Interactive content identification method and related device

Abstract

The invention discloses an interactive content identification method and a related device, and the method comprises the steps: searching corresponding video context information from a mapping relation between a timestamp and video context information according to the timestamp of the interactive content when the interactive content in a video is monitored, and the video context information comprises a mapping relation between an object and a video line, and performing preset information detection on the interactive content and the video context information to obtain an information detection result, splicing the interactive content ID, the interactive content, the video context information and the information detection result into a feature sequence, and identifying the feature sequence to obtain an interactive content identification result. According to the method, the interaction content and the video context information are associated creatively, and traditional single interaction text analysis is upgraded to scene interaction analysis combining the interaction content and the video context, so that the accuracy of interaction content identification is improved, and the probability of occurrence of missing report or false report is effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to an interactive content recognition method and related apparatus. Background Technology

[0002] Currently, users typically interact with video content played on video playback platforms through methods such as bullet comments and real-time comments to enhance the viewing experience. Existing methods for identifying interactive content mainly rely on the analysis of interactive text within the content. Identification methods may include: (1) Keyword / regular expression filtering: setting up a dictionary containing sensitive words and identifying and blocking interactive content through string matching. (2) Text classification models: using machine learning or deep learning models (such as TextCNN, BERT) to classify interactive text and determine whether it involves pornography, violence, advertising, personal attacks, etc.

[0003] Traditional interactive content recognition mainly targets single interactive text, resulting in low accuracy and a high likelihood of missed or false alarms. Summary of the Invention

[0004] In view of this, the present invention discloses an interactive content recognition method and related apparatus to improve the accuracy of interactive content recognition and effectively reduce the probability of missed or false alarms.

[0005] An interactive content recognition method, comprising:

[0006] When interactive content is detected in the video, the video context information corresponding to the timestamp is found from the pre-built mapping relationship between timestamps and video context information based on the timestamp of the interactive content. The video context information includes the mapping relationship between objects and video dialogue.

[0007] Pre-defined information detection is performed on the interactive content and the video context information to obtain information detection results;

[0008] The interactive content ID, the interactive content, the video context information, and the information detection result are concatenated into a feature sequence;

[0009] The feature sequence is identified to obtain the interactive content recognition result.

[0010] Optionally, the process of constructing the mapping relationship between the timestamp and video context information includes:

[0011] The audio in the video is segmented to obtain different object segments;

[0012] Each object segment is matched with the actor's voiceprint database to identify the actor ID corresponding to the object segment;

[0013] Based on the list of names of characters in the cast of the video, determine the role corresponding to the actor ID;

[0014] Extract video dialogue from the audio track of the video;

[0015] A multi-dimensional mapping relationship is established between the actor ID, the role, and the video dialogue corresponding to the same timestamp, wherein the objects include: the actor ID and the corresponding role.

[0016] Optionally, the step of performing preset information detection on the interactive content and the video context information to obtain information detection results includes:

[0017] Sensitive word detection is performed on the interactive content to obtain the sensitive word detection results of the interactive content;

[0018] Sensitive word detection is performed on the video dialogue to obtain the sensitive word detection results.

[0019] From a pre-defined database of sensitive individuals, query the sensitivity level of the person corresponding to the object, wherein each record in the database of sensitive individuals contains the person's name and the corresponding sensitivity level.

[0020] The results of sensitive word detection in the interactive content, the results of sensitive word detection in the video dialogue, and the sensitivity level of the characters are determined as the information detection results.

[0021] Optionally, querying the sensitivity level of the person corresponding to the person from a preset sensitive person database includes:

[0022] From the sensitive actor database, query the actor sensitivity level corresponding to the actor ID. The sensitive actor database is contained in the sensitive person database, and the actor ID is contained in the object.

[0023] From the sensitive character database, query the character sensitivity level of the character corresponding to the actor ID. The sensitive character database is contained in the sensitive person database, and the character is contained in the object.

[0024] The higher sensitivity level between the actor's sensitivity level and the character's sensitivity level is taken as the character's sensitivity level.

[0025] Optionally, the step of identifying the feature sequence to obtain the interactive content identification result includes:

[0026] The interactive content and the video context information are input into a pre-trained text classification model to predict the risk type and risk probability.

[0027] The rule engine uses a preset weighting formula to calculate the person sensitivity weighting value and sensitive word hit weighting value in the information detection results. The sensitive word hit weighting value includes: interactive content sensitive word hit weighting value and video dialogue sensitive word hit weighting value.

[0028] Using the risk probability as the initial value of the risk score, the sum of the initial risk score, the weighted value of the person's sensitivity, and the weighted value of the sensitive word hit is calculated to obtain the risk score.

[0029] The interactive content ID, the risk type, and the risk score are determined as the interactive content identification result.

[0030] Optionally, it also includes:

[0031] Determine the target risk score range that includes the risk score, and from the pre-set correspondence between risk levels and risk score ranges, use the target risk level corresponding to the target risk score range as the risk level of the interactive content.

