A physical merchant short video generation method and device

Abstract

The application belongs to the technical field of video generation, and particularly relates to a short video generation method and device for an entity merchant. The method comprises the following steps: S100, acquiring merchant information and generating a keyword according to the merchant information; S200, acquiring short video data according to the keyword; S300, identifying a marketing intention of the short video data, extracting multi-dimensional features from the short video data with different marketing intentions respectively, and generating a short video script according to the multi-dimensional features; S400, driving a digital human model to broadcast according to the short video script, and generating a short video; S500, determining publishing information and publishing the short video; and S600, acquiring back-drawing data after the short video is published, and optimizing the short video according to the back-drawing data. The application can quickly generate multiple high-quality short videos with different marketing intentions, realizes closed-loop optimization of the short video, improves the efficiency of short video generation for the entity merchant, and reduces the cost.

Description

Technical Field

[0001] This application belongs to the field of video generation technology, and specifically relates to a method and apparatus for generating short videos for physical businesses. Background Technology

[0002] Currently, short videos have become an important customer acquisition channel for many brick-and-mortar businesses. However, most businesses lack the professional capabilities for video content planning, shooting, editing, and operation. Although digital human technology and AI copywriting generation technology exist to assist in solving shooting problems, these technologies are often independent and have not yet formed an end-to-end automated process. If businesses want to independently complete short video marketing, they still need to invest a lot of time and energy in learning and practice, or rely on costly third-party operation services. Therefore, many small and medium-sized businesses find it difficult to efficiently utilize short video platforms for marketing, missing out on important customer acquisition opportunities. In summary, existing technological solutions for supporting brick-and-mortar businesses in acquiring customers through short videos still suffer from pain points such as insufficient content creation capabilities, complex production processes, and a lack of data feedback and optimization mechanisms. They have not yet formed an automated, intelligently optimizable closed-loop solution, making it difficult to meet the urgent needs of brick-and-mortar businesses for efficient and low-cost short video customer acquisition.

[0003] Therefore, there is an urgent need for a technical solution to overcome or mitigate at least one of the aforementioned defects in the existing technology. Summary of the Invention

[0004] The purpose of this application is to provide a method and apparatus for generating short videos for physical businesses, in order to solve at least one of the problems in the existing technology, namely, insufficient short video content creation capabilities, complex production processes, and lack of data feedback and optimization mechanisms.

[0005] The technical solution of this application is:

[0006] The first aspect of this application provides a method for generating short videos for physical businesses, including:

[0007] S100: Obtain merchant information and generate keywords based on the merchant information;

[0008] S200. Obtain short video data based on the keywords;

[0009] S300. Identify the marketing intent of the short video data, extract multi-dimensional features from the short video data with different marketing intents, and generate short video copy based on the multi-dimensional features.

[0010] S400: The digital human model is driven to broadcast according to the short video script, generating a short video;

[0011] S500: Determine the publishing information and publish the short video;

[0012] S600. Obtain the recollection data after the short video is published, and optimize the short video based on the recollection data.

[0013] In at least one embodiment of this application, in step S100, obtaining merchant information and generating keywords based on the merchant information includes:

[0014] S101. Obtain merchant information, which includes store name, main products / services, store features and advantages, target customer groups, and marketing materials;

[0015] S102. Construct an industry category identification network to identify the industry category to which the merchant belongs based on the merchant information. The industry category identification network is as follows:

[0016] ;

[0017] Where y is the industry category probability vector, Softmax is the normalization function, W is the weight of the classification layer, b is the bias of the classification layer, T is the text information, I is the image / video information, α and β are the corresponding weights, MLP_fusion is the multilayer perceptron used for feature fusion, and GraphBuilder(S) is the knowledge graph of the merchant information S.

[0018] S103. Generate keywords based on the industry category.

[0019] In at least one embodiment of this application, in step S200, obtaining short video data based on the keywords includes:

[0020] S201. Collect short video data that meets the set conditions from the short video platform based on the keywords;

[0021] S202. Generate a structured short video database based on the short video data.

[0022] In at least one embodiment of this application, in step S300, identifying the marketing intent of the short video data, extracting multi-dimensional features from the short video data with different marketing intents, and generating short video copy based on the multi-dimensional features, includes:

[0023] S301. Identify the marketing intent of the short video data, including the marketing intent of driving traffic to stores, guiding group purchases, launching new products, and promoting brands.

[0024] S302. Extract multi-dimensional features from the short video data with different marketing intentions, including intent features, cross-type features, and industry style features;

[0025] S303. For the multi-dimensional features of the short video data for each marketing intent, extract N feature information from each feature dimension to generate a multi-dimensional feature matrix I. M×N M is the feature dimension;

[0026] S304. Generate short video script based on the multi-dimensional feature matrix.

[0027] In at least one embodiment of this application, in step S400, the short video is generated by driving the digital human model to broadcast based on the short video script, including:

[0028] S401. Obtain the digital human model;

[0029] S402. Convert the short video script into a digital human voice;

[0030] S403. Based on the digital human's voice, drive the digital human model's lip movements, facial expressions, and body movements to broadcast, generating a first short video;

[0031] S404. Add audiovisual auxiliary information to the first short video to generate a second short video. The audiovisual auxiliary information includes background information, background music information, subtitle information, transition effect information, and special effects information.

