Video polyphonic ringtone improving method based on large language model

By introducing large language model technology into the ring ring system, intelligently generating and recommending personalized video ring rings, users' dissatisfaction with personalized needs and the cumbersome process of ring ring production are solved, and user experience and operator market expansion are improved.

CN119025690APending Publication Date: 2024-11-26广州市申迪计算机系统有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411148696.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-21
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Users’ personalized needs for ringtones have not been fully met. The existing technology has problems such as homogeneity of content, single conception and insufficient novelty. At the same time, users’ participation in ringtone production and ordering process is cumbersome, and the content quality is not high.

Method used

Introduce large language model technology to optimize the ring ring system architecture, add human-computer interaction module, content generation module and user portrait module, and intelligently generate and recommend personalized video ring rings through multimedia large models, providing convenient interaction methods and personalized recommendations.

Benefits of technology

It improves the personalization and user experience of the ringtone, lowers the production threshold, and enhances user participation and operator market expansion capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119025690A_ABST
    Figure CN119025690A_ABST
Patent Text Reader

Abstract

The invention aims to provide a video polyphonic ringtone improving method based on a large language model. The method comprises the following steps: optimizing a polyphonic ringtone system architecture according to a multimedia large model; the polyphonic ringtone content is intelligently generated according to user requirements; and intelligently recommending the generated polyphonic ringtone content to the user according to the multimedia large model. The generation type AI technology represented by the large model is introduced into the polyphonic ringtone service, key technical scheme optimization of polyphonic ringtone system upgrading, content creation and recommendation scenes is promoted, fusion innovation of the generation type AI technology, the polyphonic ringtone system, key scenes, key technologies and the like is deepened, more personalized polyphonic ringtone service new experience is provided for users, and the user experience is improved. And market expansion and value improvement of the coloring ring back tone service of an operator are facilitated.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a video color ringback tone enhancement method based on a large language model. Background Art

[0002] As a value-added service created by operators and combining the characteristics of operators and the Internet, CRBT has given new value and new business opportunities to the "golden 15 seconds" in call waiting scenarios. Currently, with the rapid development of 5G, video CRBT has gradually replaced audio CRBT and become an important carrier for personal social networking, brand communication, and public welfare publicity. At the same time, it has played an important role in improving user experience and driving operators' revenue growth.

[0003] However, as people's demand for digital content continues to increase, users also have higher expectations for ringback tones. On the one hand, users hope that the ringback tones they order will be more imaginative and personalized. Through research, it is found that due to the current restrictions on humanistic creativity, creative materials and tools, users are prone to encounter problems such as content homogeneity, single ideas and lack of novelty when choosing and participating in the production of ringback tones. Generative AI technologies such as big models can accurately capture the personalized needs of a large number of users and generate original and diverse audio and video content, which can effectively solve this development bottleneck on the supply side. On the other hand, users hope to use convenient interactive methods to participate in the production of ringback tones or service subscriptions. Although current operators have provided the "one-click generation" video ringback tones function, users still need to participate in tuning multiple times, and the quality of the generated video content is not high. The "human-like" interactive capabilities and powerful generation capabilities of generative AI technologies such as big models can effectively lower the production threshold and meet user needs.

[0004] Users hope that the ordered ringtones are more imaginative and personalized. However, due to the current limitations of humanistic creativity, creative materials and tools, users are prone to encounter problems such as content homogeneity, single ideas and lack of novelty when choosing and participating in the production of ringtones. On the other hand, users hope to use convenient interactive methods to participate in the production of ringtones or service subscriptions. Although current operators have provided a "one-click generation" video ringtone function, users still need to participate in tuning multiple times, and the quality of the generated video content is not high. Although the ringtone service has provided a personalized recommendation function, it is limited by factors such as algorithms, low activity of ringtone users, and inaccurate ringtone content labels, resulting in a low proportion of users clicking or watching the system-recommended ringtones, and the conversion rate of users choosing to order recommended ringtones has not increased significantly. Summary of the invention

[0005] The purpose of the present invention is to provide a method for improving video ringback tone based on a large language model. The method promotes the optimization of key technical solutions for ringback tone system upgrades, content creation and recommendation scenarios by introducing generative AI technology represented by a large model into the ringback tone service, deepens the integration and innovation of generative AI technology with the ringback tone system, key scenarios and key technologies, provides users with a more personalized new ringback tone service experience, and helps operators expand the market and enhance the value of their ringback tone services.