[0032] Optionally, it also includes:

[0033] From the pre-set review priority classification criteria, the target review priority corresponding to the risk level and the sensitivity level of the person is determined, wherein the sensitivity level of the person is included in the information detection result;

[0034] From the pre-set interactive content risk processing queue, determine the target review priority and the target interactive content risk processing queue corresponding to the risk type;

[0035] The interactive content is sent to the target interactive content risk processing queue for processing.

[0036] An interactive content recognition device includes:

[0037] The context information acquisition unit is used to, when interactive content in a video is detected, find the video context information corresponding to the timestamp from a pre-built mapping relationship between timestamps and video context information based on the timestamp of the interactive content. The video context information includes: a mapping relationship between objects and video dialogue.

[0038] An information detection unit is used to perform preset information detection on the interactive content and the video context information to obtain information detection results;

[0039] A sequence determination unit is used to concatenate the interactive content ID, the interactive content, the video context information, and the information detection result into a feature sequence;

[0040] The recognition unit is used to recognize the feature sequence and obtain the interactive content recognition result.

[0041] A computer storage medium storing at least one instruction that, when executed by a processor, implements any interactive content recognition method.

[0042] An electronic device, comprising: a memory and a processor;

[0043] The memory is used to store at least one instruction;

[0044] The processor is used to execute the at least one instruction to implement any interactive content recognition method.

[0045] As can be seen from the above technical solution, this invention discloses an interactive content recognition method and related apparatus. When interactive content in a video is detected, the video context information corresponding to the timestamp is found from a pre-constructed mapping relationship between timestamps and video context information based on the timestamp of the interactive content. This video context information includes a mapping relationship between objects and video dialogue. Pre-defined information detection is performed on the interactive content and video context information to obtain information detection results. The interactive content ID, interactive content, video context information, and information detection results are concatenated into a feature sequence, and the feature sequence is recognized to obtain the interactive content recognition result. This invention innovatively associates interactive content with video context information for pre-defined information detection during interactive content recognition. By concatenating the interactive content ID, interactive content, video context information, and information detection results into a feature sequence, recognition is completed. Therefore, this invention achieves the recognition of information after the combination of interactive content and video context information, upgrading traditional single interactive text analysis to contextual interaction analysis combining interactive content and video context, thereby significantly improving the accuracy of interactive content recognition and effectively reducing the probability of false negatives or negatives. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the published drawings without creative effort.

[0047] Figure 1 This is a flowchart of an interactive content recognition method disclosed in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the structure of an interactive content recognition device disclosed in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of the present invention. Detailed Implementation

[0050] Through research, the inventors discovered that traditional interactive content identification primarily targets individual interactive texts. Based on this, users often exploit "implicit expressions" to circumvent censorship. For example, when a person with a specific identity utters a line in a video, a user might send a comment (such as "He's right" or "I support you"). However, this comment, combined with the video dialogue and the person's identity, constitutes a clear political stance or implicit support for sensitive statements. Existing solutions, focusing on identifying individual interactive texts without associating "interactive content" with "video dialogue / object," completely fail to identify such risks, resulting in low accuracy in interactive content identification.

[0051] The inventors' research also revealed that in video scenarios, such as in game videos, comments like "Kill him" are considered normal descriptions; however, in realistic films and television dramas, they could be interpreted as inciting violence. Due to the lack of contextual information, many normal comments are easily misreported, severely impacting user experience. Furthermore, the combined risks of interactive content and video contextual information, as mentioned above, are often missed because the interactive content itself does not contain sensitive words.

[0052] Based on this, embodiments of the present invention disclose an interactive content recognition method and related apparatus. When performing interactive content recognition, it innovatively associates interactive content with video context information for preset information detection. Recognition is achieved by concatenating the interactive content ID, interactive content, video context information, and information detection results into a feature sequence. Therefore, the present invention realizes the recognition of information combining interactive content and video context information, upgrading traditional single interactive text analysis to contextual interaction analysis combining interactive content and video context. This significantly improves the accuracy of interactive content recognition and effectively reduces the probability of false negatives or negatives.

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] See Figure 1 This invention discloses a flowchart of an interactive content recognition method. This method is applied to video playback platforms with interactive functions (such as "bullet comments" and "real-time comments"). Users can interact with the video content played on the video playback platform through bullet comments, real-time comments, etc., thereby enhancing the viewing experience.

[0055] Application scenarios where users interact with video playback platforms can include:

[0056] (1) Video on demand platform: Users send bullet comments when watching movies, documentaries, and self-made videos.

[0057] (2) Live streaming platform: Viewers send real-time comments when the host is live streaming games, shows, sports events, etc.

[0058] (3) Online education platform: Students can ask questions and participate in discussions via bullet comments while watching course recordings or live broadcasts.

