzing the inter-frame time interval; audio sampling identification determines the sampling rate, bit depth and channel configuration of the audio through frequency spectrum analysis; and text encoding detection identifies the encoding format of the text through character set feature recognition. These parameters form the feature identifier of each media file, which is constructed into a media feature map.

[0023] According to the media feature map, further through the cross coding conversion of heterogeneous media format spatio-temporal synchronization processing. Cross coding conversion will be different sources of media files into a unified standard format. Video transcoding adopts H.264 / AVC coding standard, with 25fps frame rate and 1080p resolution as the benchmark; audio transcoding is unified as 44.1kHz sampling rate, 16-bit depth AAC format; text content is converted into the standard text format of UTF-8 encoding. Spatio-temporal synchronization processing ensures that audio and video are accurately aligned with text on the time axis, generating a unified standard media stream. The unified standard media stream is subjected to semantic segmentation processing, and key frame positioning technology is applied to video, audio and text content slicing. Video key frame positioning is based on shot transition detection algorithm, which identifies scene change points by analyzing the visual difference between adjacent frames; audio segmentation is based on acoustic feature change and silence detection, which identifies the natural separation point of sound content; text segmentation is based on semantic units and syntax structure, which divides continuous text into meaningful sentence paragraphs. Each media slice is associated and matched according to the timestamp to form a semantic association slice set, each slice containing video, audio and text content in the same time period.

[0024] The semantic association slice set is classified by multi-dimensional cross mapping technology, which establishes the association relationship model between different dimensional data. According to the content theme, emotional tendency, character object and other dimensions, the slices are marked with features; then through similarity calculation, the slices with similar features are clustered and grouped; the association between slices is established to form a multi-level content association network. The content association network reflects the logical structure and semantic relationship of the film and television content, which is expressed in matrix form to form a structured content matrix. The structured content matrix is written into the data warehouse through the distributed storage mechanism, and the large-scale data is stored in multiple nodes by sharding technology, and multi-dimensional retrieval channels are established to support efficient query. During storage, an index structure based on content features, timestamps, emotional tendencies and other dimensions is established to support multi-angle, multi-level data retrieval and analysis, forming a complete multi-dimensional film and television content data set, providing a data basis for subsequent supervision and analysis.

[0025] Taking a popular network drama as an example, the adaptive capture network captures the distribution of the drama on three major video platforms, sets up a targeted crawling strategy, and collects a full spectrum of original video data including 100 episodes, 50 behind-the-scenes videos, and 3000 user comments. Subsequently, digital fingerprinting is performed to detect that the video format includes MP4 and FLV, the frame rate is 24fps and 30fps, the audio sampling rate is 44.1kHz and 48kHz, and the text encoding includes UTF-8 and GBK, generating a media feature map. Based on the feature map, cross-encoding conversion is performed to convert all videos to H.264 encoding, 25fps frame rate MP4 format, audio to 44.1kHz, 16-bit AAC format, and text to UTF-8 encoding, achieving temporal and spatial synchronization to obtain a unified standard media stream. Then semantic segmentation is performed, 1200 scene transition points are identified from the video through key frame algorithm, 850 sound change points are identified through audio analysis, and 3500 semantic units are extracted through text analysis, forming a semantic association slice set according to the timestamp matching. Through multidimensional cross mapping, the slices are classified, and a content association network is constructed according to the three dimensions of "plot development-character interaction-emotional change", and a structured content matrix is generated. The matrix data is stored in multiple nodes of the data warehouse, a retrieval channel based on content type, time sequence, and semantic association is established, and a multidimensional video content dataset is generated, laying a foundation for content supervision and analysis of the popular drama.

[0026] In a specific embodiment, the process of performing step S102 can specifically include the following steps: (1) Extracting video key frame sequences from the multidimensional video content dataset, performing edge detection and region division on video pictures through scene segmentation technology to obtain video scene combination features; (2) Classifying theme elements using a semantic understanding engine on the video scene combination features to extract key visual elements and context relationships in the scene, obtaining a scene semantic mapping set; (3) Separating audio data streams from the multidimensional video content dataset, performing waveform decomposition on audio emotional features through spectral analysis to obtain emotional audio feature spectra; (4) Multi-level labeling of emotional audio feature spectra, emotional classification of tone, speech rate, and rhythm changes through emotional intensity gradient quantization, obtaining multi-level emotional labeling data; (5) Extracting text content from the multidimensional video content dataset, extracting theme words and value orientation analysis of the text through a semantic network to obtain cultural value tendency indicators; (6) Integrating and associating the scene semantic mapping set, multi-level emotional labeling data, and cultural value tendency indicators according to the time axis to establish the time sequence mapping relationship between content elements, obtaining a time sequence associated content label list.

[0027] Specifically, representative frames are selected from the standardized video data. The extraction process combines frame difference method and edge energy threshold, calculates the pixel difference value between adjacent frames, marks the scene cut point when the difference value exceeds the preset threshold, and generates a key frame set. After obtaining the key frame, edge detection and region division are performed on the picture through scene segmentation technology, which specifically includes: using Canny edge detection algorithm to identify the object contour lines in the picture; using region growing method to combine similar texture and color pixel groups into semantic regions; related regions are combined according to spatial adjacency and feature similarity through hierarchical clustering algorithm to form video scene combination features, which include the spatial distribution, shape features and visual attribute information of the main objects in the picture. The video scene combination features are classified by a semantic understanding engine to establish a visual element dictionary containing basic element categories such as characters, scenes and objects; then the visual regions in the picture are mapped to the element categories in the dictionary through feature matching to identify the specific content objects contained in the scene; then the spatial position relationship and interaction state between the objects are analyzed to extract the context correlation between the elements; scene semantic description labels are generated to construct a scene semantic mapping set, which records the main visual elements, element categories, element combination relationships and overall semantic expressions contained in each scene, providing a semantic understanding basis for visual content for subsequent supervision and judgment.

