Artificial intelligence-based material generation and recommendation method, device, equipment and medium

Abstract

The application relates to the technical field of artificial intelligence, and discloses a material generation and recommendation method, device, equipment and medium based on artificial intelligence, which comprises the following steps: obtaining target product basic information uploaded by a user, and performing multi-source collaborative analysis on the target product basic information; integrating the results of the multi-source collaborative analysis into integrated information, and generating a target script scheme for the integrated information by using a text large model; generating a plurality of candidate materials for the target script scheme by combining a video generation large model; constructing a user portrait for the user to generate a user feature vector, and converting each candidate material into a corresponding video feature vector; selecting a target material from the plurality of candidate materials by combining the user feature vector and the video feature vector, and pushing the target material to the user. The application can be applied to the field of financial technology and the like, and can improve the material generation effect and enhance the material marketing effect.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to methods, apparatus, devices and media for generating and recommending materials based on artificial intelligence. Background Technology

[0002] As short videos become a mainstream communication channel for financial marketing, users are increasingly demanding higher efficiency, greater personalization, and more precise targeting of short video marketing content. Currently, the short video marketing system for financial products suffers from significant technological shortcomings in content production, personalized content customization, and intelligent recommendation, severely hindering the overall efficiency and user conversion rates of financial short video marketing.

[0003] In terms of marketing video content production, the existing production models for financial product video marketing materials are immature and inefficient, with lengthy production cycles and high costs. A complete video production process involves multiple stages, including creative planning, scriptwriting, filming, post-production editing, compliance review, and final product launch. This process typically takes several weeks, is cumbersome, and has a slow iteration speed, making it difficult to adapt to dynamic changes in financial market conditions, user needs, and industry trends. This hinders the rapid updating and accurate adaptation of marketing content, resulting in a severe lack of timely coverage of marketing scenarios. Furthermore, the industry currently widely uses template-based video generation tools to create marketing materials. These tools rigidly define production logic and content frameworks, leading to significant homogenization of the produced videos. Content formats, marketing rhetoric, and visual presentations are highly similar, failing to tailor to the risk preferences, financial needs, and cognitive levels of different user groups. This makes it difficult to achieve deep personalized content customization, resulting in weak marketing appeal and low user acceptance.

[0004] At the level of intelligent video recommendation, the existing technical architecture of financial short video recommendation systems is relatively traditional, and the recommendation logic has obvious limitations. Most current recommendation algorithms rely solely on shallow features such as video text and static image tags for matching and recommendation, failing to deeply mine the rich semantic information of short video visuals, audio, plot, marketing focus, and product attributes. This results in insufficient semantic understanding of the core marketing content, target audience, and product selling points. This technical deficiency directly leads to low recommendation matching accuracy and poor content delivery precision, easily resulting in a mismatch between marketing content and user needs. This significantly reduces user viewing experience and click-through conversion intentions, ultimately causing low user conversion rates and wasted marketing resources in financial short video marketing.

[0005] In summary, existing financial short video marketing technologies suffer from numerous problems, including low content production efficiency, insufficient personalization, and poor accuracy in intelligent recommendations. The overall marketing chain is inefficient and cannot meet the core needs of financial institutions for efficient marketing, precise customer acquisition, and high conversion rates on mainstream short video channels. The industry urgently needs a new intelligent marketing solution for financial short videos to address the various shortcomings of existing technologies and comprehensively improve the marketing efficiency and conversion effect of financial short videos. Summary of the Invention

[0006] This invention provides a method, apparatus, computer device, and storage medium for generating and recommending content based on artificial intelligence, aiming to improve the quality of content generation and enhance the effectiveness of content promotion.

[0007] In a first aspect, embodiments of the present invention provide a method for generating and recommending materials based on artificial intelligence, including: Obtain basic information about the target product uploaded by the user, and perform multi-source collaborative analysis on the basic information about the target product; The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model. Multiple candidate materials are generated for the target script scheme by combining a large video generation model; A user profile is constructed for the user to generate a user feature vector, and each of the candidate materials is converted into a corresponding video feature vector; The target material is selected from multiple candidate materials by combining the user feature vector and the video feature vector, and the target material is pushed to the user.

[0008] Secondly, embodiments of the present invention provide an artificial intelligence-based material generation and recommendation device, comprising: The information analysis unit is used to obtain basic information about the target product uploaded by the user and to perform multi-source collaborative analysis on the basic information about the target product. The script generation unit is used to integrate the results of the multi-source collaborative analysis into integrated information, and use a large text model to generate a target script scheme from the integrated information. The material generation unit is used to generate multiple candidate materials based on the target script scheme by combining the video generation big model; The vector generation unit is used to construct a user profile for the user, generate user feature vectors, and convert each of the candidate materials into a corresponding video feature vector. The material push unit is used to select target material from multiple candidate materials by combining the user feature vector and the video feature vector, and push the target material to the user.

[0009] Thirdly, embodiments of the present invention provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the above-described artificial intelligence-based material generation and recommendation method.

[0010] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described artificial intelligence-based material generation and recommendation method.