[0059] In the aforementioned interactive application scenarios, the interactive content recognition method disclosed in this invention includes the following steps:

[0060] Step S101: When interactive content in the video is detected, the video context information corresponding to the timestamp is found from the pre-built mapping relationship between timestamps and video context information based on the timestamp of the interactive content.

[0061] During video playback, when interactive content is detected in the video, such as voice comments, text comments, real-time comments, etc., the timestamp of the interactive content (i.e. the specific time point when the interactive content occurs) is obtained.

[0062] The pre-constructed mapping relationship between timestamps and video context information in this invention can be viewed as a structured video context database with time as the axis. Each timestamp on the timeline records the video context information for that timestamp, including a mapping relationship between objects and video dialogue. Objects can be characters in the video, and characters include actor IDs and their corresponding roles. In other words, objects can include actor IDs and their corresponding roles, and each timestamp on the timeline records a mapping relationship of "actor ID - role - video dialogue." The actor ID is a digital or symbolic code used to uniquely identify an actor. In the film and television industry, content management systems, or databases, actor IDs can accurately distinguish different actors, avoiding erroneous associations caused by identical names, aliases, or information confusion.

[0063] When interactive content is detected in a video, the interactive content is no longer an isolated string, but can be used as a trigger. This trigger can query the video context information corresponding to the timestamp of the interactive content from the mapping relationship. Thus, this application can conduct an overall risk assessment of the interactive content based on the triple consisting of "interactive content" + "object" + "video dialogue".

[0064] In practical applications, the role corresponding to the actor ID can include not only the role name, but also the actor's facial expressions (such as anger, sadness), body movements (such as saluting, holding weapons), and the scene in which the role is performed (such as historical sites, battlefields), etc., to achieve full-modal recognition of "visual (facial expressions, body movements, scenes), auditory (video dialogue), and text (interactive content)".

[0065] Step S102: Perform preset information detection on the interactive content and the video context information to obtain information detection results.

[0066] Specifically, the process of performing pre-defined information detection on interactive content and video context information may include:

[0067] (1) Perform sensitive word detection on the interactive content and obtain the sensitive word detection results of the interactive content.

[0068] In practical applications, a database containing sensitive words can be set up. By matching the interactive content with the sensitive words in the database, the sensitive words in the interactive content can be detected, and the results of the sensitive word detection can be obtained.

[0069] The results of sensitive word detection in interactive content can include: whether the interactive content contains sensitive words, and the number of sensitive words detected in the interactive content.

[0070] Sensitive words refer to words involving violence, personal attacks, negative emotions, etc.

[0071] (2) Perform sensitive word detection on the video dialogue to obtain the sensitive word detection results of the video dialogue.

[0072] Similarly, the video dialogue is matched with sensitive words in the database to detect sensitive words in the video dialogue and obtain the sensitive word detection results.

[0073] The results of sensitive word detection for video dialogue can include: whether the video dialogue contains sensitive words, and the number of sensitive words detected in the video dialogue.

[0074] (3) Query the sensitivity level of the person from the preset sensitive person database.

[0075] In practical applications, each record in the sensitive person database contains {person name, person type (actor / role), person sensitivity level (high / medium / low), reason for classification, update time}. When the person type is actor, the person name is actor ID; when the person type is role, the person name is role name.

[0076] In this embodiment, the characters include: actor ID and corresponding role. In practical applications, the risk may come from the actor himself or from the role played by the actor. Therefore, when determining the character sensitivity level, it is necessary to determine both the actor sensitivity level and the role sensitivity level. The final character sensitivity level is determined by comprehensively considering both the actor sensitivity level and the role sensitivity level.

[0077] Specifically, the sensitive figures database includes: a sensitive actors database and a sensitive characters database.

[0078] From the sensitive actor database, query the actor sensitivity level corresponding to the actor ID;

[0079] From the sensitive roles database, query the role sensitivity level corresponding to the actor ID;

[0080] The higher sensitivity level between the actor's sensitivity level and the character's sensitivity level will be used as the character's sensitivity level.

[0081] The actors in the sensitive actor database can be identified based on online public opinion, news reports, regulatory notices, and social media trending topics. Furthermore, the sensitivity level of actors can be categorized into high-risk, medium-risk, and low-risk levels, or, based on the specific circumstances, levels 1, 2, 3, etc.

[0082] The sensitive roles database can be determined based on factors such as social controversy surrounding the role and whether it touches on historically sensitive issues. Furthermore, the sensitivity level of a role can be categorized into high-risk, medium-risk, and low-risk levels based on the specific circumstances.

[0083] Therefore, this invention proposes a method for determining a character's sensitivity level based on dual sensitivity assessment. This method performs dual assessments of "actor sensitivity level" and "role sensitivity level," and determines the higher of the two sensitivity levels as the final character sensitivity level. Through a cross-validation mechanism of "actor-role" sensitivity, this method effectively avoids the limitations of single-dimensional assessment and significantly improves the accuracy and reliability of character sensitivity level determination.