[0028] The audio data stream is separated from the multi-dimensional video content data set, and independent sound information is obtained by extracting the audio channel. The audio data is subjected to frequency spectrum analysis, and short-time Fourier transform (STFT) is performed to convert the time domain signal into time-frequency domain representation; then MFCC (Mel frequency cepstrum coefficient) features are extracted, which simulate the perception characteristics of human ears to sound; then spectral analysis is performed to identify the energy distribution, tone change and rhythm features in the audio; through waveform decomposition technology, complex audio is separated into components of different frequency bands to form an emotional audio feature spectrum, which contains the frequency distribution, energy intensity and modulation characteristics of the audio signal in different time segments, providing basic data features for emotion recognition. The emotional audio feature spectrum is subjected to multi-level labeling processing to establish an emotional acoustic parameter model, which contains the mapping relationship between key acoustic features such as tone height, speech rate, volume size and rhythm change and emotional state; then the parameters in the audio feature spectrum are quantitatively calculated to extract the emotional expression features; then through emotional intensity gradient quantization technology, the emotional intensity is divided into multiple levels, such as low intensity, medium intensity and high intensity; according to the acoustic feature combination mode, the audio segment is labeled with emotional categories, such as basic emotional types of anger, joy, sadness and calmness, to obtain multi-level emotional labeling data, which contains emotional categories, emotional intensity levels and emotional change trend information of each time segment of the audio.

[0029] Text content is extracted from multi-dimensional video content data set, including dialogue subtitles, voice-over commentary and other text information. The text is analyzed by semantic network technology to perform Chinese word segmentation and part-of-speech tagging, and identify the basic language units in the text. Then, the importance weight of the words is calculated by TF-IDF algorithm to extract the key theme words in the text. Then, a word co-occurrence network is established to analyze the semantic association strength between words. A value orientation dictionary is introduced to analyze the value orientation of the text content, identify the positive values, negative values or neutral expressions contained therein, and form a cultural value tendency index. The index quantifies the expression strength and orientation tendency of various value concepts in the text content.

[0030] The scene semantic mapping set, multi-level emotional labeling data and cultural value tendency index are integrated and associated according to the time axis to establish a unified time reference, ensuring the alignment of the three data in the time dimension. Then, the multi-modal data in the same time window is associated and matched by time window sliding method. Then, a time sequence association matrix between content elements is constructed to record the mutual influence relationship between visual, auditory and text content. A time sequence association content labeling list is generated, which records the comprehensive content characteristics of each segment of the video work in chronological order, including visual scene description, emotional expression state and cultural value orientation.

[0031] Taking a certain animation film with a youth theme as an example, the key frame sequence is extracted from the multi-dimensional video content data set, and the character image, environment scene and interactive action in the picture are identified by scene segmentation to generate video scene combination features. These features are analyzed by semantic understanding to identify key visual elements such as "fighting scene" and "weapon use", forming a scene semantic mapping set. Meanwhile, the frequency spectrum features are extracted from the audio to analyze the characteristics of high pitch and fast rhythm in the sound, generating emotional audio feature spectrum. These audio features are labeled as "moderate tension", "slight fear" and other emotional categories by emotional intensity quantification, forming multi-level emotional labeling data. Key words are extracted from the subtitle text to analyze the value-oriented expressions such as "courage" and "justice" contained therein, generating a cultural value tendency index. The three kinds of data are integrated according to the time axis, and it is found that in a certain fighting scene, the visual presents intense fighting, the audio expresses tension, and the text emphasizes the value of "fighting for justice". The comprehensive analysis of these multi-modal data forms a time sequence association content labeling list.

[0032] In a specific embodiment, the process of performing step S103 can specifically include the following steps: (1) Element screening is performed on the time sequence association content labeling list to separate sensitive elements and core content in the video content by key point extraction technology to obtain a content key element set; (2) Quantify and score the key elements of the content according to the social impact intensity standard, and evaluate the social sensitivity and guiding role of the content elements with reference to public opinion to obtain the social impact intensity index; (3) Construct a communication link model based on the key elements of the content, calculate the diffusion speed and coverage of the content among different groups through user behavior analysis, and obtain the communication penetration depth value; (4) Measure psychological acceptance based on the key elements of the content, analyze the psychological acceptance of people of different age groups and cultural backgrounds through audience portrait matching, and obtain the psychological acceptance level; (5) Perform multi-dimensional cross-correlation on the social impact intensity index, communication penetration depth value and psychological acceptance level, and dynamically weight the three indicators through the weight distribution matrix to obtain comprehensive score data; (6) The comprehensive score data is hierarchically mapped, and the content score is segmented by threshold adjustment to establish a variable weight scoring system and obtain a dynamic weight content score table.

[0033] Specifically, key point extraction technology is used to identify and separate sensitive elements from core content in film and television content. This technology employs a variant of the TF-IDF algorithm, weighting each content element in the time series markers based on their frequency of occurrence and discriminability to identify key content points of regulatory significance. This process calculates the characteristic weight of each content element and ranks them according to their importance within the overall content. A threshold is then set and elements with weights exceeding the threshold are extracted to form a set of key content elements. This set of key content elements includes video scenes, audio emotions, and text expressions requiring specific supervision. These elements are assigned unique identifiers for subsequent scoring and tracking. The set of key content elements is quantitatively scored based on the intensity of social impact, which refers to the degree to which content can lead, stimulate, or influence social groups. This quantification process utilizes social public opinion reference technology, matching content elements with cases in a historical public opinion database to analyze the intensity of social reactions similar content has elicited in the past. The assessment dimensions include sensitivity, controversy, and guidance. Each dimension is scored on a five-point scale, and a weighted average is used to calculate the social impact intensity index. The index reflects the extent to which content elements may affect social values, public order or group behavior, providing an important reference for subsequent content classification.

[0034] Based on the key elements of content, a propagation link model is constructed to analyze the propagation characteristics of content among different groups of people. The propagation link model is based on information diffusion theory and uses user behavior analysis technology to mathematically model the content propagation process. The model calculation process can be expressed as: in, Indicates the propagation penetration depth value; Indicates the number of user groups; represents the weight coefficient of the g-th group; represents the group influence coefficient; represents the number of users in the gth group; represents the communication ability index of the i-th user; represents the time decay coefficient of the gth population; Indicates the duration of transmission; Indicates the number of communication channels; represents the influence coefficient of the j-th channel; represents the coverage area of ​​the jth channel; represents the maximum channel influence coefficient; Represents the total target coverage area. This formula takes into account the characteristics of different groups, user communication behavior, time decay factors, and communication channel effects to calculate the content's penetration depth, reflecting the content's diffusion speed and coverage.

