Cross-video interaction data migration and video generation method based on AIGC prompt words

Abstract

Embodiments of the present disclosure provide an AIGC prompt word-based cross-video interaction data migration and video generation method, which is applied to the technical field of computer application; the method comprises: generating a video file in response to an AIGC prompt word input by a user; at the same time, performing multi-modal semantic retrieval based on the AIGC prompt word to recall historical videos similar in semantics; extracting interaction data associated with the historical videos to perform portrait label annotation, and performing matching filtering according to a target audience portrait; mapping original time stamps of the matching filtered interaction data to the AIGC video according to rhythm metadata of the historical videos and the AIGC video to generate a barrage configuration file of the AIGC video; and binding the video file and the barrage configuration file to form an AIGC video with a barrage, so as to realize personalized barrage display, and let the AIGC video go online with real barrages fitting the picture and the audience without manual intervention, solve the cold start problem, save computing power and operation cost, and improve the watching immersion and the complete play rate.

Description

Technical Field

[0001] This disclosure relates to the field of computer application technology, and in particular to a method for cross-video interactive data migration and video generation based on AIGC prompt words. Background Technology

[0002] With the rapid development of AI-Generated Content (AIGC) technology, especially the application of large-scale video generation models, platforms can quickly and massively generate entirely new video content based on user-input prompts. This technology significantly lowers the barrier to entry and cost of video content creation, improving content supply efficiency. However, these AIGC-generated videos generally face a cold start problem of "zero buzz and zero interaction" in their initial release. Due to the lack of real user comments, bullet screens, and other interactive data, the videos appear lifeless, resulting in low user completion and retention rates, severely impacting the content's dissemination and commercial value.

[0003] Current technologies utilize artificial intelligence to automatically generate "machine-generated bullet comments" or "bot comments," simulating user interaction through scripts to enliven the video's atmosphere. However, this method has significant drawbacks: AI-generated text typically originates from general corpora, lacking the emotional fluctuations, internet memes, community resonance, and real-time reactions to visual details experienced by real humans in specific contexts. This results in mechanical and rigid bullet comment content that users easily recognize and find objectionable, potentially degrading the user experience. Alternatively, existing interactive data (such as comments and bullet comments) from other similar videos can be directly copied or pasted into the new video. While utilizing real human interactive text, this approach leads to the following technical shortcomings in practical applications: Directly copied bullet comments usually retain their timestamps from the original video. However, since different videos vary in length, editing rhythm, and emotional climaxes, direct translation can lead to a significant time misalignment between bullet comments (such as "High energy ahead" or "Explosive video quality") and the actual content of the new video, resulting in an abrupt and uncomfortable viewing experience. The original interaction data often contains words that point to specific subjects or events in the original video, such as "The uploader is so handsome," "This video is really good," "Please update," and "Donate coins." These words have strong contextual dependence. If they are directly transferred to new videos generated by AIGC (such as a pure landscape video with no one on screen), it will produce serious semantic errors and a sense of incongruity, destroying the user's immersion. Existing migration or filling methods are mostly "full-volume, static" distribution models, meaning that all viewers see the same bullet comment content. This ignores the differences in interests, preferences, and cultural circles among different audience groups, resulting in monotonous bullet comment content that is difficult to resonate with specific viewers and fails to achieve personalized recommendations of interactive content. Summary of the Invention

[0004] This disclosure provides a method, apparatus, device, and storage medium for cross-video interactive data migration and video generation based on AIGC prompt words, which solves the technical problems of existing AIGC videos having no real interaction during cold start, stiff machine-generated bullet comments, semantic errors due to direct copying of bullet comments, timing misalignment, inability to personalize distribution, high latency in serial processing links, and difficulty in achieving immediate online release of generated videos.

[0005] According to a first aspect of this disclosure, a method for cross-video interactive data migration and video generation based on AIGC prompt words is provided, the method comprising: responding to AIGC prompt words input by a user, and generating a video file based on the AIGC prompt words; While generating video files, Multimodal semantic retrieval is performed based on the AIGC prompt words to recall historical videos that are semantically similar to the AIGC prompt words; Extract the interaction data associated with the historical videos, label them with profile tags, and then match and filter them according to the target audience profile of the current AIGC video; Based on the rhythm metadata of the historical video and the AIGC video, the original timestamps of the matched and filtered interactive data are mapped to the AIGC video to generate the bullet screen configuration file of the AIGC video. The video file is bound to the bullet screen configuration file to form an AIGC video with bullet screens.

[0006] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the step of performing multimodal semantic retrieval based on the AIGC prompt words to recall historical videos semantically similar to the AIGC prompt words includes: The AIGC prompts are input into a pre-trained embedding model to obtain a high-dimensional semantic feature vector; Based on the high-dimensional semantic feature vector, a similarity search is performed in a preset vector database to recall historical videos whose semantic similarity with the high-dimensional semantic feature vector meets a preset threshold, and to obtain the video identifier of each historical video; the vector database is used to store historical videos, the video semantic feature vectors corresponding to each historical video, and the unique video identifier corresponding to each historical video.