[0084] It should be noted that in this embodiment, the results of sensitive word detection in interactive content, sensitive word detection in video dialogue, and sensitivity level of characters are determined as information detection results.

[0085] In summary, the pre-defined information detection in this invention is a core component of video platform content governance. By detecting sensitive words in interactive content and video dialogue, and determining the sensitivity level of individuals, it enables the identification of information resulting from the combination of interactive content and video context, thus providing data support for improving interactive content recognition. Furthermore, by considering individual sensitivity, this invention can focus the identification efforts on the nodes most likely to trigger public opinion risks, thereby improving review efficiency and the risk response capabilities of video playback platforms.

[0086] Step S103: Concatenate the interactive content ID, the interactive content, the video context information, and the information detection result into a feature sequence.

[0087] This embodiment integrates the originally separate information (interactive content ID, interactive content, video context information, and information detection results) into a logically coherent single feature sequence that allows the model to perform comprehensive reasoning. This feature sequence is not a simple text concatenation, but rather uses a structured format with delimiters or tags to combine this information, which helps to distinguish the interactive content ID, interactive content, video context information, and information detection results in the feature sequence, thereby enabling targeted correlation.

[0088] Step S104: Identify the feature sequence to obtain the interactive content identification result.

[0089] The interactive content identification results include: the interactive content ID and its corresponding risk type and risk score.

[0090] Risk types include, for example, political metaphors and inappropriate associations.

[0091] It should be noted that the interactive content ID is a unique identifier used to distinguish each interactive content. Each interactive content sent by a user on a video will carry the corresponding ID information.

[0092] In summary, this invention discloses an interactive content recognition method. When interactive content is detected in a video, the method retrieves the corresponding video context information based on the timestamp of the interactive content from a pre-constructed mapping relationship between timestamps and video context information. This video context information includes a mapping relationship between objects and video dialogue. Pre-defined information detection is performed on the interactive content and video context information to obtain information detection results. The interactive content ID, interactive content, video context information, and information detection results are concatenated into a feature sequence to obtain the interactive content recognition result. This invention innovatively associates interactive content with video context information for pre-defined information detection during interactive content recognition. By concatenating the interactive content ID, interactive content, video context information, and information detection results into a feature sequence, recognition is completed. Therefore, this invention achieves the recognition of information combining interactive content and video context information, upgrading traditional single interactive text analysis to contextual interaction analysis combining interactive content and video context, thereby significantly improving the accuracy of interactive content recognition and effectively reducing the probability of false negatives or negatives.

[0093] In one embodiment, the process of constructing the mapping relationship includes:

[0094] (1) The audio in the video is segmented to obtain different object segments.

[0095] Specifically, object recognition technology can be used to segment the audio in a video to obtain different object segments.

[0096] (2) Match each object segment with the actor voiceprint database to identify the actor ID corresponding to the object segment.

[0097] In practical applications, speaker recognition (SR) technology can be used to analyze the unique biometric features in speech signals to identify or verify the identity of an individual.

[0098] (3) Based on the list of names of the actors in the video, determine the role corresponding to the actor ID.

[0099] (4) Extract video dialogue from the audio track of the video.

[0100] Audio tracks refer to the sound portion of a video, including background music, original sound, and dubbing.

[0101] In practical applications, automatic speech recognition technology can be used to extract video dialogue from the audio track of the video.

[0102] The goal of Automatic Speech Recognition (ASR) is to convert the lexical content of human speech into computer-readable input, such as keystrokes, binary codes, or character sequences.

[0103] In practical applications, automatic speech recognition technology can be used to process the audio track of the entire video to extract the video dialogue. Alternatively, automatic speech recognition technology can be used to process the audio tracks of the segmented video clips separately to extract the video dialogue from each track.

[0104] (5) Establish a multi-dimensional mapping relationship between actor ID, role and video lines corresponding to the same timestamp.

[0105] The objects include: actor IDs and their corresponding roles.

[0106] This embodiment establishes a mapping relationship of "actor ID-role-video dialogue" for each timestamp based on timestamps. The "actor ID-role-video dialogue" can serve as the video context information under that timestamp.

[0107] The multidimensional mapping relationship corresponding to each timestamp can be in the form of a mapping table or an entry in a time series database.

[0108] It should be noted that the speaker recognition technology and automatic speech recognition technology used in this invention both belong to AI (Artificial Intelligence) technology.

[0109] In summary, this invention constructs a multi-dimensional mapping relationship between actor ID, role, and video dialogue corresponding to each timestamp. In other words, by constructing video context information corresponding to each timestamp, it facilitates the association of interactive content with video context information through timestamps, thereby enabling the identification of information after the combination of interactive content and video context information. This upgrades traditional single interactive text analysis to contextual interactive analysis that combines interactive content and video context, thus significantly improving the accuracy of interactive content identification and effectively reducing the probability of missed or false alarms.

[0110] In one embodiment, a feature sequence composed of the interactive content ID, interactive content, video context information, and information detection results can be input into a pre-trained recognition module to obtain the interactive content recognition result.