[0006] A video ringback tone enhancement method based on a large language model, comprising: Optimize the ringback tone system architecture based on the multimedia big model; Intelligently generate ringback tone content based on user needs; The generated ringback tone content is intelligently recommended to users based on the multimedia big model.

[0007] Preferably, the optimizing the color ring back tone system architecture according to the multimedia big model comprises: Add human-computer interaction module, ring tone content generation module and ring tone user portrait module to the ring tone system architecture; The human-computer interaction module is used to obtain user needs and push video ringback tone according to user needs; The color ring back tone content generation module is used to generate a video color ring back tone according to user needs; The color ring back tone user portrait module is used to classify and mark users according to their needs.

[0008] Preferably, the intelligent generation of color ring back tone content according to user needs includes: Import all user requirements; Pre-process all user needs; Generate ringback tone content based on pre-processed user needs.

[0009] Preferably, the import of all user requirements includes: Use human-computer interaction templates to obtain customer needs; Add 5G letter reading, 5G messaging and ringback tone video customer service functions to provide media display for users.

[0010] Preferably, the preprocessing of all user requirements includes: Convert customer requirements into text and normalize them; Filter out bad information from normalized customer demands; Categorize the filtered customer demands according to tracks, genres, singers, video themes and storylines.

[0011] Preferably, generating the color ring back tone content according to the pre-processed user demand includes: Convert user input requirements into keywords; Generate ring back tone content according to the keyword.

[0012] Preferably, generating the color ring back tone content according to the keyword includes: Based on the audio and video requirements, ringback tone scripts, and user portrait information in the keywords, a set of integrated audio and video ringback tones is intelligently generated, and the generated content is sent to the user side through the human-computer interaction module; Obtain different feedback from users based on the generated content.

[0013] Preferably, generating the color ring back tone content according to the keyword includes: Based on the audio and video requirements in the keywords and the key information of the user profile, audio ringback tone and video-only ringback tone are generated respectively, and sent to the terminal side in batches or all at once according to user needs; If the user issues a deterministic feedback or an optimized feedback on the generated audio or video ringback tone, the service flow corresponding to the deterministic feedback or the optimized feedback is entered until the user confirms that both the audio and video meet the requirements; After the audio and video intelligent fitting request is issued, the content generation module will call the multimedia big model again, and intelligently fit the audio and video selected by the user through the alignment algorithm and content consistency check technology, and output the video ringtone with integrated audio and video; If the user has optimization requirements for the fitted video ringback tone, the service will enter the optimization feedback process.

[0014] Preferably, the intelligently recommending the generated color ring back tone content to the user according to the multimedia macro model comprises: Use multimedia big models to accurately analyze user portraits and needs, quickly identify and mark ringback tone content, and intelligently predict user behavior; Calling the multimedia big model to intelligently analyze the user's basic profile and service usage frequency, and classify users into high-frequency users, users to be mined, and new users; Develop login-type entry push and proactive recommendation methods based on user type.

[0015] Preferably, the method of formulating login-type entry push and active recommendation according to user type includes: For high-frequency users, the multimedia big model analyzes usage habits, preferences, and historical choices, combines business browsing records and retrieval information data, and intelligently generates the next business demand forecast; Set up positive feedback, loyalty rewards or privileges based on predictive recommendations to continuously improve the accuracy of content recommendations; For users to be mined, we deeply mine user basic portraits, historical data and behavior pattern information based on multimedia big models, capture the potential interests of users in idols and music styles, and push strategies through login-type portals; When users log in to the APP, selected ringtones are recommended. If users click to watch, like, collect or order, positive feedback is formed, leading to conversion into high-frequency users. For new users, we use a large multimedia model to analyze registration information, first-time usage data, and basic questionnaire data, and push recent selections and popular thematic ringtones that match the style and content preferences.