[0007] In addition to the aspects and any possible implementations described above, a further implementation is provided, wherein the extraction of interaction data associated with the historical video for profile tagging includes: Based on the video identifier, read the interaction data associated with the historical video, including bullet screen data and comment data; A weighted popularity score is calculated based on the interaction metrics associated with each interaction data point. The data is then filtered based on the weighted popularity score to obtain the interaction data after popularity filtering. The interaction metrics include the number of likes and replies of the original interaction data and the number of views of the historical videos. A lightweight intent classification model is used to identify contextual intent in the interaction data after popularity filtering. Text containing strong subject pronouns is removed, and general sentiment text without subject attachment is retained to obtain semantically purified interaction data. Based on the semantically purified interaction data, prompt words containing format-qualifying instructions are constructed. The prompt words and the semantically purified interaction data are then input into a large language model to obtain a tag array corresponding to each interaction data.

[0008] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the matching and filtering based on the target audience profile of the current AIGC video includes: The tag array is matched with the target audience profile of the current AIGC video to obtain the interaction data after profile matching.

[0009] In addition to the aspects and any possible implementations described above, a further implementation is provided in which the mapping of the original timestamps of the matched and filtered interactive data to the AIGC video, and the generation of the AIGC video's bullet screen configuration file, includes: Multimodal audio-visual rhythm metadata is extracted from the historical video and the AIGC video respectively. The multimodal audio-visual rhythm metadata includes visual climax points determined by analyzing the intensity of image changes between adjacent frames of the video and auditory climax points determined by analyzing the audio energy intensity at each time point in the video audio track. Based on the multimodal audio-visual rhythm metadata of the historical video, a preset climax interval of the historical video is determined; based on the multimodal audio-visual rhythm metadata of the AIGC video, the corresponding climax interval of the AIGC video is determined. Obtain the original timestamps corresponding to each interaction data in the interaction data after profile matching, and determine whether the original timestamps fall within the preset climax range of the historical video; If it falls within the climax range, a segmented interval interpolation algorithm is used to map the original timestamp to the corresponding climax range of the AIGC video to obtain a new timestamp after mapping. If it does not fall within the climax range, a new timestamp is calculated by using a linear scaling method based on the total duration of historical videos, the original timestamp, and the total duration of AIGC videos. Based on the mapped new timestamp, the corresponding bullet screen text, and the display track parameters, a bullet screen configuration file for the AIGC video is generated.

[0010] In addition to the aspects and any possible implementations described above, a further implementation is provided in which binding the video file with the bullet comment configuration file to form an AIGC video with bullet comments includes: Upload the video file to the object storage service to obtain the video storage address; Write the bullet screen configuration file into the cache database to obtain the configuration file storage identifier; Bind the video storage address and the configuration file storage identifier to generate a unique corresponding mapping relationship; When a video playback request is received, the video file and the bullet screen configuration file are called simultaneously according to the mapping relationship. Based on the video file and the bullet screen configuration file, render the video layer and the bullet screen layer to obtain an AIGC video with bullet screens.

[0011] In addition to the aspects and any possible implementations described above, a further implementation is provided in which, after binding the video file with the bullet screen configuration file, the method further includes: In response to an end user's request to play an AIGC video, the system obtains the end user's personal historical preference characteristics. Based on the personal historical preference characteristics, the bullet screen data in the bullet screen configuration file is filtered to obtain bullet screen data that matches the personal historical preference characteristics; The selected bullet screen data is sent to the player of the terminal user and rendered and displayed synchronously with the video file.

[0012] According to a second aspect of this disclosure, an apparatus for cross-video interactive data migration and video generation based on AIGC prompts is provided. The apparatus includes: a video rendering module, configured to generate a video file based on AIGC prompts input by a user; The recall module is used to generate video files while performing multimodal semantic retrieval based on the AIGC prompt words to recall historical videos that are semantically similar to the AIGC prompt words; The filtering module is used to extract the interaction data associated with the historical videos, label them with profile tags, and perform matching and filtering based on the target audience profile of the current AIGC video. The bullet screen configuration module is used to map the original timestamps of the matched and filtered interactive data to the AIGC video based on the rhythm metadata of the historical video and the AIGC video, and generate the bullet screen configuration file of the AIGC video. The video generation module is used to bind the video file with the bullet screen configuration file to form an AIGC video with bullet screens.

[0013] According to a third aspect of this disclosure, an electronic device is provided. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described above.

[0014] According to a fourth aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon that, when executed by a processor, implements the methods according to the first and / or second aspects of this disclosure.

[0015] In this disclosure, in response to user-inputted AIGC prompts, a video file is generated based on the AIGC prompts. Simultaneously, multimodal semantic retrieval is performed based on the AIGC prompts to recall historical videos semantically similar to the AIGC prompts. Interaction data associated with historical videos is extracted and labeled with profile tags, and then matched and filtered according to the target audience profile of the current AIGC video. Based on the rhythm metadata of historical videos and AIGC videos, the original timestamps of the matched and filtered interaction data are mapped to the AIGC video to generate a bullet screen configuration file for the AIGC video. The video file and the bullet screen configuration file are bound together to form an AIGC video with bullet screens. In this way, the AIGC video is launched with an immersive interactive atmosphere adapted to different viewers, significantly improving video completion rate, user retention, and content dissemination conversion effect.