[0111] The recognition module in this application may include a pre-trained text classification model and a rule engine.

[0112] Therefore, in one embodiment, step S104 may specifically include:

[0113] (1) Input the interactive content and video context information into the pre-trained text classification model to predict the risk type and risk probability.

[0114] The process of inputting interactive content and video context information into a pre-trained text classification model in this embodiment can be regarded as a preliminary assessment of "semantic risk" based on the text classification model.

[0115] The input to the text classification model is a text sequence consisting of interactive content and video context information, which integrates purely semantic information from interactive content, video dialogue, actor IDs, and characters.

[0116] The output of the text classification model includes two preliminary results:

[0117] model_base_score: A floating-point number between 0.0 and 1.0, representing the risk probability of the combination of interactive content and video context information at the purely semantic level.

[0118] predicted_risk_type: The text classification model's preliminary judgment of the risk type, such as "Political Metaphor" or "Inappropriate Association".

[0119] (2) The sensitivity weighting value of the person and the hit weighting value of the sensitive words in the information detection results are calculated by using a preset weighting formula through the rule engine.

[0120] The weighted value for sensitive word hits includes: the weighted value for sensitive word hits in interactive content and the weighted value for sensitive word hits in video dialogue.

[0121] Person sensitivity level = max(Actor sensitivity level, Character sensitivity level), which means selecting the highest sensitivity level between the actor sensitivity level and the character sensitivity level.

[0122] 1) An example illustrating the formula for calculating the weighted value of a person's sensitivity is as follows:

[0123] final_risk_score+=(person_sensitivity_level*0.2);

[0124] When the character sensitivity level is 1, the character sensitivity weighting value is 0.2 points; when the character sensitivity level is 2, the character sensitivity weighting value is 0.4 points; when the character sensitivity level is 3, the character sensitivity weighting value is 0.6 points.

[0125] 2) An example illustrating the formula for calculating the weighted value of sensitive word hits in interactive content is as follows:

[0126] if (interactive content hits a sensitive word): final_risk_score += 0.1;

[0127] That is, the weighted value for sensitive word hits in interactive content is 0.1 points.

[0128] 3) An example illustrating the formula for calculating the weighted value of sensitive word hits in video dialogue is as follows:

[0129] if (video dialogue contains sensitive words): final_risk_score += 0.15;

[0130] That is, the weighted score for the hit of sensitive words in the video dialogue is 0.15.

[0131] (3) Using the risk probability as the initial value of the risk score, calculate the sum of the initial value of the risk score, the weighted value of the person sensitivity and the weighted value of the sensitive word hit to obtain the risk score.

[0132] Assuming the risk probability is 0.1 points, the sensitivity weighting for the person is 0.2 points, the weighting for the hit of sensitive words in interactive content is 0.1 points, and the weighting for the hit of sensitive words in video dialogue is 0.15 points.

[0133] The risk score final_risk_score = 0.1 + 0.2 + 0.1 + 0.15 = 0.55.

[0134] It should be noted that the total risk score cannot exceed 1.0.

[0135] That is, final_risk_score=min(final_risk_score,1.0).

[0136] (4) The interactive content ID, risk type and risk score are determined as the interactive content identification results.

[0137] In summary, this invention processes interactive content and video context information using a text classification model to achieve a preliminary assessment of semantic risks, obtaining risk types and probabilities. A rule engine then adjusts the risk probability (i.e., the model's base score) based on high-risk features such as person sensitivity and sensitive word hits, thus obtaining the final interactive content recognition result. Because this application integrates interactive content and video context during interactive content recognition, it significantly improves the accuracy of interactive content recognition and effectively reduces the probability of false positives or negatives. Simultaneously, this invention can effectively combat the use of metaphors, allusions, and other methods to publish inappropriate content, thereby greatly enhancing the platform's ability to identify complex and varied inappropriate content.

[0138] In one embodiment, the interactive content recognition method may further include:

[0139] Determine the target risk score range that includes the risk score, and from the pre-set correspondence between risk levels and risk score ranges, use the target risk level corresponding to the target risk score range as the risk level of the interactive content.

[0140] This invention sets different risk score ranges for different risk levels. By matching the final determined risk score with each risk score range, the risk level of the interactive content can be determined, so that technicians can take corresponding measures according to the risk level of the interactive content.

[0141] In one embodiment, the interactive content recognition method may further include:

[0142] (1) Determine the target review priority corresponding to the risk level of interactive content and the sensitivity level of the person from the pre-set review priority classification criteria.

[0143] The sensitivity level of individuals is included in the information detection results.

[0144] This invention pre-sets different review priorities for interactive content based on its varying risk levels and the sensitivity levels of the individuals associated with it. For example, interactive content linked to a Level 1 sensitive individual has the highest review priority regardless of its risk level. Therefore, this invention solves the problem of dynamically allocating review resources based on the risk of interactive content, thereby improving review efficiency.