[0035] Psychological acceptance is measured based on a set of key content elements, employing audience profile matching technology to analyze the psychological acceptance levels of people from different age groups and cultural backgrounds. This psychological acceptance measurement constructs a multi-dimensional audience profile library, encompassing psychological characteristic models for groups across age groups, cultural backgrounds, and educational levels. The key content elements are then matched to each group profile to assess the potential psychological impact of the content on different groups. Based on these matching results, acceptance levels are typically categorized into four levels: high, moderate, low, and unsuitable. This psychological acceptance rating provides an audience-oriented reference for regulatory decision-making, helping to determine the appropriate audience for the content. The social impact intensity index, communication penetration value, and psychological acceptance rating are cross-correlated across multiple dimensions, and a weighting matrix is ​​used to dynamically weight these three indicators. This weighting matrix dynamically adjusts the weights of each indicator based on content type, dissemination environment, and regulatory priorities, ensuring that the scoring results are more aligned with actual regulatory needs. The weighting calculation utilizes a weighted geometric mean method, comprehensively considering the interactions among the three indicators to generate a comprehensive score. The data has multidimensional attributes. It not only contains the overall score value, but also retains the independent score and weight information of each dimension, providing comprehensive data support for subsequent grading processing.

[0036] The comprehensive score data is classified and mapped, and the content score is segmented by threshold adjustment technology. The classification mapping defines different levels of score intervals, including categories suitable for all ages, suitable for teenagers, adult supervision and restricted levels, etc.; then the threshold range of each level is dynamically adjusted according to the content characteristics, to ensure that the classification result meets the current regulatory standards; a variable weight scoring system is established, and different scoring strategies are set for different types of content to obtain a dynamic weight content score table. The score table assigns detailed rating information to each content element, including suitable audience, restriction requirements and adjustment suggestions, providing direct basis for subsequent regulatory decisions.

[0037] Taking a certain live reality show as an example, the element screening is performed on the time sequence associated content label list, and sensitive elements such as "high-intensity competition between contestants", "harsh comments from judges" and "emotional breakdown of participants" are identified to form a set of key content elements. The social influence intensity of these elements is evaluated, and by comparing the social discussion heat caused by similar content in history, the sensitivity of the "emotional breakdown" scene is rated as level 4 (five-level scale), the controversiality is rated as level 3, the guidance is rated as level 4, and the social influence intensity index is calculated as 3.8. The propagation link model is constructed to analyze the propagation characteristics of the content in the youth group. Considering the high activity of the group (γ = 0.85) and the high propagation coefficient of the short video platform (φ = 0.92), the propagation penetration depth value is calculated as 0.73 by applying the above formula. The psychological acceptance is measured, and it is found that the matching degree of the content to the 12-16 year-old youth group is low, and the psychological acceptance is rated as "low acceptance" level. The three index data are input into the weight distribution matrix, considering that the program mainly faces the youth group, the psychological acceptance index is given a high weight of 0.5, and the social influence index and the propagation depth are given weights of 0.3 and 0.2 respectively, and the comprehensive score is calculated as 3.62. Through threshold mapping, the score falls into the "caution for teenagers" interval (3.5-4.0), and is marked as the "parental guidance viewing" category in the dynamic weight content score table, and specific modification suggestions are generated, including reducing the display time of the emotional breakdown scene and increasing the positive guiding content.

[0038] In a specific embodiment, the process of performing step S104 can specifically include the following steps: (1) Extract high-weight content tags from the dynamic weight content score table, and perform matching analysis on the associated content in different video works by similarity calculation to obtain a content association link library; (2) Construct a video work time axis propagation sequence according to the content association link library, track the evolution trend of the content in different periods through time sequence data association, and obtain a time sequence propagation trajectory diagram; (3) Spatial dimension expansion of the time-series propagation trajectory map, visualization of the content propagation density and coverage range in different regions through regional distribution mapping, and obtaining regional propagation heat distribution; (4) Network topology construction of the content association link library according to the propagation relationship, identification of the key nodes and propagation paths of the content propagation through node connection analysis, and obtaining the network propagation relationship network; (5) Influence evaluation of the network propagation relationship network, quantitative analysis of the radiation range and influence degree of each node in the propagation network through influence diffusion calculation, and obtaining content influence evaluation data; (6) Multidimensional integration of the time-series propagation trajectory map, regional propagation heat distribution and content influence evaluation data, construction of a three-dimensional propagation influence model, and obtaining a three-dimensional content influence map.

[0039] Specifically, the high-weight content label refers to content elements with a weight value exceeding a preset threshold, which generally represent content features with high regulatory value. The extraction process sorts the dynamic weight content score table, rearranges the weight values in descending order, and then sets a filtering threshold, generally taking the top 30% of the weight values as high-weight labels. The high-weight labels are extracted, and their complete attribute information is retained. After obtaining the high-weight content labels, the associated content in different film and television works is matched and analyzed through similarity calculation. The similarity calculation uses the cosine similarity algorithm to calculate the cosine value between the feature vectors of the high-weight labels in different works. The closer the value is to 1, the higher the content similarity. The association relationship between the contents is constructed through the similarity matrix to form a content association link library, which records the association degree and type of content elements between different film and television works. The film and television work time axis propagation sequence is constructed according to the content association link library. This step arranges the content propagation process in chronological order through time-series data association technology. The time-series data association collects the time information of content publishing and propagation from various platforms; then arranges these time points in chronological order to form a propagation timeline; then maps the content association link information onto the timeline to mark the time nodes of content appearance, propagation peak and decay; through time window sliding analysis, the propagation heat change in each time period is calculated to generate a continuous propagation trend curve. These data integration forms a time-series propagation trajectory map, which visually displays the propagation rules and evolution trend of the content in the time dimension, providing a time reference for subsequent spatial analysis.