[0016] The beneficial effects of the present invention are: 1. The present invention promotes the optimization of key technical solutions for ringback tone system upgrades, content creation and recommendation scenarios by introducing generative AI technology represented by big models into the ringback tone service, so as to better meet users' needs for personalized ringback tone. As the generative AI technology industry such as big models matures, multimodal understanding and cross-modal generation capabilities will continue to increase; 2. The present invention can utilize rich multimedia content, detailed user portraits, ringback tone service operations and other data to continuously promote the iterative optimization of big models, continuously deepen the integration and innovation of generative AI technology with ringback tone systems, key scenarios and key technologies, provide users with a more personalized new ringback tone service experience, and help operators expand the market and enhance the value of their ringback tone services. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention. In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor. Figure 1 A flow chart of a method for improving video ringback tone based on a large language model according to the present invention; Figure 2 It is a schematic diagram of the end-to-end architecture of the optimized color ring back tone intelligent enhancement system of the present invention; Figure 3 A schematic diagram of the color ring back tone service process of the present invention; Figure 4 A logical schematic diagram of the color ring back tone service of the present invention; Figure 5 A schematic diagram of the processing logic based on deterministic and optimized feedback of the present invention; Figure 6 A schematic diagram of the personalized content recommendation service process of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0020] In addition, the descriptions of "first", "second", etc. in the present invention are only used for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0021] Users hope that the ordered ringtones are more imaginative and personalized. However, due to the current limitations of humanistic creativity, creative materials and tools, users are prone to encounter problems such as content homogeneity, single ideas and lack of novelty when choosing and participating in the production of ringtones. On the other hand, users hope to use convenient interactive methods to participate in the production of ringtones or service subscriptions. Although current operators have provided a "one-click generation" video ringtone function, users still need to participate in tuning multiple times, and the quality of the generated video content is not high. Although the ringtone service has provided a personalized recommendation function, it is limited by factors such as algorithms, low activity of ringtone users, and inaccurate ringtone content labels, resulting in a low proportion of users clicking or watching the system-recommended ringtones, and the conversion rate of users choosing to order recommended ringtones has not increased significantly.

[0022] The present invention introduces generative AI technology represented by big models into the ringback tone service, promotes the optimization of key technical solutions for ringback tone system upgrades, content creation and recommendation scenarios, and better meets users' needs for personalized ringback tone. As the generative AI technology industry such as big models matures, multimodal understanding and cross-modal generation capabilities will continue to increase. The present invention can utilize rich multimedia content, detailed user portraits, ringback tone service operations and other data to continuously promote the iterative optimization of big models, and continuously deepen the integration and innovation of generative AI technology with ringback tone systems, key scenarios and key technologies, to provide users with a more personalized new ringback tone service experience, and help operators expand the market and enhance the value of their ringback tone services.

[0023] Example 1 A video ringback tone enhancement method based on a large language model, reference Figure 1 ,include: S100, optimizes the ringback tone system architecture based on the multimedia big model; S200, intelligently generates ringback tone content based on user needs; S300, intelligently recommend the generated ringback tone content to the user according to the multimedia macro model.

[0024] The present invention provides a method and device for improving video ringback tone based on a large language model. By introducing generative AI technology represented by a large model into the ringback tone service, the key technical solutions for ringback tone system upgrade, content creation and recommendation scenarios are optimized, and the integration and innovation of generative AI technology with the ringback tone system, key scenarios and key technologies are deepened to provide users with a more personalized new ringback tone service experience, thereby helping operators to expand the market and enhance the value of their ringback tone services.

[0025] Preferably, S100, optimizing the color ring back tone system architecture according to the multimedia big model includes: Add human-computer interaction module, ring tone content generation module and ring tone user portrait module to the ring tone system architecture; The human-computer interaction module is used to obtain user needs and push video ringback tones according to user needs; The color ring tone content generation module is used to generate video color ring tone according to user needs; The ringback tone user portrait module is used to classify and mark users according to their needs.

[0026] Currently, the CRBT system includes terminals, access networks, core networks, CRBT platforms, billing platforms and business support systems. The CRBT platform includes signaling modules, resource modules and control modules. In order to meet users' needs for personalized content and convenient participation in production, this solution, based on maximizing compatibility with CRBT systems and business models, introduces modules such as big models, connects the capability middle platform and computing resource pool, and improves the service experience. The end-to-end architecture of the optimized CRBT intelligent enhancement system is as follows: Figure 2 shown.

[0027] On the one hand, the CRBT platform is upgraded to an enhanced CRBT service platform. On the basis of the original CRBT platform signaling module, resource module and control module, an intelligent human-computer interaction module, CRBT content generation module, CRBT user portrait module and multimedia large model module are added. On the other hand, the enhanced CRBT service platform is directly connected with the capability center and computing resource pool to provide basic user portrait information for CRBT content generation and personalized recommendation, as well as computing resources required for large model training and reasoning.