[0145] (2) Determine the target review priority and the target interactive content risk processing queue corresponding to the risk type from the pre-set interactive content risk processing queue.

[0146] (3) Send the interactive content to the target interactive content risk processing queue for processing.

[0147] For example, regarding the pre-set risk handling queues for interactive content: interactive content with a high risk level and a high sensitivity level for the person involved -> Sensitive Person Queue; interactive content with a high risk level and belonging to a combined risk (the person's sensitivity level is not high, and it is a combined risk consisting of the interactive content and video context information) -> Combined Risk Queue; interactive content with low risk -> Normal Queue.

[0148] Traditional methods for identifying interactive content also include scoring user accounts based on their historical behavior (such as the number of reports and violations), with low-reputation users facing stricter scrutiny. The inventors discovered that simply relying on user reputation for identification fails to recognize interactive content related to "sensitive individuals" or "sensitive events," leading to an even distribution of review resources and an inability to dynamically adjust based on the risk of the interactive content, thus impacting the efficiency of interactive content review.

[0149] Based on this, this invention constructs a review priority allocation mechanism. By comprehensively assessing the risk level of interactive content and the sensitivity level of individuals, the corresponding review priority is determined. Subsequently, combining the specific risk type of the interactive content with its review priority, the interactive content is accurately distributed to the corresponding risk processing queue. This mechanism achieves dynamic matching between review resources and the risk of interactive content—high-risk, highly sensitive content is given priority to enter the fast review channel, while low-risk content is processed through standardized procedures, thereby optimizing the allocation efficiency of review resources and improving the overall review efficiency of interactive content.

[0150] In identifying interactive content, this application considers not only the text content contained in the interactive content, but also creates user profiles (such as historical violation tendencies and interest tags) based on the text content to make personalized and accurate judgments on the risks of interactive content.

[0151] In addition, it is possible to identify risks based on the group behavior of interactive content. For example, if a large number of similar, repetitive, and consistent interactive content appears in a short period of time, these can be used as a new risk dimension for interactive content.

[0152] Regarding the interactive content recognition process, the present invention also provides a specific embodiment, as follows:

[0153] Business scenarios:

[0154] 1. Video content: A historical drama series, which includes a speech clip of character M.

[0155] 2. Key scene (assuming the video plays to 00:50:10): Actor ID is Actor A; corresponding character B, line - "All I ask for is for everyone to live and work in peace and contentment."

[0156] 3. Interactive content: When the video played to 00:50:12 (i.e., the timestamp), a user sent an interactive message that read: "Talk is better than singing."

[0157] Interactive content recognition process:

[0158] 1. Based on the timestamp of the interaction content, find the video context information corresponding to the timestamp 00:50:10-00:50:18: Actor ID is Actor A; corresponding character B, video line "All I ask for is for everyone to live and work in peace and contentment".

[0159] 2. Perform preset information detection on interactive content and video context:

[0160] 1) Sensitive word detection result in interactive content: No match (i.e. no sensitive words were detected in the interactive content).

[0161] 2) Sensitive word detection result in video dialogue: No match (i.e. no sensitive words were detected in the video dialogue).

[0162] 3) Character sensitivity level: Actor A's actor sensitivity level is the highest level, character B's character sensitivity level is the highest level, and the final character sensitivity level is the highest level.

[0163] 4) Combining 1), 2), and 3), the information detection result is: it has the highest level of sensitivity to human beings.

[0164] 3. Interactive content recognition:

[0165] The interactive content ID, interactive content, video context information, and information detection results are concatenated into a feature sequence.

[0166] The identification results of the interactive content obtained by identifying the feature sequences include:

[0167] Interactive content ID;

[0168] Risk score model_base_score=0.75;

[0169] The risk type is predicted_risk_type="Political_Sarcasm".

[0170] Corresponding to the above method embodiments, the present invention also discloses an interactive content recognition device.

[0171] See Figure 2The present invention discloses a schematic diagram of an interactive content recognition device. This device is applied to a video playback platform with interactive functions (such as "bullet comments" and "real-time comments"). Users can interact with the video content played on the video playback platform through bullet comments, real-time comments, etc., thereby enhancing the viewing experience.

[0172] The interactive content recognition device disclosed in this invention includes:

[0173] The context information acquisition unit 201 is used to, when interactive content in a video is detected, find the video context information corresponding to the timestamp from a pre-built mapping relationship between timestamps and video context information based on the timestamp of the interactive content. The video context information includes a mapping relationship between objects and video dialogue.

[0174] In practical applications, the object can include: actor ID and the corresponding role.

[0175] The information detection unit 202 is used to perform preset information detection on the interactive content and the video context information to obtain information detection results.

[0176] The sequence determination unit 203 is used to concatenate the interactive content ID, the interactive content, the video context information, and the information detection result into a feature sequence.

[0177] The recognition unit 204 is used to recognize the feature sequence and obtain the interactive content recognition result.

[0178] The interactive content identification results include: the interactive content ID and its corresponding risk type and risk score.