[0040] The time-series propagation trajectory graph is expanded in spatial dimension, and the propagation data is visualized in geographical space through regional distribution mapping technology. The regional distribution mapping divides the geographical grid of the target area, which is generally divided according to administrative division or population density. Then, the content propagation data of each region is collected, including the number of views, shares and interactions. Then, these data are mapped to the geographical grid according to the location information. Through the heat rendering algorithm, the data-intensive areas are highlighted to generate a propagation heat map. This processing forms the regional propagation heat distribution, which intuitively shows the propagation density and coverage of the content in different geographical areas, and reveals the propagation differences between regions and the distribution of hot regions. The content association link library is constructed into a network topology according to the propagation relationship. This process uses complex network modeling technology, and the content elements are regarded as network nodes, and the propagation association between the contents is regarded as network connection. The network topology construction defines the node attributes, including content type, influence level and timestamp. Then, the connection weight is defined according to the association strength, and the higher the strength, the greater the weight. Then, the connection is established according to the actual propagation relationship between nodes to form the initial network structure. Through node connection analysis technology, the network characteristics such as node connectivity, centrality and clustering coefficient are calculated to identify the key nodes and main propagation paths in the network. This processing forms a network propagation relationship network, which reveals the structural characteristics and key influence points of content propagation.

[0041] The influence of the network propagation relationship network is evaluated, and the radiation range and influence degree of each node are quantitatively analyzed through influence diffusion calculation. The influence diffusion calculation is based on the information cascade model, and its core formula can be expressed as: wherein, represents the influence index of node v; is the node inherent attribute weight; represents the content weight value of node v; is the association transmission weight; N(v) is the node set directly connected to node v; represents the connection weight of node u to v; represents the out-degree of node u; is the attenuation coefficient; is the network distance between nodes; is the content feedback weight; represents the user feedback intensity obtained by node v; is the maximum feedback intensity value; K is the number of influence factors; represents the correction coefficient of the jth external influence factor to node v. represents the upstream node directly connected to node The influence index of the formula. The formula comprehensively considers the node inherent properties, network location, connection relationship and user feedback and other factors, and comprehensively evaluates the influence of the node in the propagation network. By applying the calculation model to each node, content influence evaluation data is generated.

[0042] The time sequence propagation trajectory diagram, regional propagation heat distribution and content influence evaluation data are multi-dimensionally integrated to construct a propagation influence stereoscopic model. The multi-dimension integration adopts data fusion technology to unify the data in different dimensions to the same coordinate system. The integration process establishes a three-dimensional coordinate system of time, space and network influence; then maps the data in each dimension to the corresponding coordinate axis; then fills the data blank points through interpolation algorithm to ensure data continuity; constructs a stereoscopic data visualization model to generate a stereoscopic content influence map. The map realizes the comprehensive display of the three dimensions of time, space and influence.

[0043] In a specific embodiment, the process of performing step S105 can specifically include the following steps: (1) Extract high-risk supervision points from the stereoscopic content influence map, identify and filter the high-risk supervision points through risk weight calculation to obtain a key supervision point list; (2) Hierarchical priority sorting of the key supervision point list, determination of the processing order and resource allocation proportion of different supervision points through emergency assessment, to obtain a supervision priority sequence list; (3) Based on the supervision priority sequence list, for high-priority content nodes, precise positioning of the difference between the illegal content and the compliance standard through content comparison analysis, to obtain a precise modification suggestion list; (4) According to the content propagation characteristics in the key supervision point list, formulating a hierarchical restriction scheme for contents with different influence ranges through gradient control strategy, to obtain a gradient restriction measure scheme; (5) According to the content features and audience group information of the key supervision point list, precise definition of the content adaptation population range through hierarchical standard mapping, to obtain a personalized hierarchical requirement table; (6) Integrating and correlating the precise modification suggestion list, the gradient restriction measure scheme and the personalized hierarchical requirement table to obtain an intelligent supervision scheme.

[0044] Specifically, risk assessment and screening of content nodes are performed by risk weight calculation technology. The risk weight calculation sets multi-dimensional risk assessment indicators for each content node in the stereoscopic content influence graph, including content sensitivity, propagation speed, coverage population characteristics, influence intensity, etc.; then sets weight coefficients for each dimension, differentiates and emphasizes risk factors of different types of content; then calculates the comprehensive risk score, sums up the scores of each dimension after multiplying by the corresponding weight; sets a risk threshold, screens out content nodes with risk scores exceeding the threshold, and forms a high-risk supervision point set. Further analyze the network location and influence range of these high-risk points, focus on identifying nodes that are in the core position of the propagation network and have extensive influence, and generate a key supervision point list containing content nodes that need to be prioritized and their risk feature descriptions. The key supervision point list is prioritized by level, and an emergency assessment technique is used to scientifically sort the processing order of different supervision points. The emergency assessment establishes a multi-factor scoring standard, considering the timeliness of the content, the speed of influence diffusion, the potential harm degree, and the resource consumption; then scores each supervision point according to the scoring standard to form an emergency quantitative value; then sorts from high to low according to the emergency, and divides into four levels: emergency processing, priority processing, regular processing, and low priority; allocate processing resources according to the sorting results to determine the processing time limit and resource input proportion of each supervision point. This processing forms a supervision priority sequence list.

[0045] Based on the supervision priority sequence list, content comparison analysis is performed on high-priority content nodes. This process identifies the differences between illegal content and compliance standards through precise comparison technology. Content comparison analysis constructs a supervision standard template library covering supervision requirements and compliance standards for different content types; then compares high-priority content nodes with corresponding standard templates to identify violations and non-compliant content; then analyzes each violation point in depth to determine the violation degree, violation type, and adjustment direction; generates specific modification suggestions, including deleting content, replacing expressions, adjusting scenes, and other specific operation guidelines. This processing forms a precise modification suggestion list, providing clear compliance adjustment guidance for content producers and improving the precision and operability of supervision. According to the content propagation characteristics in the key supervision point list, a gradient control strategy technology is used to develop a hierarchical restriction scheme for content with different influence ranges. The gradient control strategy divides the control levels according to the content influence range, usually into four levels: comprehensive restriction, partial restriction, guided viewing, and unrestricted; then designs differentiated restriction measures for each level, including playback permission control, time limit, region restriction, user screening, etc.; then matches each supervision point with the most suitable restriction level and specific measures; forms a complete gradient restriction measure scheme, providing targeted control strategies for different characteristics of content, achieving fine-grained supervision.