[0016] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0017] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein: Figure 1 A flowchart is shown for a cross-video interactive data migration and video generation method based on AIGC prompt words according to an embodiment of the present disclosure; Figure 2 A block diagram of an apparatus for cross-video interactive data migration and video generation based on AIGC prompt words according to an embodiment of the present disclosure is shown; Figure 3 A block diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0019] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0020] In this disclosure, in response to user-inputted AIGC prompts, a video file is generated based on the AIGC prompts. Simultaneously, multimodal semantic retrieval is performed based on the AIGC prompts to recall historical videos semantically similar to the AIGC prompts. Interaction data associated with historical videos is extracted and labeled with profile tags, and then matched and filtered according to the target audience profile of the current AIGC video. Based on the rhythm metadata of historical videos and AIGC videos, the original timestamps of the matched and filtered interaction data are mapped to the AIGC video to generate a bullet screen configuration file for the AIGC video. The video file and the bullet screen configuration file are bound together to form an AIGC video with bullet screens. In this way, the AIGC video is launched with an immersive interactive atmosphere adapted to different viewers, significantly improving video completion rate, user retention, and content dissemination conversion effect.

[0021] Figure 1 A flowchart of a cross-video interactive data migration and video generation method 100 based on AIGC prompt words according to an embodiment of the present disclosure is shown; as follows: Figure 1 As shown, method 100 includes: S101, responding to the AIGC prompt words input by the user, generates a video file based on the AIGC prompt words; In one possible implementation, the system receives user creation instructions pushed from the front-end console, parses the core AIGC prompt word "Prompt" within the instruction, and simultaneously reads the video constraint parameters selected by the creator. The parsed standardized prompt text and all video constraint parameters are encapsulated into a structured task data package, which is then delivered to the AIGC video rendering service cluster via the distributed task queue Celery. The AIGC video rendering service cluster calls the large video generation model to complete the generation of sequence frames, rendering of visual effects, synthesis of audio background music, and generation of subtitle base based on the prompt words in the task data package, continuously outputting intermediate rendering frame streams. The rendering process is a time-consuming serial processing flow with a large number of reusable blank waiting windows. After all frames and audio tracks have been rendered and synthesized, a complete video file in a standardized encapsulation format is exported, uploaded to an object storage service, and a unique video storage address and a globally unique video identifier (VideoID) are generated, completing the video file generation process.

[0022] Meanwhile, the main thread adopts a non-blocking asynchronous I / O mechanism. At the same time as executing the delivery rendering task, it triggers an asynchronous data retrieval thread in parallel, synchronously starts a multimodal semantic retrieval link, and reuses the video rendering waiting window to process bullet screen interaction data, so as to realize asynchronous parallelism and no time loss between video generation and bullet screen preprocessing.

[0023] For example, the constraint parameters include video resolution, total duration, visual style, shot type, background music type, target audience tags, video frame rate, and rendering quality level; While generating video files, S102, perform multimodal semantic retrieval based on AIGC prompts to recall historical videos that are semantically similar to AIGC prompts; In one possible implementation, AIGC prompts are input into a pre-trained embedding model to obtain high-dimensional semantic feature vectors. Based on the high-dimensional semantic feature vectors, a similarity search is performed in a pre-defined vector database to recall historical videos whose semantic similarity to the high-dimensional semantic feature vectors meets a pre-defined threshold, and the video identifiers of each historical video are obtained. The vector database is used to store historical videos, the video semantic feature vectors corresponding to each historical video, and the unique video identifiers corresponding to each historical video.

[0024] For example, after the main thread submits the AIGC video rendering task, it builds a non-blocking asynchronous IO link based on the Celery distributed task queue and Kafka message queue. The standardized serialized AIGC prompts are encapsulated into independent retrieval messages and pushed to a dedicated retrieval topic. This does not block the main video rendering thread, reusing the video rendering downtime, and initiates the retrieval process. The multimodal retrieval microservice consumes the prompt messages from the Kafka queue, calls a pre-trained streaming embedding model (bge-large-zh) to perform semantic encoding on the complete prompt text, and outputs a fixed-dimensional high-dimensional semantic feature vector. The generated high-dimensional semantic vector is then input... The data is fed into the Milvus vector database, which is indexed by HNSW. The database pre-stores semantic feature vectors and globally unique VideoIDs corresponding to all historical videos. The matching score between the current prompt word vector and the vectors of each historical video is calculated based on the cosine similarity algorithm. A fixed semantic similarity threshold is set to filter out irrelevant historical videos with similarity below the threshold. The videos are sorted from high to low similarity, and the top 5 to top 10 highly relevant historical videos are extracted to generate a structured video list. The unique VideoID corresponding to each historical video in the list is extracted, packaged into a retrieval result data package, and sent to the downstream interactive data cleaning module to complete the similar historical video retrieval process.