[0179] In summary, this invention discloses an interactive content recognition device. When interactive content is detected in a video, the device retrieves the corresponding video context information based on the timestamp of the interactive content from a pre-constructed mapping relationship between timestamps and video context information. This video context information includes a mapping relationship between objects and video dialogue. Pre-defined information detection is performed on the interactive content and video context information to obtain information detection results. The interactive content ID, interactive content, video context information, and information detection results are concatenated into a feature sequence, and the feature sequence is then recognized to obtain the interactive content recognition result. This invention innovatively associates interactive content with video context information for pre-defined information detection during interactive content recognition. By concatenating the interactive content ID, interactive content, video context information, and information detection results into a feature sequence, recognition is completed. Therefore, this invention achieves the recognition of information after the combination of interactive content and video context information, upgrading traditional single interactive text analysis to contextual interaction analysis combining interactive content and video context, thereby significantly improving the accuracy of interactive content recognition and effectively reducing the probability of false negatives or negatives.

[0180] In one embodiment, the interactive content recognition device may further include:

[0181] Mapping relationship building unit, used for:

[0182] The audio in the video is segmented to obtain different object segments;

[0183] Each object segment is matched with the actor's voiceprint database to identify the actor ID corresponding to the object segment;

[0184] Based on the list of names of characters in the cast of the video, determine the role corresponding to the actor ID;

[0185] Extract video dialogue from the audio track of the video;

[0186] Establish a multi-dimensional mapping relationship between the actor ID, the role, and the video dialogue corresponding to the same timestamp, wherein the objects include: actor ID and the corresponding role.

[0187] In one embodiment, the information detection unit 202 can be specifically used for:

[0188] Sensitive word detection is performed on the interactive content to obtain the sensitive word detection results of the interactive content;

[0189] Sensitive word detection is performed on the video dialogue to obtain the sensitive word detection results.

[0190] From a pre-defined database of sensitive individuals, query the sensitivity level of the person corresponding to the object, wherein each record in the database of sensitive individuals contains the person's name and the corresponding sensitivity level.

[0191] The results of sensitive word detection in the interactive content, the results of sensitive word detection in the video dialogue, and the sensitivity level of the characters are determined as the information detection results.

[0192] In one embodiment, the information detection unit 202 can also be used for:

[0193] From the sensitive actor database, query the actor sensitivity level corresponding to the actor ID. The sensitive actor database is contained in the sensitive person database, and the actor ID is contained in the object.

[0194] From the sensitive character database, query the character sensitivity level of the character corresponding to the actor ID. The sensitive character database is contained in the sensitive person database, and the character is contained in the object.

[0195] The higher sensitivity level between the actor's sensitivity level and the character's sensitivity level is taken as the character's sensitivity level.

[0196] In one embodiment, the identification unit 204 can be specifically used for:

[0197] The interactive content and the video context information are input into a pre-trained text classification model to predict the risk type and risk probability.

[0198] The rule engine uses a preset weighting formula to calculate the person sensitivity weighting value and sensitive word hit weighting value in the information detection results. The sensitive word hit weighting value includes: interactive content sensitive word hit weighting value and video dialogue sensitive word hit weighting value.

[0199] Using the risk probability as the initial value of the risk score, the sum of the initial risk score, the weighted value of the person's sensitivity, and the weighted value of the sensitive word hit is calculated to obtain the risk score.

[0200] The interactive content ID, the risk type, and the risk score are determined as the interactive content identification result.

[0201] In one embodiment, the interactive content recognition device may further include:

[0202] The risk level determination unit is used to determine the target risk score range containing the risk score, and from the pre-set correspondence between risk levels and risk score ranges, to take the target risk level corresponding to the target risk score range as the risk level of the interactive content.

[0203] In one embodiment, the interactive content recognition device may further include:

[0204] The review priority determination unit is used to determine the target review priority corresponding to the risk level and the sensitivity level of the person from a pre-set review priority division standard, wherein the sensitivity level of the person is included in the information detection result;

[0205] The risk processing queue determination unit is used to determine the target interactive content risk processing queue corresponding to the target review priority and the risk type from a pre-set interactive content risk processing queue.

[0206] The initiation unit is used to send the interactive content to the target interactive content risk processing queue for processing.

[0207] It should be noted that for the specific working principles of each component in the device embodiment, please refer to the corresponding section of the method embodiment, which will not be repeated here.

[0208] Corresponding to the above embodiments, the present invention also discloses a computer storage medium that stores at least one instruction, which, when executed by a processor, implements the steps shown in the embodiments of the interactive content recognition method.

[0209] Computer storage media can be tangible media that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. Computer storage media can be machine-readable signal media or machine-readable storage media. Computer storage media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0210] Corresponding to the above embodiments, such as Figure 3 As shown, the present invention also provides a schematic diagram of the structure of an electronic device, which may include: a processor 1 and a memory 2;

[0211] The processor 1 and memory 2 communicate with each other via communication bus 3.