[0011] In the aforementioned scheme implemented by the AI-based material generation and recommendation method, apparatus, computer equipment, and storage medium, the basic information of the target product uploaded by the user can be obtained through the client, and multi-source collaborative analysis can be performed on the basic information of the target product; the results of the multi-source collaborative analysis can be integrated into integrated information, and a target script scheme can be generated from the integrated information using a large text model; multiple candidate materials can be generated from the target script scheme by combining a large video generation model; a user profile can be constructed for the user to generate user feature vectors, and each candidate material can be converted into a corresponding video feature vector; the target material can be selected from multiple candidate materials by combining the user feature vector and the video feature vector, and the target material can be pushed to the user and then returned to the client. In this embodiment of the invention, through the linkage of the large text model and the large video generation model, the entire process of production from script generation to material output can be completed automatically based on the basic information of the target product, constructing a full-link intelligent video promotion platform from creative generation, video production to precise distribution, thereby significantly shortening the production cycle of promotional videos, reducing production costs, and enabling rapid response to market dynamics and covering time-sensitive promotion scenarios. Meanwhile, in the recommendation process, multi-dimensional features of the materials are extracted to generate video feature vectors, which are then combined with the user's multi-dimensional feature vectors to achieve accurate matching. This fully explores the deep semantic information of the materials, significantly improving the recommendation matching degree and push accuracy, avoiding mismatch between marketing content and user needs, and realizing an automated closed loop of "demand input - content generation - accurate push - optimization iteration". Ultimately, this improves the user conversion rate of financial short video promotion, avoids the waste of promotion resources, and solves the problems of low content production efficiency, insufficient personalization, and poor accuracy of intelligent recommendation in existing technologies. It can meet the core needs of financial institutions for efficient marketing and accurate customer acquisition through short video channels. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of an application environment for an artificial intelligence-based material generation and recommendation method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for generating and recommending materials based on artificial intelligence, according to one embodiment of the present invention. Figure 3 This is a schematic diagram of a sub-process of a material generation and recommendation method based on artificial intelligence in one embodiment of the present invention; Figure 4 This is a schematic diagram of a material generation and recommendation device based on artificial intelligence in one embodiment of the present invention; Figure 5 This is a schematic diagram of a substructure of an artificial intelligence-based material generation and recommendation device in one embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 7 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] The AI-based material generation and recommendation method provided in this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. The server can obtain basic information about the target product uploaded by the user through the client, and perform multi-source collaborative analysis on the basic information. The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model. Multiple candidate materials are generated from the target script scheme using a large video generation model. A user profile is constructed for the user to generate user feature vectors, and each candidate material is converted into a corresponding video feature vector. The target material is selected from the multiple candidate materials by combining the user feature vector and the video feature vector, and the target material is pushed to the user and then returned to the client. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0016] Please see Figure 2 As shown, Figure 2 A flowchart illustrating an AI-based material generation and recommendation method provided in an embodiment of the present invention includes steps S101 to S105.

[0017] S101: Obtain the target product basic information uploaded by the user, and perform multi-source collaborative analysis on the target product basic information; S102: The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model; S103: Combine the video generation model to generate multiple candidate materials for the target script scheme; S104: Construct a user profile for the user to generate a user feature vector, and convert each of the candidate materials into a corresponding video feature vector; S105: Combine the user feature vector and the video feature vector to select the target material from multiple candidate materials, and push the target material to the user.

[0018] In this embodiment, firstly, basic information about the target product uploaded by the user is acquired and analyzed collaboratively from multiple sources. Then, the analysis results are integrated into consolidated information, and a target script scheme is generated using a large text model. Next, a large video generation model is used to generate multiple candidate materials based on the target script scheme. Afterward, a user profile is constructed to generate user feature vectors, and each candidate material is converted into a corresponding video feature vector. Finally, the target material is selected from the multiple candidate materials by combining the user feature vector and the video feature vector, and then pushed to the user.

[0019] This embodiment, through the linkage of a large text model and a large video generation model, can automatically complete the entire production process from script generation to candidate material output based on the target product's basic information. It constructs a full-link intelligent video promotion platform from creative generation and video production to precise distribution, significantly shortening the production cycle of promotional videos, reducing production costs, and enabling rapid response to market dynamics and covering time-sensitive promotional scenarios. Simultaneously, in the recommendation stage, by extracting multi-dimensional features of candidate materials to generate video feature vectors, and combining them with users' multi-dimensional feature vectors to achieve precise matching, it fully mines the deep semantic information of candidate materials, significantly improving recommendation matching and push accuracy. This avoids mismatch between promotional content and user needs, achieving an automated closed loop of "demand input - content generation - precise push - optimization iteration," ultimately improving the user conversion rate of financial short video promotion, avoiding the waste of marketing resources, and solving the problems of low content production efficiency, insufficient personalization, and poor intelligent recommendation accuracy in existing technologies. It can meet the core needs of financial institutions for efficient marketing and precise customer acquisition through short video channels.

[0020] In practical applications, the basic information of the target product uploaded by users can cover the type of financial product, expected return range, risk level, core selling points, etc. At the same time, they can also mark the target audience tags, such as age group, risk preference, investment habits, and attention scenarios such as "retirement financial management" and "steady value appreciation".

[0021] In one embodiment, step S101 includes: Based on the target product's basic information, the product analysis agent extracts the corresponding product structure information from a preset enterprise product database. The product information is obtained by cross-validating and completing the basic information of the target product using the product structured information. The system utilizes a public opinion analysis intelligence agent to collect market demand insights into the basic information of the target product; these market demand insights include relevant financial topic discussions and user sentiment trends.

[0022] In this embodiment, when performing multi-source collaborative analysis on the acquired basic information of the target product, the product analysis agent first extracts the structured information of the corresponding product from the company's existing product database. Through cross-validation, it completes the missing or ambiguous content in the basic information to avoid deviations in the subsequently generated content due to incomplete basic information. Then, the public opinion analysis agent simultaneously collects publicly available public opinion data related to the target product in the current market, and sorts out the market discussion hotspots, selling points that users generally care about, and overall user sentiment (such as the attention given to "low risk" and "flexible access"). The completed product information is integrated with market insight information to provide comprehensive and accurate input information for subsequent script generation.

[0023] It should be noted that the product analysis agent and public opinion analysis agent (including other agents mentioned later) described in this embodiment are all dedicated agents pre-tuned based on a large language model. They can be specifically optimized for tasks such as product information sorting and public opinion information collection in the financial field, enabling more accurate information analysis and processing in financial scenarios and outputting analysis results that conform to financial industry information standards. For example, the product analysis agent can accurately match the standardized information of similar products in the product library for given basic information of a target product, automatically complete the missing information in the basic information such as product risk level, minimum investment amount, and fee structure, and automatically correct erroneous descriptions in the input information, outputting standardized product information content, effectively ensuring the accuracy and standardization of the information in the subsequently generated content. As for the public opinion analysis agent, it can accurately capture relevant discussions on current social media platforms, financial forums, and other public channels for different categories of target products, extract core information such as the return characteristics, risk concerns, and service needs that users generally care about regarding this type of product, and quantitatively output the overall sentiment and demand of users, helping the subsequently generated marketing content to better align with the current user focus in the market.