[0028] Among them, the multimedia big model module mainly provides intelligence for intelligent human-computer interaction, ringtone content generation, and user ringtone portrait modules. Taking into account the current domestic multimedia big model technology and industry maturity, computing power resources, etc., the multimedia big model module can be a native multi-modal data parallel processing and generation capability of text, image, audio, video, etc. The media generative AI big model assumes the main function (hereinafter referred to as the "multimedia big model"), and can also be composed of multiple multimodal dedicated small models or large models. Taking the multimedia big model as an example, it can be based on the domestic and foreign open source multimodal basic general big model, and can be enhanced by injecting rich media data sets such as music, animation, movies and pictures, as well as audio and video ringtone resources, user ringtone portraits and service subscription information. The ringtone service data set is enhanced, fine-tuned and optimized to better adapt to the vertical segmentation needs of audio and video ringtone content generation. In order to improve the business experience by using generative AI technologies such as big models, combined with user needs, the key scenario technical solutions for interactive content creation and personalized content recommendation are sorted out.

[0029] Preferably, S200, intelligently generating color ring back tone content according to user demand includes: S210, import all user requirements; According to the system processing logic, interactive content creation mainly includes basic links such as user demand introduction, user demand refinement, intelligent generation of ringtone content and iterative optimization. The business process is as follows: Figure 3 shown.

[0030] S220, pre-processing all user requirements; The user demand extraction process involves demand translation, content review and preprocessing. Business logic such as Figure 4 shown.

[0031] S230, generating ring back tone content according to the pre-processed user demand.

[0032] Preferably, S210, importing all user requirements includes: Use human-computer interaction templates to obtain customer needs; Add 5G letter reading, 5G messaging and ringback tone video customer service functions to provide media display for users.

[0033] First, enrich the interactive mode. Aiming at the interactive experience needs of users of different ages, on the basis of existing services of ringback tone services such as web pages and APPs, 5G reading, 5G messaging and ringback tone video customer service are introduced to further expand service channels, provide users with more diversified and personalized content generation and interactive modes, and enhance the satisfaction and convenience of user experience. Among them, 5G reading and 5G messaging can provide rich media displays such as pictures, texts, audios, and videos compared to SMS, which is convenient for users to view generated content in real time and put forward optimization requirements; ringback tone video customer service relies on 5G new calls, intelligent human-computer interaction and large models and other modules to provide users with intelligent exclusive ringback tone customer service based on large models, as well as intelligent interactive functions such as real-time translation of multi-modal requirements, real-time viewing of generated content and feedback.

[0034] The second is to guide intelligent interaction. In order to improve the convenience of users' participation in content creation, the intelligent human-computer interaction module supports users to choose voice, text, audio, pictures and videos, and input requirements in any or according to established interaction templates. At the same time, it can also provide a variety of functions such as interaction mode based on large models, demand polishing, and auxiliary creation. For example, it provides a "one-click film" function that directly generates video ringtones based on pictures, voice, text, and videos, and a "divide and combine" function that generates audio and video based on templates such as "song name, singer, style, key lyrics" and then intelligently couples them, as well as a variety of auxiliary functions such as demand polishing, script writing, copywriting design, character customization, etc. based on large models.

[0035] Preferably, S220, preprocessing all user requirements includes: Convert customer requirements into text and normalize them; Filter out bad information from normalized customer demands; Categorize the filtered customer demands according to tracks, genres, singers, video themes and storylines.

[0036] The first is demand translation. Since the demands input by users may contain multiple types of data, and some voice and text have local characteristics, and raw data such as pictures and videos may also have problems such as low resolution and blur, when the intelligent translation module receives the demand, it first calls the multimedia large model to perform normalization processing such as voice-to-text, text formatting, audio and video file enhancement and standardization, and maps the raw data of different modes to the same or similar semantic space to achieve mutual understanding and alignment of multimodal demand information. If the user strictly enters the text demand according to the interactive prompt format, this intelligent translation process can be skipped; if the user chooses multiple modes and multiple rounds of input requirements, they can be translated one by one and entered in the next link in order, or they can be translated and merged in batches and imported into the content review and preprocessing link for centralized processing.

[0037] The second is content review. To ensure that the user input requirements and the generated ringtone content are compliant, a content review mechanism based on a large model is introduced to improve review efficiency and accuracy and reduce manual review costs. Among them, the content review module will call the language large model in the multimedia large model, and conduct intelligent review and bad information filtering on the translated user requirements in accordance with the established standards and rules of the country on content security review. After the user requirements are reviewed, they are imported into the demand preprocessing module for key information extraction.