[0046] Based on the content characteristics and audience information in the key regulatory points list, a grading standard mapping technique is used to precisely define the appropriate audience for the content. This grading standard mapping establishes a content grading system, including multiple levels such as Suitable for All Ages, Parental Guidance, Caution for Teens, and Adults Only. The system then analyzes the match between content characteristics and audience psychological characteristics, assessing the receptiveness of different age groups and cultural backgrounds to the content. Based on these matching results, the most appropriate audience range is assigned to each content node. A personalized grading requirement table is generated, clearly defining the appropriate audience, viewing conditions, and prompt requirements for each type of content, providing scientific guidance for user selection. The precise modification suggestion list, the gradient restriction measure plan, and the personalized grading requirement table are integrated and linked to form a complete intelligent regulatory solution. This integration and linking process structured the three lists, ensuring uniform data format and standardized fields. A mapping relationship is then established between the three lists, ensuring consistency in the modification suggestions, restrictions, and grading requirements for each regulatory point. The integrated plan is then prioritized according to regulatory priorities to form a well-organized implementation plan. A complete intelligent regulatory solution document is generated, including regulatory objectives, specific measures, implementation timelines, and expected results, providing comprehensive guidance for regulatory implementation.

[0047] Taking a popular emotional variety show as an example, risk nodes were extracted from its three-dimensional content impact map, revealing high-risk content features such as "excessive arguments among guests," "disclosure of emotional privacy," and "value orientation bias." A risk weighting calculation determined that "value orientation bias," considering its high content sensitivity, strong reach, and reach among youth groups, resulted in a comprehensive risk score of 1.8 times the risk threshold, identifying it as a critical regulatory point and adding it to the list of key regulatory points. An urgency assessment was conducted on the list, and given the program's peak broadcast period and high social media buzz, "value orientation bias" was assigned an urgent level of action and placed at the top of the regulatory priority list. A comparative content analysis compared the program's expressions of values ​​such as "money first" and "gamification of relationships" with the "positive and healthy emotional outlook" in the regulatory standard template. Six core problematic paragraphs were pinpointed and revised, including specific guidance such as "deleting guests' comments on judging love based on material conditions" and "increasing expert commentary on positive emotional guidance," resulting in a precise list of recommended revisions. Based on the program's dissemination characteristics, it was found that it spread quickly among young people and had a strong tendency to imitate. A gradient control strategy was adopted to divide it into the "guided viewing" level, and restrictive measures such as "adding opening prompts" and "setting broadcast time after 22:00" were formulated to form a gradient restriction measure plan. Through the mapping of classification standards, the program was defined as "suitable for audiences over 15 years old, 12-15 years old need to be accompanied by parents" and the classification reasons and precautions were detailed in the personalized classification requirements table. The three lists were integrated and linked to generate a complete intelligent supervision plan that includes modification requirements, broadcast restrictions and age ratings. It provides program producers with a clear direction for compliance adjustments and sets specific restrictions for the broadcast platform, realizing precise content supervision of the variety show.

[0048] In a specific embodiment, the process of executing step S106 may specifically include the following steps: (1) Implement and deploy the intelligent supervision plan, collect and record the time nodes and effect data of the implementation process of the measures through execution tracking, and obtain the supervision execution data flow; (2) Extract key effect indicators from the regulatory execution data stream, quantitatively compare the differences between expected regulatory effects and actual execution results through target deviation measurement, and obtain regulatory effect deviation data; (3) Conduct root cause analysis on the regulatory effect deviation data, identify and locate the key factors and influencing conditions that cause the deviation through tracing back to the source, and obtain a deviation factor attribution table; (4) Formulate a correction strategy based on the deviation factor attribution table, determine the correction direction and adjustment range of different deviation factors through correction intensity calculation, and obtain the regulatory parameter adjustment plan; (5) Optimize and verify the regulatory parameter adjustment plan, test and evaluate the effectiveness of the adjusted regulatory parameters through small-scale experiments, and obtain a parameter optimization verification report; (6) Perform parameter extraction and analysis on the effective parameters in the parameter optimization verification report to obtain the self-optimization supervision parameter set.

[0049] Specifically, the implementation process of regulatory measures is monitored throughout the entire process through execution tracking technology. Execution tracking breaks down the intelligent regulatory plan into specific execution tasks, setting the responsible entity, completion deadline, and acceptance criteria for each task; then, regulatory execution data collection points are established, and data collection triggers are set at key time nodes to record the execution status, progress, and effects of the tasks; then, through a multi-channel data collection mechanism, execution effect data is collected from multiple dimensions such as content platforms, user feedback, and social public opinion; these collected time series data are organized and stored in a unified format to form a complete regulatory execution data stream. This data stream contains data records of the entire process from the issuance to the implementation of regulatory measures, extracts key effect indicators from the regulatory execution data stream, and uses target deviation measurement technology to accurately compare expected effects with actual execution results. Key performance indicator extraction defines the evaluation dimensions of regulatory effectiveness, including compliance, dissemination control effectiveness, user feedback indicators, and other aspects. Data items related to these evaluation dimensions are then filtered from the regulatory execution data stream to extract core indicator values. These actual indicator values ​​are then compared with the target values ​​preset in the intelligent regulatory solution to calculate the deviation rate and direction. Deviation data for each dimension is structured and organized to form regulatory effect deviation data. This deviation data intuitively demonstrates the differences in the effectiveness of regulatory implementation and provides accurate data basis for subsequent optimization and adjustment.