[0024] In one embodiment, step S102 includes: The product information and market demand insights are combined into the integrated information; Multiple candidate script schemes are generated from the integrated information using a large text model; The candidate script solutions are screened to obtain the target script solution; wherein the screening method includes manual screening and / or automatic screening, and the screening criteria for automatic screening include solution compliance, solution creativity and target audience matching.

[0025] After completing multi-source collaborative analysis, this embodiment first integrates product information and market demand insights into a standardized integrated information input by generating an intelligent agent through scripting. Then, it calls a large text model fine-tuned based on financial compliance cases to quickly generate multiple candidate script schemes adapted to different audience styles. The content framework and marketing rhetoric of each candidate script are adjusted according to the characteristics of the target audience. For example, a script targeting middle-aged and elderly conservative investors will emphasize low risk and stable returns, using plain and easy-to-understand language; a script targeting younger investors will emphasize low barriers to entry and flexible redemption, using a more lively language style that fits the online context, thus satisfying the content acceptance habits of different groups. Specifically, candidate script schemes may include complete narration, scene descriptions (such as "3 seconds in the opening: animation showcasing a family financial management scene," "15 seconds: text overlay of the product's core selling points"), and built-in verification logic such as "pre-positioned risk warnings" and "compliant return descriptions" to avoid illegal content.

[0026] After generating multiple candidate script solutions, you can choose to manually or automatically select the final target script solution according to actual needs. Automatic selection will preset compliance verification rules, first eliminating candidate scripts that do not meet regulatory requirements, such as those containing illegal expressions or missing risk warnings, and then scoring and ranking the remaining candidate scripts based on the creativity of the solution and its matching with the target audience. The top-ranked scripts will be selected as the target script solutions for subsequent use. For example, 3-5 video script solutions that meet financial compliance requirements and are both creative and suitable for the audience can be selected.

[0027] In a specific embodiment, when generating multiple candidate script schemes from the integrated information using a large text model, the process includes the following steps: First, based on preset multi-script generation parameters, different style positioning and audience adaptation tags are matched for different candidate scripts. Then, the integrated information is input into the large text model according to different style positioning requirements to generate multiple candidate scripts with clear style distinctions and adapted to different audience preferences. This avoids homogenization of multiple candidate script content and ensures that the target script schemes obtained through subsequent screening can meet differentiated marketing needs.

[0028] In one embodiment, such as Figure 3 As shown, step S103 includes: S201~S204.

[0029] S201: Generate original dynamic visuals for the target script scheme using a material generation intelligent agent; S202: Add compliant elements to the original dynamic image using a compliant watermarking agent; S203: Generate corresponding synthesized speech from original dynamic images with added compliant elements through a speech synthesis intelligent agent; S204: Align and render the original dynamic visuals with added compliant elements and the synthesized speech to obtain the candidate materials.

[0030] In this embodiment, when generating candidate materials for the selected target script scheme, the material generation agent first calls the video generation model based on the storyboard description in the target script scheme, and generates original dynamic images that match the content by combining the target audience's style. For example, based on storyboard descriptions such as "showing the bank branch environment" and "dynamically demonstrating the profit growth curve," the system can automatically generate corresponding images. If matching resources exist in the company's existing compliance material library (such as brand image video clips and product footage), they are reused first through intelligent retrieval. Subsequently, the compliance watermark agent adds compliance watermarks that meet regulatory requirements and necessary product risk warning text to fixed positions on the dynamic images to ensure that the marketing content meets financial compliance display standards. For example, standardized risk warnings such as "Wealth management is not a deposit, products are risky, and investment requires caution" are added to fixed positions such as the lower right corner of the video, along with compliance elements such as the company logo and product registration number, to comply with financial regulatory requirements. Next, the system calls the speech synthesis agent to generate a synthesized voice matching the style based on the script's narration and style positioning: a standard broadcast voice with a slow pace and clear pronunciation is used for middle-aged and elderly audiences, while a light and lively youthful voice is used for younger audiences. Finally, the processed original animated visuals and synthesized speech are aligned according to the script's timeline annotations, and then the overall rendering output is completed, resulting in complete video candidate materials. Each target script solution corresponds to the generation of an independent marketing video material, thus forming multiple differentiated material results.

[0031] In a specific embodiment, the processing steps of the intelligent material generation agent described in this embodiment include: receiving the storyboard description content in the target script scheme, converting the textual storyboard description into generation prompts that can be recognized by the video generation model, and then adjusting the parameter requirements of the prompts according to the style positioning corresponding to the storyboard. For example, the visual style parameters of a storyboard targeting a young audience will be adjusted to a more minimalist and youthful design, while the visual style of a storyboard targeting a middle-aged and elderly audience will be adjusted to a clearer, more eye-catching, and higher-contrast visual style. Subsequently, the prompts with adjusted parameters are input into the video generation model to generate original dynamic scene clips for the corresponding storyboard. Then, the dynamic scene clips of all storyboards are spliced ​​and integrated according to the script timeline to obtain the complete original dynamic scene corresponding to the target script. If the intelligent agent finds compliant live-action or pre-made materials in the enterprise's compliant material library, it will directly retrieve the corresponding materials to replace the corresponding generated clips, ensuring compliance while improving the stability and brand consistency of the generated scene.

[0032] The specific processing steps of the aforementioned compliance watermarking intelligent agent include: after receiving the request to add compliance elements, automatically identifying the compliance content to be added according to preset financial regulatory rules, obtaining the filing information of the target product corresponding to the current candidate material and the risk warning content that must be disclosed, and then adding elements such as compliance warning text, product filing number, and brand logo to a fixed area of ​​the dynamic screen according to the display position, font size, and dwell time requirements of regulatory requirements. At the same time, it automatically detects whether there is already complete compliance content in the screen. If there is any missing or non-compliant content, it will be automatically adjusted and corrected to ensure that the final generated screen meets the compliance display requirements of financial marketing.

[0033] The specific processing steps of the aforementioned speech synthesis intelligent agent include: selecting a matching speech model based on the audience style and narration content corresponding to the target script, converting the narration text in the script into natural and fluent synthesized speech, and adjusting the speech speed and pauses according to the rhythm of the narration content to ensure that the speech rhythm matches the rhythm of the screen display and adapts to the viewing habits of users on short video platforms.