[0038] The third is demand preprocessing. Considering that there are certain differences between the ringback tone requirements input by users and the machine language that the big model can efficiently understand and respond to, in order to reduce the understanding deviation of the big model, a demand preprocessing module is set in the intelligent human-computer interaction module. By calling the multimedia big model to perform intent mining, key information such as tracks, styles, singers, video themes and storylines are extracted, and user needs are efficiently converted into prompts that are easy for the big model to understand. In addition, in order to improve the understanding of users, after receiving the needs, the intelligent human-computer interaction module will actively retrieve the user portrait from the capability center, extract key information and include it in the prompt, and input it into the content generation module together.

[0039] Preferably, S230, generating the color ring back tone content according to the pre-processed user demand includes: S231, converting the requirements input by the user into keywords; S232, generating ring back tone content according to the keyword.

[0040] According to the "one-click to create a film" or "divide and combine" content generation mode selected by the user, and the prompts input by the demand preprocessing module, the content generation module calls the multimedia model to generate personalized ringback rings that meet the needs. To facilitate comparison and selection, the content generation module generates 4 ringback rings by default each time. If the user has special requirements for the number of ringback rings generated at one time, specific requirements can be input based on human-computer interaction to optimize.

[0041] Preferably, S232, generating the color ring back tone content according to the keyword includes: Based on the audio and video requirements, ringback tone scripts, and user portrait information in the keywords, a set of integrated audio and video ringback tones is intelligently generated, and the generated content is sent to the user side through the human-computer interaction module; Obtain different feedback from users based on the generated content.

[0042] The content generation module directly calls the multimedia big model, or calls the audio big model and the video big model in combination, and intelligently generates a set of integrated audio and video ringback rings based on the key information such as audio and video requirements, ringback ring scripts, and user portraits in the prompts, and sends the generated content to the user side through the human-computer interaction module. Users give different feedback based on the generated content, including deterministic feedback such as checking and liking, as well as optimization feedback such as input style conversion and regeneration. The processing logic based on deterministic and optimization feedback is as follows: Figure 5 shown.

[0043] The processing flow based on optimized feedback reuses the interactive content creation process and can be executed repeatedly according to user needs until the user selects the ringback tone that meets the needs and issues a confirmation feedback. At the same time, the human-computer interaction module will update the user's ringback tone portrait information and synchronize it to the capability center.

[0044] Preferably, S232, generating the color ring back tone content according to the keyword includes: Based on the audio and video requirements in the keywords and the key information of the user profile, audio ringback tone and video-only ringback tone are generated respectively, and sent to the terminal side in batches or all at once according to user needs; If the user issues a deterministic feedback or an optimized feedback on the generated audio or video ringback tone, the service flow corresponding to the deterministic feedback or the optimized feedback is entered until the user confirms that both the audio and video meet the requirements; After the audio and video intelligent fitting request is issued, the content generation module will call the multimedia big model again, and intelligently fit the audio and video selected by the user through the alignment algorithm and content consistency check technology, and output the video ringtone with integrated audio and video; If the user has optimization requirements for the fitted video ringback tone, the service will enter the optimization feedback process.

[0045] The content generation module calls the audio big model and the video big model separately, or calls the multimedia big model multiple times, and generates audio ringback tone and ringback tone containing only video respectively based on the audio and video requirements and key information such as user portrait in the prompt, and sends them to the terminal side in batches or together according to user requirements, waiting for user feedback. If the user sends deterministic feedback such as checking and liking the generated audio or video ringback tone, or inputs optimization feedback such as style conversion and regeneration, the service will enter the business process corresponding to the deterministic feedback or optimization feedback until the user confirms that both the audio and video meet the requirements and sends the audio and video intelligent fitting requirements. The content generation module will call the multimedia big model again, and intelligently fit the audio and video selected by the user through alignment algorithms and content consistency checks. Technology, output audio and video integrated video ringback tone. If the user has optimization requirements for the fitted video ringback tone, the service will enter the optimization feedback process. This optimization feedback process can be executed repeatedly until the user confirms that the video ringback tone meets the requirements.