[0025] S103, extract the interaction data associated with historical videos, label them with profile tags, and match and filter them according to the target audience profile of the current AIGC video; In one possible implementation, interactive data associated with historical videos is read based on video identifiers. This interactive data includes bullet screen data and comment data. A weighted popularity score is calculated based on the interactive metrics associated with each interactive data point. The data is then filtered based on the weighted popularity score to obtain the popularity-filtered interactive data. The interactive metrics include the number of likes, replies, and views of historical videos from the original interactive data. A lightweight intent classification model is used to identify the contextual intent of the popularity-filtered interactive data, removing text containing strong subject pronouns and retaining general sentiment text without subject attachments to obtain semantically purified interactive data. Based on the semantically purified interactive data, prompt words containing format-qualifying instructions are constructed. The prompt words and the semantically purified interactive data are input together into a large language model to obtain a tag array corresponding to each interactive data point.

[0026] In one possible implementation, the tag array is matched with the target audience profile of the current AIGC video to obtain the interaction data after profile matching.

[0027] For example, according to the unique VideoIDs of a plurality of historical videos, a distributed database connection pool is used to access the TiDB / MySQL persistent storage repository in parallel, so as to batch read all original interaction data of bullet comments and comments bound to the corresponding videos. The text, number of likes, number of replies, total video play volume, and original publishing timestamp are encapsulated and stored in the in-memory original text queue pool. Meanwhile, preliminary deduplication is performed on the original data to eliminate completely identical texts, so as to reduce subsequent computing overhead.

[0028] For example, the Aho-Corasick automaton algorithm and regular expression streaming scanning are adopted to eliminate web links, HTML tags, user reply prefixes, and special system emoticon codes. A hard threshold for character length is set to directly discard meaningless short sentences with a length of less than 2 characters and ultra-long occluding texts with a length of more than 30 characters. A popularity weighted scoring formula is constructed: the likes and replies of a single interaction are taken as positive weights, and the total play volume of the original video is used as the base for smoothing, so as to reduce the scoring deviation caused by the natural data base of top hit videos. A dynamic popularity threshold is set to eliminate plain interaction texts with scores below the threshold and no group resonance, and retain high-popularity highlight content.

[0029] For example, a strong-subject negative vocabulary list (UP主, blogger, this episode, urging for updates, tipping coins, showing face, etc.) is pre-constructed, and the syntactic dependency analysis is used to identify the attachment relationship of the subject / object of the text. The popularity-filtered text is sent to a fine-tuned lightweight intent classification model. If the text is bound to a specific blogger, an exclusive event of the original video, or a personal persona reference, it is determined as a non-compliant text and directly eliminated. Only texts without subject attachment, such as pure scene praise, general emotional venting, and universal internet memes, are retained. Privacy desensitization is performed on the retained texts to erase sensitive information such as the original publishing user UID, nickname, and geographic location, so as to generate independent virtual bullet comment instances and complete semantic purification.

[0030] For example, standardized structured prompts are constructed to restrict the large model to output a JSON-format tag array, and the tag dimensions include three categories: crowd circle, content style, and emotional attribute. All purified interaction texts are batch input into the local quantized lightweight large model Qwen / GPT-4o-mini, and multi-dimensional crowd tags corresponding to each bullet comment (such as [Visual Enthusiast], [ACG Enthusiast], [Tech Savvy], [Comedy Oriented]) are generated one by one. The bullet comment text, popularity score, tag array, and original timestamp are bound and stored in the Redis high-speed cache to form a tagged bullet comment material pool.

[0031] For example, the target audience tag set preset in the AIGC video generation stage is read. The tag sources are divided into two types: one is automatically predicted and generated based on the semantics of AIGC prompt words, and the other is manually selected and configured by the creator on the front end. A two-dimensional matching matrix is ​​constructed with "bullet screen built-in tag set and video target audience tag set", and the intersection matching degree of the two is calculated. A minimum matching degree threshold is set, and only bullet screens whose intersection meets the threshold are retained. The exclusive bullet screen subset after profile matching and filtering is output and sent to the subsequent time-series mapping module.

[0032] S104: Based on the rhythm metadata of historical videos and AIGC videos, the original timestamps of the matched and filtered interactive data are mapped to AIGC videos to generate the bullet screen configuration file for AIGC videos. In one possible implementation, multimodal audio-visual rhythm metadata is extracted from historical videos and AIGC videos respectively. This metadata includes visual climaxes determined by analyzing the intensity of image changes between adjacent frames of the video and auditory climaxes determined by analyzing the audio energy intensity at each time point in the video's audio track. Based on the multimodal audio-visual rhythm metadata of the historical videos, a preset climax interval is determined. Based on the multimodal audio-visual rhythm metadata of the AIGC videos, the corresponding climax interval of the AIGC videos is determined. The original timestamps corresponding to each interaction data point in the profile-matched interaction data are obtained, and it is determined whether the original timestamps fall within the preset climax interval of the historical videos. If they fall within the climax interval, a segmented interval interpolation algorithm is used to map the original timestamps to the corresponding climax interval of the AIGC videos, resulting in a new mapped timestamp. If they do not fall within the climax interval, a linear proportional mapping is used to calculate the new mapped timestamp based on the total duration of the historical videos, the original timestamps, and the total duration of the AIGC videos. Based on the new mapped timestamps, the corresponding bullet screen text, and the display track parameters, a bullet screen configuration file for the AIGC videos is generated.