[0034] The above-described material generation steps can produce a batch of complete video materials that meet compliance requirements and are compatible with multiple channels. This process eliminates the need for manual editing, significantly reducing the production threshold and time cost of video materials, and enabling the rapid output of differentiated materials in batches to support multi-scenario marketing campaigns.

[0035] In one embodiment, step S104 includes: The system acquires user data in real time and generates multi-dimensional user feature vectors based on the user data; wherein the user data includes user static attributes, user dynamic behavior data, and real-time scenarios; The video understanding agent extracts semantic features, style features, and compliance features for each candidate material, and combines these features to generate a video feature vector for each candidate material.

[0036] In this embodiment, when pushing candidate materials to users in a targeted manner, a precise multi-dimensional user feature vector is first generated in real time for each user. For example, the user's static attributes can include fixed basic information such as the user's age, occupation, region, risk tolerance level, and historical investment preferences. The user's dynamic behavior data can include behavioral data such as the types of target products that the user has clicked and viewed in the past, the duration of stay, and forwarding and interaction records. The real-time scenario can include real-time scenario information such as the user's current advertising channel, browsing time, and the financial categories that the user is currently interested in. After integrating and encoding these three types of information, a user feature vector that can completely represent the user's needs and preferences can be formed.

[0037] Simultaneously, for the multiple candidate materials already generated, the video understanding agent is invoked to extract the core semantic features (i.e., the core selling points of the marketing content and the corresponding product category, such as "retirement financial management" or "low-risk and stable"), style features (i.e., the style of the target audience and the type of distribution channel, such as "animated science popularization" or "live-action explanation"), and compliance features (i.e., compliance level and applicable regulatory scope, such as the corresponding product type). After fusing and encoding these three types of features, a video feature vector corresponding to each candidate material is generated, providing a data foundation for subsequent matching and recommendation.

[0038] In a specific embodiment, the specific processing steps of the video understanding agent include: firstly, sampling the entire marketing video segment, extracting the visual frame image features and audio semantic features of the video respectively, then fusing the two types of features across modalities to obtain the overall semantic features of the marketing video, further extracting information such as the core selling points of the corresponding product, the target audience, and the suitable distribution channels to refine the style features of the video; simultaneously, identifying the completeness and display standardization of compliance prompts in the video to obtain compliance features, and after completing the extraction and integration of the three types of features, encoding and generating a standardized video feature vector corresponding to the candidate material.

[0039] In one embodiment, step S105 includes: The cosine similarity is calculated on the user feature vector and the video feature vector corresponding to each candidate material to obtain an initial similarity score; The corresponding real-time scene is obtained based on the user feature vector, and recommendation weights are generated based on the real-time scene. The initial similarity score is recalculated using real-time recommendation weights to obtain a new similarity score. Select the target similarity score with the highest similarity score, and set the candidate material corresponding to the target similarity score as the target material.

[0040] This embodiment, after obtaining the user feature vector and the video feature vectors corresponding to each candidate material, first calculates the cosine similarity between the two vectors to obtain an initial matching score between user preferences and material features. Then, it adjusts the recommendation weights based on real-time information from the current recommendation scenario. For example, when recommending to mobile users during weekday morning rush hour, the weight of short, fast-paced materials is appropriately increased; in weekend financial news scenarios, the weight of detailed, clearly explained materials is appropriately increased. The weights can also be adjusted for materials with specific compliance levels according to channel requirements. The adjusted recommendation weights are then used to correct the initial similarity score, resulting in the final matching similarity score. Finally, the candidate material with the highest score is selected as the target material to be pushed, completing a precise recommendation for the corresponding user. This weighted adjustment matching method ensures that the recommended materials align with users' long-term preferences and adapt to their current browsing needs, effectively improving the click-through rate and conversion rate of the recommended content.

[0041] Furthermore, this embodiment also supports A / B testing configuration, which allows for comparison of the recommendation effects of different materials. Specifically, random group push can be configured according to testing needs, continuously collecting exposure, click, and conversion data of each group of materials. Through comparative analysis, the optimal material scheme can be determined. Subsequently, the parameter settings of model generation and recommendation can be optimized based on the results, forming a data-driven closed-loop optimization mechanism, thereby continuously improving the quality of material generation and recommendation accuracy.

[0042] In one embodiment, the AI-based material generation and recommendation method further includes: Obtain feedback effect data of the candidate materials; The feedback data is analyzed from multiple dimensions, and the text model and video generation model are optimized and updated based on the results of the multi-dimensional analysis.

[0043] After the material is pushed out, this embodiment will continuously collect feedback data on the candidate materials throughout the entire delivery process, such as exposure rate, click-through rate, completion rate, conversion rate, and negative user feedback data. This allows for analysis of the performance differences of materials with different styles and selling points from dimensions such as exposure performance, user interaction, and conversion effects. It can identify which types of content are more recognized by the target user group and which content has compliance risks or low user acceptance (e.g., "30-40 year old conservative users have the highest click-through rate for animated financial videos" or "videos describing certain returns have low completion rates"). Subsequently, these analysis results will be used as training data to iteratively optimize the parameters of the text model and video generation model, continuously improving the content quality, audience suitability, and compliance of subsequently generated materials. This forms a complete closed loop from demand input, material generation, intelligent recommendation to effect optimization, continuously enhancing the material generation and recommendation effect of the entire system.

[0044] Of course, it should be noted that in addition to optimizing and updating the large text model and the large video generation model, this embodiment will also simultaneously optimize the parameters of other agent models used for feature extraction and matching calculation. This will continuously improve the accuracy of user feature extraction, material feature encoding, and matching recommendation based on feedback data accumulation, thereby further enhancing the overall system performance. For example, the script generation model (optimizing creative direction), the video understanding model (improving feature extraction accuracy), and the matching recommendation model (adjusting weight strategies) can be fine-tuned. The optimized models will be directly applied to the script generation and recommendation stages in the next cycle, forming a self-reinforcing closed loop of "data-optimization-iteration".