[0046] Preferably, S300, intelligently recommending the generated color ring back tone content to the user according to the multimedia macro model includes: S310 uses a large multimedia model to accurately analyze user portraits and needs, quickly identify and mark ringback tone content, and intelligently predict user behavior; S320, calls the multimedia big model to intelligently analyze the user's basic profile and service usage frequency, and divides the users into high-frequency users, users to be mined, and new users; S330, formulate login-type entry push and active recommendation methods according to user type.

[0047] Currently, although the CRBT service has provided personalized recommendation functions, it is limited by factors such as algorithms, low activity of CRBT users, and inaccurate CRBT content labels, resulting in a low percentage of users clicking or watching the CRBT recommended by the system, and the conversion rate of users choosing to order recommended CRBTs has not increased significantly. In order to improve the activity rate of CRBT users and service satisfaction, the multimedia big model can be used to accurately analyze user portraits and needs, quickly identify and mark CRBT content, and intelligently predict user behavior to better understand users' potential needs and preferences and enhance the accuracy of CRBT recommendations. The personalized content recommendation service process is as follows: Figure 6 shown.

[0048] Preferably, S330, formulating the login-type entry push and active recommendation methods according to the user type includes: For high-frequency users, the multimedia big model analyzes usage habits, preferences, and historical choices, combines business browsing records and retrieval information data, and intelligently generates the next business demand forecast; Set up positive feedback, loyalty rewards or privileges based on predictive recommendations to continuously improve the accuracy of content recommendations; For users to be mined, we deeply mine user basic portraits, historical data and behavior pattern information based on multimedia big models, capture the potential interests of users in idols and music styles, and push strategies through login-type portals; When users log in to the APP, selected ringtones are recommended. If users click to watch, like, collect or order, positive feedback is formed, leading to conversion into high-frequency users. For new users, we use a large multimedia model to analyze registration information, first-time usage data, and basic questionnaire data, and push recent selections and popular thematic ringtones that match the style and content preferences.

[0049] The enhanced ringback tone service platform regularly uses the multimedia big model to intelligently analyze data such as user basic portraits and service usage frequency, and divides users into three categories: high-frequency users, users to be discovered, and new users. It formulates recommendation strategies such as login-type entry push and proactive recommendation to further improve service conversion rate.

[0050] Specifically, for high-frequency users, the multimedia big model analyzes data such as usage habits, preferences, and historical choices, and combines data such as business browsing records and retrieval information to intelligently generate the next business demand forecast, and set positive feedback, loyalty rewards or privileges based on the predicted recommendation to continuously improve the accuracy of content recommendations. For users to be mined, the multimedia big model deeply mines user basic portraits, historical data, behavior patterns and other information, accurately captures potential points of interest such as user idols and music styles, and uses a login-type entry push strategy. If a user logs in to the APP, a selected ringtone recommendation is provided, or the first new song ringtone of the favorite singer is regularly sent through SMS and 5G messages; if the user clicks to watch, likes, collects or orders, positive feedback is formed to guide them to gradually become high-frequency users. For new users, the multimedia big model analyzes registration information, first-time use data, basic questionnaires and other data, and pushes recent selections, hot topics and other ringtones that match the music style and content tendency to them, so as to deeply explore the user's individual needs and cultivate them into long-term loyal users. If a new user guidance process is set up after user registration, popular ringtones and questionnaires can be actively pushed through SMS, 5G messages, APP, etc., and ringtone portraits such as personal preferences can be widely collected to provide richer data information for personalized content recommendations of multimedia large models.

[0051] At the same time, taking into account that users' preferences and attitudes towards singers, music styles, themes, etc. may change at any time, a user demand feedback mechanism is established on the application side. Through the multimedia big model, user search keywords, viewing habits, and recommended content click-through rate, download rate, satisfaction and other indicators are analyzed in real time, and the intelligent recommendation algorithm based on the multimedia big model is further adjusted to continuously improve the accuracy of ringback tone recommendations and user satisfaction.

[0052] The present invention introduces generative AI technology represented by big models into the ringback tone service, promotes the optimization of key technical solutions for ringback tone system upgrades, content creation and recommendation scenarios, and better meets users' needs for personalized ringback tone. As the generative AI technology industry such as big models matures, multimodal understanding and cross-modal generation capabilities will continue to increase. The present invention can utilize rich multimedia content, detailed user portraits, ringback tone service operations and other data to continuously promote the iterative optimization of big models, and continuously deepen the integration and innovation of generative AI technology with ringback tone systems, key scenarios and key technologies, to provide users with a more personalized new ringback tone service experience, and help operators expand the market and enhance the value of their ringback tone services.