[0033] For example, for each recalled historical video: the video analysis component is invoked to calculate the intensity of changes between adjacent frames using video optical flow, marking visual climax points; the short-time energy spectrum of the audio track is analyzed to identify auditory climax points with sudden increases in audio loudness; all visual climax points and auditory climax points are integrated and divided into continuous intervals to obtain a set of climax intervals for historical videos, while simultaneously recording the total duration of historical videos and the original millisecond timestamps of all bullet comments. For the currently rendered AIGC video: the scene transition, special effects rendering, and BGM energy structured metadata exported during the rendering stage are reused, and the same optical flow and audio energy analysis logic is used to divide the climax intervals corresponding to the AIGC video, recording the complete total duration of the AIGC video. The total duration of both types of videos and the start and end millisecond timestamps of all climax intervals are encapsulated into a rhythm metadata data package and stored in memory for time-series mapping calculations.

[0034] For example, the interaction data after the profile matching filter is traversed one by one, and the original timestamp of the historical video bound to each bullet comment is extracted; the original timestamp is compared with the climax interval of the corresponding historical video one by one to determine whether the timestamp falls within the climax interval of any historical video, and two types of bullet comments are distinguished: emotional highlight bullet comments (falling into the climax interval) and regular prelude bullet comments (not falling into the climax interval).

[0035] For example, if the bullet comment is an emotional highlight (i.e., the original timestamp is located in the climax range of the historical video), a segmented interval interpolation nonlinear mapping algorithm is used to locate the historical climax range to which the bullet comment belongs, as well as the corresponding climax range with the highest emotional matching degree in the AIGC video. Interpolation is performed according to the relative offset ratio within the interval, and the bullet comment is forcibly projected into the matching climax range of the AIGC video to ensure that the highlight bullet comment is accurately aligned with the audiovisual explosion scene of the new video. If the bullet comment is a regular prelude (i.e., the original timestamp is not in the climax range of the historical video), a global linear scaling algorithm is used to scale it proportionally according to the overall duration to adapt to the basic timeline of the new video.

[0036] For example, the calculation formula is: New timestamp = Original timestamp × Total duration of AIGC videos ÷ Total duration of historical videos.

[0037] For example, perturbation optimization is added to the new millisecond-level timestamp for completing the mapping calculation: randomly assign the horizontal scrolling track_id of the bullet comments to avoid multiple bullet comments overlapping and obscuring the screen at the same time; standardize the basic parameters of bullet comment font size, scrolling speed, and transparency; and bind the cleaned bullet comment text without any flaws, the unique ID of the virtual bullet comment, the circle tag, and the popularity score.

[0038] For example, all bullet screen data that have completed timestamp mapping and parameter configuration are uniformly organized, sorted in ascending order by playback time, and encapsulated into a streaming standard JSON format file. Each entry in the file contains structured fields such as time_offset_ms, track_id, bullet screen text, tags, and popularity, resulting in a unique bullet screen configuration file adapted to the current AIGC video. This configuration file is then written to the Redis cache and a unique configuration identifier is generated.

[0039] S105 binds the video file with the bullet screen configuration file to create an AIGC video with bullet screens.

[0040] In one possible implementation, the video file is uploaded to an object storage service to obtain the video storage address; the bullet screen configuration file is written to a cache database to obtain the configuration file storage identifier; the video storage address and the configuration file storage identifier are bound together to generate a unique mapping relationship; when a video playback request is received, the video file and the bullet screen configuration file are called simultaneously according to the mapping relationship; based on the video file and the bullet screen configuration file, the video layer and the bullet screen layer are rendered to obtain an AIGC video with bullet screens.

[0041] For example, after the AIGC rendering cluster completes video compositing and export, the distributed main control program pushes the complete video file to the object storage service for persistent storage. After storage, the object storage returns a globally unique video access address and a unique video identifier (VideoID), which the system then associates and caches. The generated JSON format bullet screen configuration file is retrieved and written to the Redis cluster cache. The cache service assigns a unique identifier (ConfigID) to the configuration file, and the cache synchronously retains all structured data of the bullet screen, ensuring millisecond-level reads during playback requests. Using VideoID as the primary key, the video storage address, ConfigID, and basic video attributes (duration, resolution, target audience tags) are encapsulated into a binding mapping record and written to the distributed persistent database. Atomic transaction operations are used to ensure a one-to-one correspondence between the video and bullet screen configurations, avoiding anomalies such as mismatched or lost video and bullet screen configurations. An externally callable video resource index is generated based on the binding mapping record. The index contains both the video playback link and the bullet screen configuration reading link, and is uniformly marked as an AIGC video resource with bullet screens, completing the integration at the resource level. For example, when the terminal initiates the video playback request, the streaming media server can quickly query the mapping record based on the VideoID and distribute the video slice stream and bullet screen configuration segment data in parallel to provide data support for the player's layered rendering.