[0045] In summary, the AI-based content generation and recommendation method provided in this embodiment comprehensively optimizes the financial short video marketing system from three core dimensions: marketing efficiency, ad conversion, and intelligent compliance, effectively addressing existing technological shortcomings. In terms of marketing efficiency, it significantly reduces the traditional weekly video production cycle to the hourly level, supports the batch generation of massive amounts of differentiated candidate content, and can quickly keep up with market trends, user needs, and industry hotspots, implementing an agile marketing model and completely solving the problems of slow iteration and delayed response in traditional content production. In terms of ad conversion, relying on multimodal semantic understanding technology to deeply mine the full-dimensional content information of videos and combine it with refined user profiles to achieve precise matching, enabling personalized content recommendations that cater to diverse user needs. This accurately delivers financial marketing videos tailored to user needs to the target audience, effectively improving the click-through rate and completion rate of short videos, and significantly increasing the user's final purchase conversion rate. In terms of content compliance and quality optimization, the solution incorporates a compliance verification mechanism throughout the entire content generation process, strictly ensuring that all marketing content complies with financial industry regulatory standards. At the same time, it continuously iterates and improves the overall quality of content through a data-driven closed-loop optimization mechanism, effectively reducing compliance risks and content trial-and-error costs in financial marketing, and achieving intelligent, standardized, and high-quality financial short video marketing operations.

[0046] Figure 4 This is a schematic block diagram of an AI-based material generation and recommendation device 400 provided in an embodiment of the present invention. The AI-based material generation and recommendation device 400 includes: The information analysis unit 401 is used to obtain the target product basic information uploaded by the user and to perform multi-source collaborative analysis on the target product basic information; The script generation unit 402 is used to integrate the results of the multi-source collaborative analysis into integrated information, and use a large text model to generate a target script scheme from the integrated information. The material generation unit 403 is used to generate multiple candidate materials for the target script scheme by combining the video generation large model; The vector generation unit 404 is used to construct a user profile for the user, generate a user feature vector, and convert each of the candidate materials into a corresponding video feature vector. The material push unit 405 is used to select target material from multiple candidate materials by combining the user feature vector and the video feature vector, and push the target material to the user.

[0047] In this embodiment, firstly, basic information about the target product uploaded by the user is acquired and analyzed collaboratively from multiple sources. Then, the analysis results are integrated into consolidated information, and a target script scheme is generated using a large text model. Next, a large video generation model is used to generate multiple candidate materials based on the target script scheme. Afterward, a user profile is constructed to generate user feature vectors, and each candidate material is converted into a corresponding video feature vector. Finally, the target material is selected from the multiple candidate materials by combining the user feature vector and the video feature vector, and then pushed to the user.

[0048] This embodiment, through the linkage of a large text model and a large video generation model, can automatically complete the entire production process from script generation to material output based on the target product's basic information. It constructs a full-link intelligent video promotion platform from creative generation and video production to precise distribution, significantly shortening the production cycle of marketing videos, reducing production costs, and enabling rapid response to market dynamics and covering time-sensitive promotion scenarios. Simultaneously, in the recommendation stage, by extracting multi-dimensional features of the materials to generate video feature vectors, and combining them with the user's multi-dimensional feature vectors to achieve precise matching, it fully mines the deep semantic information of the materials, significantly improving recommendation matching and push accuracy. This avoids mismatch between promotional content and user needs, achieving an automated closed loop of "demand input - content generation - precise push - optimization iteration," ultimately improving the user conversion rate of financial short video promotion, avoiding waste of promotional resources, and solving the problems of low content production efficiency, insufficient personalization, and poor intelligent recommendation accuracy in existing technologies. It can meet the core needs of financial institutions for efficient marketing and precise customer acquisition through short video channels.

[0049] In practical applications, the basic information of the target product uploaded by users can cover the type of financial product, expected return range, risk level, core selling points, etc. At the same time, they can also mark the target audience tags, such as age group, risk preference, investment habits, and attention scenarios such as "retirement financial management" and "steady value appreciation".

[0050] In one embodiment, the information analysis unit 401 includes: Based on the target product's basic information, the product analysis agent extracts the corresponding product structure information from a preset enterprise product database. The product information is obtained by cross-validating and completing the basic information of the target product using the product structured information. The system utilizes a public opinion analysis intelligence agent to collect market demand insights into the basic information of the target product; these market demand insights include relevant financial topic discussions and user sentiment trends.

[0051] In this embodiment, when performing multi-source collaborative analysis on the acquired basic information of the target product, the product analysis agent first extracts the structured information of the corresponding product from the company's existing product database. Through cross-validation, it completes the missing or ambiguous content in the basic information to avoid deviations in the subsequently generated content due to incomplete basic information. Then, the public opinion analysis agent simultaneously collects publicly available public opinion data related to the target product in the current market, and sorts out the market discussion hotspots, selling points that users generally care about, and overall user sentiment (such as the attention given to "low risk" and "flexible access"). The completed product information is integrated with market insight information to provide comprehensive and accurate input information for subsequent script generation.

[0052] It should be noted that the product analysis agent and public opinion analysis agent (including other agents mentioned later) described in this embodiment are all dedicated agents pre-tuned based on a large language model. They can be specifically optimized for tasks such as product information sorting and public opinion information collection in the financial field, enabling more accurate information analysis and processing in financial scenarios and outputting analysis results that conform to financial industry information standards. For example, the product analysis agent can accurately match the standardized information of similar products in the product library for given basic information of a target product, automatically complete the missing information in the basic information such as product risk level, minimum investment amount, and fee structure, and automatically correct erroneous descriptions in the input information, outputting standardized product information content, effectively ensuring the accuracy and standardization of the information in the subsequently generated content. As for the public opinion analysis agent, it can accurately capture relevant discussions on current social media platforms, financial forums, and other public channels for different categories of target products, extract core information such as the return characteristics, risk concerns, and service needs that users generally care about regarding this type of product, and quantitatively output the overall sentiment and demand of users, helping the subsequently generated marketing content to better align with the current user focus in the market.

[0053] In one embodiment, the script generation unit 402 includes: The product information and market demand insights are combined into the integrated information; Multiple candidate script schemes are generated from the integrated information using a large text model; The candidate script solutions are screened to obtain the target script solution; wherein the screening method includes manual screening and / or automatic screening, and the screening criteria for automatic screening include solution compliance, solution creativity and target audience matching.