[0212] Processor 1, for executing at least one instruction;

[0213] Memory 2 is used to store at least one instruction;

[0214] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0215] Memory 2 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0216] The processor executes at least one instruction to implement the steps shown in the embodiment of the interactive content recognition method.

[0217] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0218] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0219] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An interactive content recognition method, characterized in that, include: When interactive content is detected in the video, the video context information corresponding to the timestamp is found from the pre-built mapping relationship between timestamps and video context information based on the timestamp of the interactive content. The video context information includes the mapping relationship between objects and video dialogue. Pre-defined information detection is performed on the interactive content and the video context information to obtain information detection results; The interactive content ID, the interactive content, the video context information, and the information detection result are concatenated into a feature sequence; The feature sequence is identified to obtain the interactive content identification result.

2. The interactive content recognition method according to claim 1, characterized in that, The process of constructing the mapping relationship between the timestamp and video context information includes: The audio in the video is segmented to obtain different object segments; Each object segment is matched with the actor's voiceprint database to identify the actor ID corresponding to the object segment; Based on the list of names of characters in the cast of the video, determine the role corresponding to the actor ID; Extract video dialogue from the audio track of the video; A multi-dimensional mapping relationship is established between the actor ID, the role, and the video dialogue corresponding to the same timestamp, wherein the objects include: the actor ID and the corresponding role.

3. The interactive content recognition method according to claim 1, characterized in that, The step of performing preset information detection on the interactive content and the video context information to obtain information detection results includes: Sensitive word detection is performed on the interactive content to obtain the sensitive word detection results of the interactive content; Sensitive word detection is performed on the video dialogue to obtain the sensitive word detection results. From a pre-defined database of sensitive individuals, query the sensitivity level of the person corresponding to the object, wherein each record in the database of sensitive individuals contains the person's name and the corresponding sensitivity level. The results of sensitive word detection in the interactive content, the results of sensitive word detection in the video dialogue, and the sensitivity level of the characters are determined as the information detection results.

4. The interactive content recognition method according to claim 3, characterized in that, The step of querying the sensitivity level of a person from a preset database of sensitive individuals includes: From the sensitive actor database, query the actor sensitivity level corresponding to the actor ID. The sensitive actor database is contained in the sensitive person database, and the actor ID is contained in the object. From the sensitive character database, query the character sensitivity level of the character corresponding to the actor ID. The sensitive character database is contained in the sensitive person database, and the character is contained in the object. The higher sensitivity level between the actor's sensitivity level and the character's sensitivity level is taken as the character's sensitivity level.

5. The interactive content recognition method according to any one of claims 1 to 4, characterized in that, The process of identifying the feature sequence to obtain the interactive content identification result includes: The interactive content and the video context information are input into a pre-trained text classification model to predict the risk type and risk probability. The rule engine uses a preset weighting formula to calculate the person sensitivity weighting value and sensitive word hit weighting value in the information detection results. The sensitive word hit weighting value includes: interactive content sensitive word hit weighting value and video dialogue sensitive word hit weighting value. Using the risk probability as the initial value of the risk score, the sum of the initial risk score, the weighted value of the person's sensitivity, and the weighted value of the sensitive word hit is calculated to obtain the risk score. The interactive content ID, the risk type, and the risk score are determined as the interactive content identification result.

6. The interactive content recognition method according to any one of claims 1 to 4, characterized in that, Also includes: Determine the target risk score range that includes the risk score, and from the pre-set correspondence between risk levels and risk score ranges, use the target risk level corresponding to the target risk score range as the risk level of the interactive content.

7. The interactive content recognition method according to claim 6, characterized in that, Also includes: From the pre-set review priority classification criteria, the target review priority corresponding to the risk level and the sensitivity level of the person is determined, wherein the sensitivity level of the person is included in the information detection result; From the pre-set interactive content risk processing queue, determine the target review priority and the target interactive content risk processing queue corresponding to the risk type; The interactive content is sent to the target interactive content risk processing queue for processing.

8. An interactive content recognition device, characterized in that, include: The context information acquisition unit is used to, when interactive content in a video is detected, find the video context information corresponding to the timestamp from a pre-built mapping relationship between timestamps and video context information based on the timestamp of the interactive content. The video context information includes: a mapping relationship between objects and video dialogue. An information detection unit is used to perform preset information detection on the interactive content and the video context information to obtain information detection results; A sequence determination unit is used to concatenate the interactive content ID, the interactive content, the video context information, and the information detection result into a feature sequence; The recognition unit is used to recognize the feature sequence and obtain the interactive content recognition result.

9. A computer storage medium, characterized in that, The computer storage medium stores at least one instruction, which, when executed by a processor, implements the interactive content recognition method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, The electronic device includes: a memory and a processor; The memory is used to store at least one instruction; The processor is used to execute the at least one instruction to implement the interactive content recognition method as described in any one of claims 1 to 7.

Images