[0042] In one possible implementation, in response to an end user's request to play an AIGC video, the system obtains the end user's personal historical preference characteristics; based on these characteristics, it filters the bullet comment data in the bullet comment configuration file to obtain bullet comment data that matches the personal historical preference characteristics; and it sends the filtered bullet comment data to the end user's player for synchronous rendering and display with the video file.

[0043] For example, when an end user initiates an AIGC video playback request, the streaming media gateway receives the request message and parses the globally unique user UID, the target AIGC video VideoID, the terminal device model, and the network transmission protocol identifier (HTTP / 2 or WebSocket) carried in the message. The gateway then accesses the user profile Redis cache cluster with the user UID and reads the set of historical preference features accumulated by the end user. The preference features include the user's long-term viewed video circle tags, high-frequency interactive bullet screen types, preferred emotional style, frequently viewed content themes, and historical bullet screen click / stay behavior weights. Simultaneously, based on the VideoID, the gateway queries the established binding mapping relationship, reads the original structured data of the complete bullet screen configuration file of the video from the cache, and temporarily loads it into the memory processing queue.

[0044] For example, three types of tag sets are extracted: the circle / style tags carried by the bullet comments themselves; the target audience tags preset by AIGC videos; and the personal historical preference tags of the current terminal user. A three-dimensional tag matching matrix is ​​constructed, and the number of overlapping intersections of the three types of tags in a single bullet comment is calculated. A matching threshold is preset, and only bullet comments with an intersection number reaching the threshold are retained. An additional secondary weighted filtering is applied: the original interaction popularity score of the bullet comments is combined with a weighted sorting to remove bullet comments that meet the matching criteria but have extremely low popularity and no resonance value, so as to prevent low-quality bullet comments from crowding the playback screen. Finally, a customized bullet comment dataset for this user is generated.

[0045] For example, the filtered exclusive bullet comments are reordered in ascending order according to the video playback timestamp time_offset_ms, and split into multiple lightweight segmented data packets to adapt to the streaming media chunked transmission logic; the bullet comment segmented data is asynchronously sent to the terminal player through a WebSocket long connection or an HTTP / 2 multiplexed channel, and the sending timing is executed in parallel with the video slice stream transmission, without blocking the video loading; the transmission process carries the rendering parameters of each bullet comment: scroll track_id, display duration, transparency, and font size, saving the terminal secondary parsing and calculation overhead.

[0046] For example, the player internally constructs a two-layer independent rendering architecture: the bottom hardware decoding layer is responsible for decoding and outputting the video image, while the upper WebGL / Canvas lightweight bullet screen rendering layer independently manages the bullet screen text drawing logic; the player has a built-in global clock counter that synchronizes the video playback progress millisecond timestamps in real time; when the clock value matches the bullet screen time_offset_ms, the upper rendering layer is scheduled to allocate the corresponding bullet screen text to an idle track for horizontal scrolling display; the multi-layer layers are refreshed synchronously in real time, eliminating the time difference between the video image and the bullet screen scrolling, and realizing synchronous output of audio, video and bullet screen; when users with different preferences open the same AIGC video, the terminal only renders the bullet screen content that matches their own interests, achieving a personalized interactive display effect.

[0047] For example, when a user inputs an AIGC prompt related to anime / manga scenery in the front-end console, the system simultaneously launches two asynchronous parallel threads. The main thread encapsulates the prompt with constraints such as resolution, video duration, and anime / manga target audience tags into a task package and delivers it to the AIGC rendering cluster to generate a video file. The asynchronous retrieval thread simultaneously sends the serialized prompt into the bge-large-zh embedding model to generate a high-dimensional semantic vector. Similarity retrieval is then performed in the Milvus vector library with an HNSW index, recalling the top 8 historical videos of the same theme that meet the similarity criteria and obtaining their corresponding VideoIDs. Based on the VideoIDs, the system batch-reads the original interactive data of each video's bullet comments and comments from the database. First, it uses an AC automaton and regular expressions to clean links, special characters, and invalid reply prefixes, filtering texts shorter than 2 characters and longer than 30 characters. It then combines likes, replies, and a weighted popularity score based on the original video's play count to filter out low-resonance content. Next, a lightweight intent classification model is used to perform syntactic dependency analysis, removing texts containing strong subject references such as "UP master" or "urging for updates," and completing user privacy desensitization. The system calls the Qwen large model to output a JSON tag array of circle, style, and emotion for each purified bullet comment. The bullet comment tag set is then matched and filtered with the preset anime audience tags of the current AIGC video to obtain a bullet comment material pool adapted to the video positioning. Next, the system extracts the optical flow visual climax, audio short-time energy auditory burst rhythm metadata of historical videos and AIGC videos to be released, divides the emotional climax intervals of the two videos, and determines the original timestamp of each bullet comment. Bullets falling into the climax interval of the historical video are non-linearly mapped to the corresponding highlight interval of the AIGC video using segmented interval interpolation. The remaining bullet comments are converted to new millisecond timestamps using a global duration linear ratio. After randomly assigning a scrolling track to each bullet comment, they are integrated to generate a standardized JSON bullet comment configuration file and stored in Redis cache. After the AIGC video is rendered and uploaded to object storage to obtain the video access address, the system uses VideoID as the primary key to bind the video storage address with the unique ConfigID of the bullet comment configuration file through atomic transactions to generate a global mapping relationship, forming a complete AIGC video resource with bullet comments.When a user interested in anime and manga initiates a video playback request, the streaming media gateway parses the user's UID, retrieves their historical preferences for anime and manga elements and special effects, and performs a three-dimensional intersection matching of user preference tags, video audience tags, and the bullet comment's own tags, combined with a secondary filtering based on popularity. This generates a user-specific bullet comment data packet, which is then sent to the terminal player in parallel with the video slice stream via HTTP / 2. The player operates synchronously using the underlying video decoding layer and the upper-layer WebGL bullet comment rendering layer, relying on its built-in millisecond clock to match the bullet comment timestamps to complete layered scrolling rendering. Ultimately, this allows for differentiated and adapted bullet comments to be displayed when users with different interests watch the same AIGC video. The entire bullet comment migration preprocessing process is asynchronously completed during the rendering idle window. This addresses multiple shortcomings of AIGC new videos, such as zero interaction during cold starts, stiff machine-generated bullet comments, semantic glitches in cross-video bullet comments, misaligned bullet comment timing, and a lack of resonance among different user groups due to unified bullet comment distribution. Furthermore, it eliminates the need for manual intervention, ensuring simultaneous loading and smooth playback of the video and bullet comments, improving user immersion, completion rate, and retention.