[0054] After completing multi-source collaborative analysis, this embodiment first integrates product information and market demand insights into a standardized integrated information input by generating an intelligent agent through scripting. Then, it calls a large text model fine-tuned based on financial compliance cases to quickly generate multiple candidate script schemes adapted to different audience styles. The content framework and marketing rhetoric of each candidate script are adjusted according to the characteristics of the target audience. For example, a script targeting middle-aged and elderly conservative investors will emphasize low risk and stable returns, using plain and easy-to-understand language; a script targeting younger investors will emphasize low barriers to entry and flexible redemption, using a more lively language style that fits the online context, thus satisfying the content acceptance habits of different groups. Specifically, candidate script schemes may include complete narration, scene descriptions (such as "3 seconds in the opening: animation showcasing a family financial management scene," "15 seconds: text overlay of the product's core selling points"), and built-in verification logic such as "pre-positioned risk warnings" and "compliant return descriptions" to avoid illegal content.

[0055] After generating multiple candidate script solutions, you can choose to manually or automatically select the final target script solution according to actual needs. Automatic selection will preset compliance verification rules, first eliminating candidate scripts that do not meet regulatory requirements, such as those containing illegal expressions or missing risk warnings, and then scoring and ranking the remaining candidate scripts based on the creativity of the solution and its matching with the target audience. The top-ranked scripts will be selected as the target script solutions for subsequent use. For example, 3-5 video script solutions that meet financial compliance requirements and are both creative and suitable for the audience can be selected.

[0056] In a specific embodiment, when generating multiple candidate script schemes from the integrated information using a large text model, the process includes the following steps: First, based on preset multi-script generation parameters, different style positioning and audience adaptation tags are matched for different candidate scripts. Then, the integrated information is input into the large text model according to different style positioning requirements to generate multiple candidate scripts with clear style distinctions and adapted to different audience preferences. This avoids homogenization of multiple candidate script content and ensures that the target script schemes obtained through subsequent screening can meet differentiated marketing needs.

[0057] In one embodiment, such as Figure 5 As shown, the material generation unit 403 includes: The image generation unit 501 is used to generate original dynamic images based on the target script scheme through the material generation intelligent agent. The element adding unit 502 is used to add compliant elements to the original dynamic image through the compliant watermarking agent; The speech generation unit 503 is used to generate corresponding synthesized speech from original dynamic images with added compliant elements through a speech synthesis agent; The alignment rendering unit 504 is used to align and render the original dynamic images with added compliant elements and the synthesized speech to obtain the candidate materials.

[0058] In this embodiment, when generating candidate materials for the selected target script scheme, the material generation agent first calls the video generation model based on the storyboard description in the target script scheme, and generates original dynamic images that match the content by combining the target audience's style. For example, based on storyboard descriptions such as "showing the bank branch environment" and "dynamically demonstrating the profit growth curve," the system can automatically generate corresponding images. If matching resources exist in the company's existing compliance material library (such as brand image video clips and product footage), they are reused first through intelligent retrieval. Subsequently, the compliance watermark agent adds compliance watermarks that meet regulatory requirements and necessary product risk warning text to fixed positions on the dynamic images to ensure that the marketing content meets financial compliance display standards. For example, standardized risk warnings such as "Wealth management is not a deposit, products are risky, and investment requires caution" are added to fixed positions such as the lower right corner of the video, along with compliance elements such as the company logo and product registration number, to comply with financial regulatory requirements. Next, the system calls the speech synthesis agent to generate a synthesized voice matching the style based on the script's narration and style positioning: a standard broadcast voice with a slow pace and clear pronunciation is used for middle-aged and elderly audiences, while a light and lively youthful voice is used for younger audiences. Finally, the processed original animated visuals and synthesized speech are aligned according to the script's timeline annotations, and then the overall rendering output is completed, resulting in complete video candidate materials. Each target script solution corresponds to the generation of an independent marketing video material, thus forming multiple differentiated material results.

[0059] In a specific embodiment, the processing steps of the intelligent material generation agent described in this embodiment include: receiving the storyboard description content in the target script scheme, converting the textual storyboard description into generation prompts that can be recognized by the video generation model, and then adjusting the parameter requirements of the prompts according to the style positioning corresponding to the storyboard. For example, the visual style parameters of a storyboard targeting a young audience will be adjusted to a more minimalist and youthful design, while the visual style of a storyboard targeting a middle-aged and elderly audience will be adjusted to a clearer, more eye-catching, and higher-contrast visual style. Subsequently, the prompts with adjusted parameters are input into the video generation model to generate original dynamic scene clips for the corresponding storyboard. Then, the dynamic scene clips of all storyboards are spliced ​​and integrated according to the script timeline to obtain the complete original dynamic scene corresponding to the target script. If the intelligent agent finds compliant live-action or pre-made materials in the enterprise's compliant material library, it will directly retrieve the corresponding materials to replace the corresponding generated clips, ensuring compliance while improving the stability and brand consistency of the generated scene.

[0060] The specific processing steps of the aforementioned compliance watermarking intelligent agent include: after receiving the request to add compliance elements, automatically identifying the compliance content to be added according to preset financial regulatory rules, obtaining the filing information of the target product corresponding to the current candidate material and the risk warning content that must be disclosed, and then adding elements such as compliance warning text, product filing number, and brand logo to a fixed area of ​​the dynamic screen according to the display position, font size, and dwell time requirements of regulatory requirements. At the same time, it automatically detects whether there is already complete compliance content in the screen. If there is any missing or non-compliant content, it will be automatically adjusted and corrected to ensure that the final generated screen meets the compliance display requirements of financial marketing.

[0061] The specific processing steps of the aforementioned speech synthesis intelligent agent include: selecting a matching speech model based on the audience style and narration content corresponding to the target script, converting the narration text in the script into natural and fluent synthesized speech, and adjusting the speech speed and pauses according to the rhythm of the narration content to ensure that the speech rhythm matches the rhythm of the screen display and adapts to the viewing habits of users on short video platforms.

[0062] The above-described material generation steps can produce a batch of complete video materials that meet compliance requirements and are compatible with multiple channels. This process eliminates the need for manual editing, significantly reducing the production threshold and time cost of video materials, and enabling the rapid output of differentiated materials in batches to support multi-scenario marketing campaigns.