[0050] Root cause analysis is performed on the regulatory effect deviation data, and the key factors causing the deviation are further explored through traceability technology. Root cause analysis sorts the deviation data according to the deviation degree, and prioritizes the analysis of indicators with larger deviation. Then, the "5W2H" method is used to decompose the problem of each significant deviation, and the specific performance and impact range of the deviation are clarified. Then, through cause-effect chain analysis, the cause chain of the deviation is traced layer by layer, and the root cause is identified. The identified key factors are classified and arranged according to the impact dimension and impact degree, and a deviation factor attribution table is constructed. The attribution table clearly presents the factors causing the regulatory effect deviation and their action mechanisms. Based on the deviation factor attribution table, a correction strategy is developed, and the adjustment scheme for different deviation factors is scientifically planned through correction effort calculation technology. The correction strategy prioritizes the factors in the attribution table, determining the primary and secondary order of adjustment. Then, for each high-priority factor, targeted adjustment measures are designed, and the direction and means of adjustment are clarified. Then, through correction effort calculation, the adjustment amplitude is determined according to the degree of deviation and the impact weight of the factor, avoiding excessive or insufficient adjustment. The adjustment measures are integrated to form a complete regulatory parameter adjustment scheme. The scheme includes specific parameter correction values, adjustment steps, and expected effects.

[0051] Optimization verification is performed on the regulatory parameter adjustment scheme, and the effectiveness of the adjusted parameters is scientifically evaluated through small-scale test technology. Optimization verification selects representative content samples and user groups to construct a test environment. Then, the adjusted regulatory parameters are applied to the test samples to simulate the actual regulatory scenario. Then, multiple detection points are set to collect effect data during the test process, including user response, content change, and propagation effect. Comprehensive analysis of the test data evaluates the effectiveness and feasibility of parameter adjustment, forming a parameter optimization verification report. The report details the test results of each parameter adjustment.

[0052] Effective parameters in the parameter optimization verification report are integrated and refined to construct a self-improving regulatory parameter system through parameter extraction analysis technology. Parameter extraction analysis filters the parameters that can effectively improve the regulatory effect from the verification report, and retains those parameters. Then, these effective parameters are integrated with the original parameters to update the regulatory parameter library. Then, a correlation model between parameters is constructed to clarify the synergistic effect and constraint relationship between different parameters. A complete self-optimizing regulatory parameter set is formed, which has dynamic adjustment capability and can automatically optimize parameter configuration according to the actual regulatory effect, realizing the continuous improvement of the regulatory system.

[0053] Taking a network animation drama as an example, after the implementation and deployment of the intelligent supervision scheme, the execution data is collected at key nodes such as platform online, user interaction and social communication through the execution tracking technology, the completion of the modified content, the execution state of the restriction measures and the user viewing data are recorded, and a complete supervision execution data flow is formed. From the data flow, key effect indicators are extracted, including the compliance rate of violent scene modification, the success rate of access control for minors and the proportion of negative discussions on social media, and the expected target is compared to find that the success rate of access control for minors is 20 percentage points lower than the expected target, forming a supervision effect deviation data. In view of this significant deviation, root cause analysis is carried out, and through traceability tracking, three key factors are identified, including the imperfect age verification bypass mechanism, the insufficient platform execution and the unclear supervision prompt information, and a detailed deviation factor attribution table is constructed. Based on the attribution table, a correction strategy is developed, a double verification process is designed for the age verification mechanism, a reward and punishment mechanism is added to the platform execution, and the supervision prompt information is improved in user-friendly manner, forming a supervision parameter adjustment scheme. The scheme is optimized and verified, and three popular episodes of the animation drama and multiple age groups of test user groups are selected for small-scale testing to detect the effect of the adjusted supervision parameters, and it is found that the double verification mechanism reduces the bypass rate of minors by 85%, and the optimized prompt information has higher user acceptance, forming a detailed parameter optimization verification report. From the report, effective parameters are extracted, including the optimized age verification parameter configuration, the platform execution supervision threshold and the user-friendly prompt template, which are integrated with the original supervision parameters to form a self-optimizing supervision parameter set with self-adjusting ability, providing a parameter basis for the long-term supervision of this type of animation content.

[0054] The above describes the method for analyzing the supervision of films and television based on data mining in the embodiments of the present application, and the following describes the system for analyzing the supervision of films and television based on data mining in the embodiments of the present application. Please refer to Figure 2 An embodiment of the system for analyzing the supervision of films and television based on data mining in the embodiments of the present application includes: A collection module is configured to collect panoramic data of films and television works on a video platform, and to perform hierarchical standardization processing on video, audio and text information through cross-media transcoding to obtain a multi-dimensional film and television content data set; An extraction module is configured to perform scene semantic extraction on video frames, multi-level emotion recognition on audio and cultural value tendency analysis on text based on the multi-dimensional film and television content data set through content feature recognition to obtain a time sequence associated content marking list; A scoring module is configured to perform cross-scoring on key elements according to social influence intensity, transmission penetration depth and audience psychological acceptance degree based on the time sequence associated content marking list through hierarchical scoring to obtain a dynamic weight content scoring table; The analysis module is configured to perform deviation correction analysis on the execution effect of the supervision measures according to the intelligent supervision scheme through effect feedback, and obtain a self-optimizing supervision parameter set. The sorting module is configured to perform priority sorting on the supervision points according to the stereoscopic content influence graph through a decision rule, and obtain an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individual grading requirements. The analysis module is configured to perform deviation correction analysis on the execution effect of the supervision measures according to the intelligent supervision scheme through effect feedback, and obtain a self-optimizing supervision parameter set.