[0048] In this disclosure, in response to user-inputted AIGC prompts, a video file is generated based on the AIGC prompts. Simultaneously, multimodal semantic retrieval is performed based on the AIGC prompts to recall historical videos semantically similar to the AIGC prompts. Interaction data associated with historical videos is extracted and labeled with profile tags, and then matched and filtered according to the target audience profile of the current AIGC video. Based on the rhythm metadata of historical videos and AIGC videos, the original timestamps of the matched and filtered interaction data are mapped to the AIGC video to generate a bullet screen configuration file for the AIGC video. The video file and the bullet screen configuration file are bound together to form an AIGC video with bullet screens. In this way, the AIGC video is launched with an immersive interactive atmosphere adapted to different viewers, improving user retention and content dissemination conversion effects, realizing personalized bullet screen display, and allowing the AIGC video to have real bullet screens that fit the picture and the audience without manual intervention. This solves the cold start problem, saves computing power and operating costs, and improves the viewing immersion and completion rate.

[0049] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.

[0050] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.

[0051] Figure 2 A block diagram of an AIGC-based cross-video interactive data migration and video generation apparatus 200 according to an embodiment of the present disclosure is shown. Figure 2 As shown, the device 200 includes: The video rendering module 201 is used to generate a video file in response to the AIGC prompt words input by the user; The recall module 202 is used to generate video files while performing multimodal semantic retrieval based on the AIGC prompt words to recall historical videos that are semantically similar to the AIGC prompt words. The filtering module 203 is used to extract the interaction data associated with the historical videos, label them with profile tags, and perform matching and filtering based on the target audience profile of the current AIGC video. The bullet screen configuration module 204 is used to map the original timestamp of the matched and filtered interactive data to the AIGC video according to the rhythm metadata of the historical video and the AIGC video, and generate the bullet screen configuration file of the AIGC video. The video generation module 205 is used to bind the video file with the bullet screen configuration file to form an AIGC video with bullet screens.

[0052] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0053] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0054] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0055] Figure 3 A schematic block diagram of an electronic device 300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0056] Electronic device 300 includes a computing unit 301, which can perform various appropriate actions and processes according to a computer program stored in ROM 302 or a computer program loaded into RAM 303 from storage unit 308. RAM 303 can also store various programs and data required for the operation of electronic device 300. The computing unit 301, ROM 302, and RAM 303 are interconnected via bus 304. I / O interface 305 is also connected to bus 304.

[0057] Multiple components in electronic device 300 are connected to I / O interface 305, including: input unit 306, such as keyboard, mouse, etc.; output unit 307, such as various types of displays, speakers, etc.; storage unit 308, such as disk, optical disk, etc.; and communication unit 309, such as network card, modem, wireless transceiver, etc. Communication unit 309 allows electronic device 300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0058] The computing unit 301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 301 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 308. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 300 via ROM 302 and / or communication unit 309. When the computer program is loaded into RAM 303 and executed by the computing unit 301, one or more steps of method 100 described above may be performed. Alternatively, in other embodiments, the computing unit 301 may be configured to perform method 100 by any other suitable means (e.g., by means of firmware).

[0059] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0060] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0061] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium 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 fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0062] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including voice input, speech input, or tactile input).

[0063] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0064] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0065] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0066] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for cross-video interactive data migration and video generation based on AIGC prompts, characterized in that, include: In response to user-inputted AIGC prompts, a video file is generated based on the AIGC prompts. While generating video files, Multimodal semantic retrieval is performed based on the AIGC prompt words to recall historical videos that are semantically similar to the AIGC prompt words; Extract the interaction data associated with the historical videos, label them with profile tags, and then match and filter them according to the target audience profile of the current AIGC video; Based on the rhythm metadata of the historical video and the AIGC video, the original timestamps of the matched and filtered interactive data are mapped to the AIGC video to generate the bullet screen configuration file of the AIGC video. The video file is bound to the bullet screen configuration file to form an AIGC video with bullet screens.