[0063] In one embodiment, the vector generation unit 404 includes: The system acquires user data in real time and generates multi-dimensional user feature vectors based on the user data; wherein the user data includes user static attributes, user dynamic behavior data, and real-time scenarios; The video understanding agent extracts semantic features, style features, and compliance features for each candidate material, and combines these features to generate a video feature vector for each candidate material.

[0064] In this embodiment, when pushing candidate materials to users in a targeted manner, a precise multi-dimensional user feature vector is first generated in real time for each user. For example, the user's static attributes can include fixed basic information such as the user's age, occupation, region, risk tolerance level, and historical investment preferences. The user's dynamic behavior data can include behavioral data such as the types of target products that the user has clicked and viewed in the past, the duration of stay, and forwarding and interaction records. The real-time scenario can include real-time scenario information such as the user's current advertising channel, browsing time, and the financial categories that the user is currently interested in. After integrating and encoding these three types of information, a user feature vector that can completely represent the user's needs and preferences can be formed.

[0065] Simultaneously, for the multiple candidate materials already generated, the video understanding agent is invoked to extract the core semantic features (i.e., the core selling points of the marketing content and the corresponding product category, such as "retirement financial management" or "low-risk and stable"), style features (i.e., the style of the target audience and the type of distribution channel, such as "animated science popularization" or "live-action explanation"), and compliance features (i.e., compliance level and applicable regulatory scope, such as the corresponding product type). After fusing and encoding these three types of features, a video feature vector corresponding to each candidate material is generated, providing a data foundation for subsequent matching and recommendation.

[0066] In a specific embodiment, the specific processing steps of the video understanding agent include: firstly, sampling the entire marketing video segment, extracting the visual frame image features and audio semantic features of the video respectively, then fusing the two types of features across modalities to obtain the overall semantic features of the marketing video, further extracting information such as the core selling points of the corresponding product, the target audience, and the suitable distribution channels to refine the style features of the video; simultaneously, identifying the completeness and display standardization of compliance prompts in the video to obtain compliance features, and after completing the extraction and integration of the three types of features, encoding and generating a standardized video feature vector corresponding to the candidate material.

[0067] In one embodiment, the material delivery unit 405 includes: The cosine similarity is calculated on the user feature vector and the video feature vector corresponding to each candidate material to obtain an initial similarity score; The corresponding real-time scene is obtained based on the user feature vector, and recommendation weights are generated based on the real-time scene. The initial similarity score is recalculated using real-time recommendation weights to obtain a new similarity score. Select the target similarity score with the highest similarity score, and set the candidate material corresponding to the target similarity score as the target material.

[0068] This embodiment, after obtaining the user feature vector and the video feature vectors corresponding to each candidate material, first calculates the cosine similarity between the two vectors to obtain an initial matching score between user preferences and material features. Then, it adjusts the recommendation weights based on real-time information from the current recommendation scenario. For example, when recommending to mobile users during weekday morning rush hour, the weight of short, fast-paced materials is appropriately increased; in weekend financial news scenarios, the weight of detailed, clearly explained materials is appropriately increased. The weights can also be adjusted for materials with specific compliance levels according to channel requirements. The adjusted recommendation weights are then used to correct the initial similarity score, resulting in the final matching similarity score. Finally, the candidate material with the highest score is selected as the target material to be pushed, completing a precise recommendation for the corresponding user. This weighted adjustment matching method ensures that the recommended materials align with users' long-term preferences and adapt to their current browsing needs, effectively improving the click-through rate and conversion rate of the recommended content.

[0069] Furthermore, this embodiment also supports A / B testing configuration, which allows for comparison of the recommendation effects of different materials. Specifically, random group push can be configured according to testing needs, continuously collecting exposure, click, and conversion data of each group of materials. Through comparative analysis, the optimal material scheme can be determined. Subsequently, the parameter settings of model generation and recommendation can be optimized based on the results, forming a data-driven closed-loop optimization mechanism, thereby continuously improving the quality of material generation and recommendation accuracy.

[0070] In one embodiment, the AI-based material generation and recommendation device 400 further includes: Obtain feedback data on the candidate materials; wherein, the feedback data includes exposure rate, click-through rate, completion rate, redirection conversion rate, and user negative feedback data; The feedback data is analyzed from multiple dimensions, and the text model and video generation model are optimized and updated based on the results of the multi-dimensional analysis.

[0071] After the material is pushed out, this embodiment will continuously collect feedback data on the candidate materials throughout the entire delivery process, such as exposure rate, click-through rate, completion rate, conversion rate, and negative user feedback data. This allows for analysis of the performance differences of materials with different styles and selling points from dimensions such as exposure performance, user interaction, and conversion effects. It can identify which types of content are more recognized by the target user group and which content has compliance risks or low user acceptance (e.g., "30-40 year old conservative users have the highest click-through rate for animated financial videos" or "videos describing certain returns have low completion rates"). Subsequently, these analysis results will be used as training data to iteratively optimize the parameters of the text model and video generation model, continuously improving the content quality, audience suitability, and compliance of subsequently generated materials. This forms a complete closed loop from demand input, material generation, intelligent recommendation to effect optimization, continuously enhancing the material generation and recommendation effect of the entire system.

[0072] Of course, it should be noted that in addition to optimizing and updating the large text model and the large video generation model, this embodiment will also simultaneously optimize the parameters of other agent models used for feature extraction and matching calculation. This will continuously improve the accuracy of user feature extraction, material feature encoding, and matching recommendation based on feedback data accumulation, thereby further enhancing the overall system performance. For example, the script generation model (optimizing creative direction), the video understanding model (improving feature extraction accuracy), and the matching recommendation model (adjusting weight strategies) can be fine-tuned. The optimized models will be directly applied to the script generation and recommendation stages in the next cycle, forming a self-reinforcing closed loop of "data-optimization-iteration".