[0053] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be 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 present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A video ringback tone enhancement method based on a large language model, characterized in that: include: Optimize the ringback tone system architecture based on the multimedia big model; Intelligently generate ringback tone content based on user needs; The generated ringback tone content is intelligently recommended to users based on the multimedia big model.

2. The method for improving video ringback tone based on a large language model according to claim 1, characterized in that: Optimizing the color ring back tone system architecture according to the multimedia big model includes: Add human-computer interaction module, ring tone content generation module and ring tone user portrait module to the ring tone system architecture; The human-computer interaction module is used to obtain user needs and push video ringback tone according to user needs; The color ring back tone content generation module is used to generate a video color ring back tone according to user needs; The color ring back tone user portrait module is used to classify and mark users according to their needs.

3. The video ringback tone enhancement method based on a large language model according to claim 1, characterized in that: The intelligent generation of ringback tone content according to user needs includes: Import all user requirements; Pre-process all user needs; Generate ringback tone content based on pre-processed user needs.

4. The method for improving video ringback tone based on a large language model according to claim 3, characterized in that: All requirements of the imported users include: Use human-computer interaction templates to obtain customer needs; Add 5G letter reading, 5G messaging and ringback tone video customer service functions to provide media display for users.

5. The method for improving video ringback tone based on a large language model according to claim 3, characterized in that: The preprocessing of all user requirements includes: Convert customer requirements into text and normalize them; Filter out bad information from normalized customer demands; Categorize the filtered customer demands according to tracks, genres, singers, video themes and storylines.

6. The method for improving video ringback tone based on a large language model according to claim 3, characterized in that: Generating the color ring tone content according to the pre-processed user demand includes: Convert user input requirements into keywords; Generate ringback tone content according to the keyword.

7. The method for improving video ringback tone based on a large language model according to claim 6, characterized in that: Generating the ringback tone content according to the keyword includes: Based on the audio and video requirements, ringback tone scripts, and user portrait information in the keywords, a set of integrated audio and video ringback tones is intelligently generated, and the generated content is sent to the user side through the human-computer interaction module; Obtain different feedback from users based on the generated content.

8. The method for improving video ringback tone based on a large language model according to claim 6, characterized in that: Generating the color ring back tone content according to the keyword includes: Based on the audio and video requirements in the keywords and the key information of the user profile, audio ringback tone and video-only ringback tone are generated respectively, and sent to the terminal side in batches or all at once according to user needs; If the user issues a deterministic feedback or an optimized feedback on the generated audio or video ringback tone, the service flow corresponding to the deterministic feedback or the optimized feedback is entered until the user confirms that both the audio and video meet the requirements; After the audio and video intelligent fitting request is issued, the content generation module will call the multimedia big model again, and intelligently fit the audio and video selected by the user through the alignment algorithm and content consistency check technology, and output the video ringtone with integrated audio and video; If the user has optimization requirements for the fitted video ringback tone, the service will enter the optimization feedback process.

9. The method for improving video ringback tone based on a large language model according to claim 1, characterized in that: The intelligently recommending the generated ringback tone content to the user according to the multimedia big model comprises: Use multimedia big models to accurately analyze user portraits and needs, quickly identify and mark ringback tone content, and intelligently predict user behavior; Calling the multimedia big model to intelligently analyze the user's basic profile and service usage frequency, and classify users into high-frequency users, users to be mined, and new users; Develop login-type entry push and proactive recommendation methods based on user type.

10. The method for improving video ringback tone based on a large language model according to claim 9, characterized in that: The login-type entry push and active recommendation methods formulated according to user types include: For high-frequency users, the multimedia big model analyzes usage habits, preferences, and historical choices, combines business browsing records and retrieval information data, and intelligently generates the next business demand forecast; Set up positive feedback, loyalty rewards or privileges based on predictive recommendations to continuously improve the accuracy of content recommendations; For users to be mined, we deeply mine user basic portraits, historical data and behavior pattern information based on multimedia big models, capture the potential interests of users in idols and music styles, and push strategies through login-type portals; When users log in to the APP, selected ringtones are recommended. If users click to watch, like, collect or order, positive feedback is formed, leading to conversion into high-frequency users. For new users, we use a large multimedia model to analyze registration information, first-time usage data, and basic questionnaire data, and push recent selections and popular thematic ringtones that match the style and content preferences.