2. The method according to claim 1, characterized in that, The step of performing multimodal semantic retrieval based on the AIGC prompt words to recall historical videos semantically similar to the AIGC prompt words includes: The AIGC prompts are input into a pre-trained embedding model to obtain a high-dimensional semantic feature vector; Based on the high-dimensional semantic feature vector, a similarity search is performed in a preset vector database to recall historical videos whose semantic similarity with the high-dimensional semantic feature vector meets a preset threshold, and to obtain the video identifier of each historical video; the vector database is used to store historical videos, the video semantic feature vectors corresponding to each historical video, and the unique video identifier corresponding to each historical video.

3. The method according to claim 2, characterized in that, The step of extracting and tagging the interaction data associated with the historical videos includes: Based on the video identifier, read the interaction data associated with the historical video, including bullet screen data and comment data; A weighted popularity score is calculated based on the interaction metrics associated with each interaction data point. The data is then filtered based on the weighted popularity score to obtain the interaction data after popularity filtering. The interaction metrics include the number of likes and replies of the original interaction data and the number of views of the historical videos. A lightweight intent classification model is used to identify contextual intent in the interaction data after popularity filtering. Text containing strong subject pronouns is removed, and general sentiment text without subject attachment is retained to obtain semantically purified interaction data. Based on the semantically purified interaction data, prompt words containing format-qualifying instructions are constructed. The prompt words and the semantically purified interaction data are then input into a large language model to obtain a tag array corresponding to each interaction data.

4. The method according to claim 3, characterized in that, The matching and filtering based on the target audience profile of the current AIGC video includes: The tag array is matched with the target audience profile of the current AIGC video to obtain the interaction data after profile matching.

5. The method according to claim 4, characterized in that, The step of mapping the original timestamps of the matched and filtered interactive data to the AIGC video to generate the bullet screen configuration file for the AIGC video includes: Multimodal audio-visual rhythm metadata is extracted from the historical video and the AIGC video respectively. The multimodal audio-visual rhythm metadata includes visual climax points determined by analyzing the intensity of image changes between adjacent frames of the video and auditory climax points determined by analyzing the audio energy intensity at each time point in the video audio track. Based on the multimodal audio-visual rhythm metadata of the historical video, a preset climax interval of the historical video is determined; based on the multimodal audio-visual rhythm metadata of the AIGC video, the corresponding climax interval of the AIGC video is determined. Obtain the original timestamps corresponding to each interaction data in the interaction data after profile matching, and determine whether the original timestamps fall within the preset climax range of the historical video; If it falls within the climax range, a segmented interval interpolation algorithm is used to map the original timestamp to the corresponding climax range of the AIGC video to obtain a new timestamp after mapping. If it does not fall within the climax range, a new timestamp is calculated by using a linear scaling method based on the total duration of historical videos, the original timestamp, and the total duration of AIGC videos. Based on the mapped new timestamp, the corresponding bullet screen text, and the display track parameters, a bullet screen configuration file for the AIGC video is generated.

6. The method according to claim 5, characterized in that, The step of binding the video file with the bullet comment configuration file to form an AIGC video with bullet comments includes: Upload the video file to the object storage service to obtain the video storage address; Write the bullet screen configuration file into the cache database to obtain the configuration file storage identifier; Bind the video storage address and the configuration file storage identifier to generate a unique corresponding mapping relationship; When a video playback request is received, the video file and the bullet screen configuration file are called simultaneously according to the mapping relationship. Based on the video file and the bullet screen configuration file, render the video layer and the bullet screen layer to obtain an AIGC video with bullet screens.

7. The method according to claim 1, characterized in that, After binding the video file with the bullet screen configuration file, the method further includes: In response to an end user's request to play an AIGC video, the system obtains the end user's personal historical preference characteristics. Based on the personal historical preference characteristics, the bullet screen data in the bullet screen configuration file is filtered to obtain bullet screen data that matches the personal historical preference characteristics; The selected bullet screen data is sent to the player of the terminal user and rendered and displayed synchronously with the video file.

8. A cross-video interactive data migration and video generation device based on AIGC prompt words, characterized in that, include: The video rendering module is used to generate a video file in response to the AIGC prompt words input by the user; The recall module is used to generate video files while performing multimodal semantic retrieval based on the AIGC prompt words to recall historical videos that are semantically similar to the AIGC prompt words; The filtering module is used to extract the interaction data associated with the historical videos, label them with profile tags, and perform matching and filtering based on the target audience profile of the current AIGC video. The bullet screen configuration module is used to map the original timestamps of the matched and filtered interactive data to the AIGC video based on the rhythm metadata of the historical video and the AIGC video, and generate the bullet screen configuration file of the AIGC video. The video generation module is used to bind the video file with the bullet screen configuration file to form an AIGC video with bullet screens.

9. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 7.

Images