[0073] In summary, the AI-based content generation and recommendation device provided in this embodiment comprehensively optimizes the financial short video marketing system from three core dimensions: marketing efficiency, ad conversion, and intelligent compliance, effectively addressing existing technological shortcomings. In terms of marketing efficiency, it significantly reduces the traditional weekly video production cycle to the hourly level, supports the batch generation of massive amounts of differentiated candidate content, and can quickly keep up with market trends, user needs, and industry hotspots, implementing an agile marketing model and completely solving the problems of slow iteration and delayed response in traditional content production. In terms of ad conversion, relying on multimodal semantic understanding technology to deeply mine the full-dimensional content information of videos and combine it with refined user profiles to achieve precise matching, enabling personalized content recommendations that cater to diverse user needs. This accurately delivers financial marketing videos tailored to user requirements to the target audience, effectively improving the click-through rate and completion rate of short videos, and significantly increasing the user's final purchase conversion rate. In terms of content compliance and quality optimization, the solution incorporates a compliance verification mechanism throughout the entire content generation process, strictly ensuring that all marketing content complies with financial industry regulatory standards. At the same time, it continuously iterates and improves the overall quality of content through a data-driven closed-loop optimization mechanism, effectively reducing compliance risks and content trial-and-error costs in financial marketing, and achieving intelligent, standardized, and high-quality financial short video marketing operations.

[0074] Specific limitations regarding the AI-based content generation and recommendation device can be found in the limitations of the AI-based content generation and recommendation method described above, and will not be repeated here. Each module in the aforementioned AI-based content generation and recommendation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0075] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements server-side functions or steps of an artificial intelligence-based material generation and recommendation method.

[0076] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 7 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of an artificial intelligence-based material generation and recommendation method.

[0077] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain basic information about the target product uploaded by the user, and perform multi-source collaborative analysis on the basic information about the target product; The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model. Multiple candidate materials are generated for the target script scheme by combining a large video generation model; A user profile is constructed for the user to generate a user feature vector, and each of the candidate materials is converted into a corresponding video feature vector; The target material is selected from multiple candidate materials by combining the user feature vector and the video feature vector, and the target material is pushed to the user.

[0078] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: Obtain basic information about the target product uploaded by the user, and perform multi-source collaborative analysis on the basic information about the target product; The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model. Multiple candidate materials are generated for the target script scheme by combining a large video generation model; A user profile is constructed for the user to generate a user feature vector, and each of the candidate materials is converted into a corresponding video feature vector; The target material is selected from multiple candidate materials by combining the user feature vector and the video feature vector, and the target material is pushed to the user.

[0079] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0080] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0081] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0082] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0083] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for generating and recommending materials based on artificial intelligence, characterized in that, include: Obtain basic information about the target product uploaded by the user, and perform multi-source collaborative analysis on the basic information about the target product; The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model. Multiple candidate materials are generated for the target script scheme by combining a large video generation model; A user profile is constructed for the user to generate a user feature vector, and each of the candidate materials is converted into a corresponding video feature vector; The target material is selected from multiple candidate materials by combining the user feature vector and the video feature vector, and the target material is pushed to the user.

2. The method for generating and recommending materials based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the target product's basic information uploaded by the user and performing multi-source collaborative analysis on the target product's basic information includes: Based on the target product's basic information, the product analysis agent extracts the corresponding product structure information from a preset enterprise product database. The product information is obtained by cross-validating and completing the basic information of the target product using the product structured information. The system utilizes a public opinion analysis intelligence agent to collect market demand insights into the basic information of the target product; these market demand insights include relevant financial topic discussions and user sentiment trends.

3. The method for generating and recommending materials based on artificial intelligence according to claim 2, characterized in that, The results of the multi-source collaborative analysis are integrated into integrated information, and a target script scheme is generated from the integrated information using a large text model, including: The product information and market demand insights are combined into the integrated information; Multiple candidate script schemes are generated from the integrated information using a large text model; The target script scheme is obtained by filtering multiple candidate script schemes.

4. The method for generating and recommending materials based on artificial intelligence according to claim 1, characterized in that, The method combines a large video generation model to generate multiple candidate materials for the target script scheme, including: The intelligent agent generates original dynamic visuals for the target script scheme using the material generation method; The original animated images are added with compliant elements using a compliant watermarking agent. The speech synthesis agent generates corresponding synthesized speech from original dynamic images with added compliant elements. The candidate materials are obtained by aligning and rendering the original dynamic images with compliant elements and the synthesized speech.

5. The method for generating and recommending materials based on artificial intelligence according to claim 1, characterized in that, The step of constructing a user profile for the user to generate a user feature vector, and converting each of the candidate materials into a corresponding video feature vector, includes: The system acquires user data in real time and generates multi-dimensional user feature vectors based on the user data; wherein the user data includes user static attributes, user dynamic behavior data, and real-time scenarios; The video understanding agent extracts semantic features, style features, and compliance features for each candidate material, and combines these features to generate a video feature vector for each candidate material.

6. The method for generating and recommending materials based on artificial intelligence according to claim 5, characterized in that, The step of selecting target material from multiple candidate materials by combining the user feature vector and the video feature vector, and pushing the target material to the user, includes: The cosine similarity is calculated on the user feature vector and the video feature vector corresponding to each candidate material to obtain an initial similarity score; The corresponding real-time scene is obtained based on the user feature vector, and recommendation weights are generated based on the real-time scene. The initial similarity score is recalculated using real-time recommendation weights to obtain a new similarity score. Select the target similarity score with the highest similarity score, and set the candidate material corresponding to the target similarity score as the target material.

7. The method for generating and recommending materials based on artificial intelligence according to claim 1, characterized in that, Also includes: Obtain feedback effect data of the candidate materials; The feedback data is analyzed from multiple dimensions, and the text model and video generation model are optimized and updated based on the results of the multi-dimensional analysis.

8. A material generation and recommendation device based on artificial intelligence, characterized in that, include: The information analysis unit is used to obtain basic information about the target product uploaded by the user and to perform multi-source collaborative analysis on the basic information about the target product. The script generation unit is used to integrate the results of the multi-source collaborative analysis into integrated information, and use a large text model to generate a target script scheme from the integrated information. The material generation unit is used to generate multiple candidate materials based on the target script scheme by combining the video generation big model; The vector generation unit is used to construct a user profile for the user, generate user feature vectors, and convert each of the candidate materials into a corresponding video feature vector. The material push unit is used to select target material from multiple candidate materials by combining the user feature vector and the video feature vector, and push the target material to the user.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the artificial intelligence-based material generation and recommendation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the artificial intelligence-based material generation and recommendation method as described in any one of claims 1 to 7.

Images