[0055] Through the cooperation of the above components, the layered standardized arrangement of video, audio and text information is realized through panoramic data acquisition and cross-media transcoding processing, which overcomes the problems of data acquisition fragmentation and format inconsistency in traditional film and television supervision, establishes a multi-dimensional film and television content data set, and provides a high-quality data basis for subsequent analysis; through the content feature recognition technology, scene semantic extraction is performed on the video frame, multi-level emotion recognition is performed on the audio, and cultural value tendency analysis is performed on the text, which breaks through the limitations of single media analysis, realizes the semantic association of cross-media content, forms a time sequence associated content marking list, and enhances the identification ability of implicit harmful content; based on the grading scoring mechanism, the key elements are cross scored according to the social influence intensity, the transmission penetration depth and the audience psychological acceptance, a multi-dimensional evaluation system is innovatively introduced, a dynamic weight content score table is generated, and the supervision and evaluation are more comprehensive and objective; through the association tracking technology, the network transmission path of the film and television works is analyzed, the space-time dynamic characteristics of content transmission are revealed, a three-dimensional content influence map is constructed, and a decision basis is provided for accurate supervision; based on the decision rule, the priority of the supervision points is sorted, an intelligent supervision scheme containing accurate modification suggestions, gradient restriction measures and individualized grading requirements is formed, and the optimization configuration and differential processing of supervision resources are realized; through the effect feedback mechanism, the deviation correction analysis of the execution effect of the supervision measures is performed, a self-optimizing supervision parameter set is generated, and the system has the ability of self-learning and continuous optimization. It is particularly worth emphasizing that the scheme fully plays the technical contribution of artificial intelligence algorithms and models in specific functions and application fields, such as the deep learning visual recognition model applied in scene semantic extraction, which can accurately identify sensitive content in complex scenes; the sentiment computing algorithm used in multi-level emotion recognition, which can capture the emotional changes implied in the sound; the natural language processing technology introduced in cultural value tendency analysis, which can understand the deep semantics of the text; the network propagation algorithm used in transmission path analysis, which can predict the influence range and transmission trend of the content; the multi-objective optimization algorithm applied in supervision decision, which can balance the supervision effect and resource consumption; the organic combination of these algorithm characteristics makes the scheme show significant technical advantages in the field of massive multimedia content supervision, which not only improves the accuracy and efficiency of supervision, but also reduces the labor cost.

[0056] Reference Figure 3 In the embodiment of the present application, a computer device is also provided, which can be a server, and its internal structure can be as shown in Figure 3The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store corresponding data in the embodiment. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0057] Those skilled in the art can understand that, Figure 3 The structure shown in the above embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied.

[0058] The computer readable storage medium in the embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0059] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to the memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. The non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. The volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM) and memory bus dynamic RAM, etc.

[0060] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, system and unit can refer to the corresponding processes in the foregoing method embodiments, and will not be described here.

[0061] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0062] The above-described and above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A film and television supervision analysis method based on data mining, characterized in that: The film and television supervision analysis method based on data mining includes: Perform panoramic data collection on film and television works on video platforms, and organize video, audio, and text information in a hierarchical and standardized manner through cross-media transcoding processing to obtain a multi-dimensional film and television content dataset; Based on the multi-dimensional film and television content dataset, scene semantics are extracted from video frames, multi-level emotion recognition is performed on audio, and cultural value tendency analysis is performed on text to obtain a time-series related content tag list; Based on the time-series related content tag list, cross-scoring is performed on key elements according to the intensity of social impact, depth of communication penetration, and psychological acceptance of the audience through hierarchical scoring to obtain a dynamic weighted content scoring table; Based on the dynamic weight content scoring table, the network propagation path of the film and television works is analyzed through association tracking to obtain a three-dimensional content influence map; Based on the three-dimensional content impact map, regulatory points are prioritized through decision-making rules to obtain an intelligent regulatory solution that includes precise modification suggestions, gradient restriction measures, and personalized grading requirements; According to the intelligent supervision scheme, the execution effect of the supervision measures is analyzed for deviation correction through effect feedback to obtain a self-optimizing supervision parameter set.

2. The film and television supervision analysis method based on data mining according to claim 1 is characterized in that: The panoramic data collection of film and television works on the video platform is carried out, and the video, audio and text information are hierarchically standardized and organized through cross-media transcoding processing to obtain a multi-dimensional film and television content data set, including: Through the adaptive capture network, the film and television content of multi-source video platforms is deeply penetrated and collected to obtain the full spectrum of film and television raw data; Perform digital fingerprint recognition on the full spectrum of film and television raw data, perform intelligent detection on video frame rate, audio sampling and text encoding, and obtain a media feature map; According to the media feature graph, heterogeneous media formats are subjected to spatiotemporal synchronization processing through cross-coding conversion to obtain a unified standard media stream; Performing semantic segmentation on the unified standard media stream, accurately slicing the video, audio, and text using keyframe positioning technology to obtain a semantically associated slice set; hierarchically classifying the semantically associated slice set through multi-dimensional cross-mapping, constructing a content association network, and obtaining a structured content matrix; The structured content matrix is ​​written into a data warehouse through distributed storage, a multi-dimensional retrieval channel is established, and a multi-dimensional film and television content data set is obtained.

3. The film and television supervision analysis method based on data mining according to claim 1 is characterized in that: According to the multi-dimensional film and television content dataset, scene semantics are extracted from video frames, multi-level emotion recognition is performed on audio, and cultural value tendency analysis is performed on text through content feature recognition to obtain a time-series related content tag list, including: Extracting a video key frame sequence from the multi-dimensional film and television content data set, performing edge detection and region division on the video screen using scene segmentation technology, and obtaining a video scene combination feature; Using a semantic understanding engine to classify the subject elements of the video scene combination features, extract key visual elements and contextual relationships in the scene, and obtain a scene semantic mapping set; Separating an audio data stream from the multidimensional film and television content data set, performing waveform decomposition on the audio emotional features through spectrum analysis to obtain an emotional audio feature spectrum; Performing multi-level labeling on the emotional audio feature spectrum, and performing emotional classification on the pitch, speech speed and rhythm changes by emotional intensity gradient quantization to obtain multi-level emotional labeling data; Extracting text content from the multidimensional film and television content dataset, performing keyword extraction and value orientation analysis on the text through a semantic network, and obtaining a cultural value orientation index; The scene semantic mapping set, the multi-level emotion tag data and the cultural value tendency index are integrated and associated according to the time axis, a temporal mapping relationship between content elements is established, and a temporal associated content tag list is obtained.

4. The film and television supervision analysis method based on data mining according to claim 1 is characterized in that: Based on the time-series related content tag list, the key elements are cross-scored according to the social impact intensity, communication penetration depth and audience psychological acceptance through hierarchical scoring to obtain a dynamic weighted content scoring table, including: Screening the time-series related content tag list for elements, identifying and separating the sensitive elements and core content in the film and television content using key point extraction technology, and obtaining a set of key content elements; The key elements of the content are quantitatively scored according to the social impact intensity standard, and the social sensitivity and guiding role of the content elements are evaluated by referring to public opinion to obtain the social impact intensity index; Constructing a communication link model based on the key elements of the content, calculating the diffusion speed and coverage of the content among different groups of people through user behavior analysis, and obtaining a communication penetration depth value; Measure psychological acceptance based on the key elements of the content, analyze the psychological acceptance of people of different age groups and cultural backgrounds through audience portrait matching, and obtain the psychological acceptance level; Perform multi-dimensional cross-correlation on the social impact intensity index, the communication penetration depth value, and the psychological acceptance level, and dynamically weight the three indicators through a weight distribution matrix to obtain comprehensive score data; The comprehensive scoring data is hierarchically mapped, the content scoring is segmented by threshold adjustment, a variable weight scoring system is established, and a dynamic weight content scoring table is obtained.

5. The film and television supervision analysis method based on data mining according to claim 1 is characterized in that: The method of performing a mesh propagation path analysis of film and television works based on the dynamic weight content scoring table through association tracking to obtain a three-dimensional content influence map includes: Extract high-weight content tags from the dynamic weight content scoring table, perform matching analysis on related content in different film and television works through similarity calculation, and obtain a content association link library; Constructing a timeline propagation sequence of film and television works based on the content association link library, tracking the evolution trend of the content in different periods through time series data association, and obtaining a time series propagation trajectory diagram; Expanding the spatial dimension of the time-series propagation trajectory diagram, visualizing the propagation density and coverage of the content in different regions through regional distribution mapping, and obtaining the regional propagation heat distribution; The content association link library is constructed into a network topology according to the propagation relationship, and the key nodes and propagation paths of the content propagation are identified through node connection analysis to obtain a mesh propagation relationship network; Conducting influence evaluation on the mesh communication relationship network, quantitatively analyzing the radiation range and influence degree of each node in the communication network through influence diffusion calculation, and obtaining content influence evaluation data; The time-series propagation trajectory diagram, the regional propagation heat distribution and the content influence evaluation data are multi-dimensionally integrated to construct a three-dimensional propagation influence model and obtain a three-dimensional content influence map.

6. The film and television supervision analysis method based on data mining according to claim 1 is characterized in that: According to the three-dimensional content impact map, the supervision points are prioritized through decision rules to obtain an intelligent supervision solution containing precise modification suggestions, gradient restriction measures and personalized grading requirements, including: Extracting high-risk regulatory points from the three-dimensional content impact map, identifying and screening the high-risk regulatory points through risk weight calculation, and obtaining a list of key regulatory points; Prioritizing the key regulatory points in a hierarchical manner, determining the processing order and resource allocation ratio of different regulatory points through urgency assessment, and obtaining a regulatory priority sequence list; Based on the regulatory priority sequence table, for high-priority content nodes, the differences between illegal content and compliance standards are accurately located through content comparison analysis, and a precise list of modification suggestions is obtained; According to the content propagation characteristics in the key regulatory point list, a layered restriction scheme is formulated for content with different impact ranges through a gradient control strategy to obtain a gradient restriction measure scheme; Based on the content characteristics and audience information of the key regulatory points list, the scope of the content's applicable population is accurately defined through grading standard mapping to obtain a personalized grading requirement table; The precise modification suggestion list, the gradient restriction measure plan and the personalized grading requirement table are integrated and associated to obtain an intelligent supervision plan.

7. The film and television supervision analysis method based on data mining according to claim 1 is characterized in that: According to the intelligent supervision scheme, the execution effect of the supervision measures is analyzed for deviation correction through effect feedback to obtain a self-optimizing supervision parameter set, including: Implement and deploy the intelligent supervision solution, collect and record the time nodes and effect data of the implementation process of the measures through execution tracking, and obtain the supervision execution data flow; Extract key effect indicators from the regulatory execution data stream, quantitatively compare the difference between the expected regulatory effect and the actual execution result through target deviation measurement, and obtain regulatory effect deviation data; Conduct root cause analysis on the regulatory effect deviation data, identify and locate the key factors and influencing conditions that cause the deviation through tracing back to the source, and obtain a deviation factor attribution table; Formulate a correction strategy based on the deviation factor attribution table, determine the correction direction and adjustment range of different deviation factors through correction strength calculation, and obtain a regulatory parameter adjustment plan; Optimize and verify the regulatory parameter adjustment plan, test and evaluate the effectiveness of the adjusted regulatory parameters through small-scale experiments, and obtain a parameter optimization verification report; The effective parameters in the parameter optimization verification report are subjected to parameter extraction and analysis to obtain a self-optimizing regulatory parameter set.

8. A film and television supervision analysis system based on data mining, used to implement the film and television supervision analysis method based on data mining according to any one of claims 1 to 7, characterized in that: The film and television supervision analysis system based on data mining includes: The acquisition module is used to collect panoramic data of film and television works on the video platform, and to organize the video, audio and text information in a layered and standardized manner through cross-media transcoding processing to obtain a multi-dimensional film and television content data set; An extraction module is used to extract scene semantics from video frames, perform multi-level emotion recognition on audio, and analyze cultural value trends on text based on the multi-dimensional film and television content dataset through content feature recognition to obtain a list of time-series related content tags; A scoring module is used to cross-score key elements according to the social impact intensity, communication penetration depth and audience psychological acceptance through hierarchical scoring based on the time-related content tag list to obtain a dynamic weighted content scoring table; An analysis module is used to analyze the network propagation path of the film and television works through association tracking based on the dynamic weight content scoring table to obtain a three-dimensional content influence map; A sorting module is used to prioritize regulatory points based on the three-dimensional content impact map using decision rules to obtain an intelligent regulatory solution that includes precise modification suggestions, gradient restriction measures, and personalized grading requirements; The analysis module is used to perform deviation correction analysis on the execution effect of the supervision measures through effect feedback according to the intelligent supervision solution to obtain a self-optimizing supervision parameter set.

9. A computer device, characterized in that: It includes a memory and a processor, the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, it implements the film and television supervision analysis method based on data mining described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the processor is enabled to execute the film and television supervision analysis method based on data mining as described in any one of claims 1 to 7.

Images