[0032] In at least one embodiment of this application, in step S500, determining the publishing information and publishing the short video includes:

[0033] S501. For each marketing intent, obtain the set of publishing account types, the set of short video tags, and the set of publishing times for the short video data;

[0034] S502. Extract the publishing account types from the set of publishing account types;

[0035] S502. Extract short video tags from the short video tag set;

[0036] S503. Count the number of release times falling into different time periods in the release time set, and extract the release time from the time period with the most release times;

[0037] S504. Publish the short video according to the published account type, the short video tag, and the published time.

[0038] In at least one embodiment of this application, in step S600, acquiring the recaptured data after the short video is published, and optimizing the short video based on the recaptured data, includes:

[0039] S601. Obtain the data collected after the short video is published, including exposure data, interaction data, conversion data, and user behavior data.

[0040] S602. The short video is comprehensively evaluated from multiple evaluation dimensions based on the collected data;

[0041] S603. Optimize the short video text, the digital human model, the publishing account type, the short video tags, and / or the publishing time based on the comprehensive evaluation results.

[0042] A second aspect of this application provides a short video generation device for physical stores, comprising:

[0043] The merchant information acquisition module is used to acquire merchant information and generate keywords based on the merchant information;

[0044] A short video data acquisition module is used to acquire short video data based on the keywords;

[0045] The short video copy generation module is used to identify the marketing intent of the short video data, extract multi-dimensional features from the short video data with different marketing intents, and generate short video copy based on the multi-dimensional features.

[0046] The short video generation module is used to drive the digital human model to broadcast based on the short video script and generate a short video.

[0047] The short video publishing module is used to determine the publishing information and publish the short video;

[0048] The short video optimization module is used to obtain the data collected after the short video is published, and to optimize the short video based on the data collected.

[0049] A third aspect of this application provides an electronic device including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the physical merchant short video generation method described above.

[0050] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, enables the physical merchant short video generation method described above.

[0051] The invention has at least the following beneficial technical effects:

[0052] The method for generating short videos for physical stores in this application can quickly generate multiple high-quality short videos with different marketing intentions. It also achieves closed-loop optimization of short videos, improves the efficiency of short video generation for physical stores, reduces costs, and enables physical stores to acquire customers efficiently and accurately through short videos without having professional video production capabilities. Attached Figure Description

[0053] Figure 1 This is a flowchart of a method for generating short videos for physical merchants according to one embodiment of this application;

[0054] Figure 2 This is a schematic diagram of a short video generation device for physical merchants according to one embodiment of this application;

[0055] Figure 3 This is the merchant information acquisition interface of the physical merchant short video generation device according to one embodiment of this application;

[0056] Figure 4 This is a short video generation interface of a physical merchant short video generation device according to one embodiment of this application;

[0057] Figure 5 This is a schematic diagram of the structure of a computer device suitable for implementing the embodiments of this application, specifically a terminal or server.

[0058] in:

[0059] 100 - Merchant Information Acquisition Module; 200 - Short Video Data Acquisition Module; 300 - Short Video Script Generation Module; 400 - Short Video Generation Module; 500 - Short Video Publishing Module; 600 - Short Video Optimization Module; 700 - Computer Equipment; 701 - CPU; 702 - ROM; 703 - RAM; 704 - Bus; 705 - I / O Interface; 706 - Input Section; 707 - Output Section; 708 - Storage Section; 709 - Communication Section; 710 - Driver; 711 - Removable Media. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be described in more detail below with reference to the accompanying drawings. In the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The described embodiments are some, but not all, embodiments of this application. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. The embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0061] The following is in conjunction with the appendix Figures 1 to 5 This application will be described in further detail.

[0062] The first aspect of this application provides a method for generating short videos for physical stores, such as... Figure 1 As shown, it includes the following steps:

[0063] S100. Obtain merchant information and generate keywords based on the merchant information;

[0064] S200. Obtain short video data based on keywords;

[0065] S300: Identify the marketing intent of short video data, extract multi-dimensional features from short video data with different marketing intents, and generate short video copy based on the multi-dimensional features;

[0066] S400: Drives a digital human model to broadcast based on short video scripts, generating short videos;

[0067] S500: Determine the information to be published and publish the short video;

[0068] S600: Obtain the data collected after the short video is published, and optimize the short video based on the collected data.

[0069] The method for generating short videos for physical merchants in this application first involves obtaining merchant information in step S100, and then generating keywords based on that information, including:

[0070] S101. Obtain merchant information, which includes store name, main products / services, store features and advantages, target customer groups, and marketing materials.

[0071] S102. Construct an industry category identification network to identify the industry category to which a merchant belongs based on the merchant information;

[0072] S103. Generate keywords based on industry category.

[0073] Merchants can enter their basic store information through text input (such as filling out forms or providing free descriptions) and voice input (converting speech to text for processing). Merchant information includes, but is not limited to, the following core information fields:

[0074] Shop names: For example, Lao Wang Barbecue Shop, Huayang Nail Salon, etc.;

[0075] Main products / services: such as specialty crayfish, Japanese-style manicures, personal fitness training courses, etc.

[0076] Store features and advantages: such as a 20-year-old brand, a popular spot for online influencers, imported ingredients, professional technicians, and limited-time promotional activities;

[0077] Target customer groups: such as young white-collar workers, families, etc.

[0078] Marketing materials: such as store logo images, promotional images, product videos, etc.

[0079] Deep learning-based text classification models, such as Deepseek and the Doubao series models, can be used to perform semantic understanding on the text information input by merchants. Through pre-training and fine-tuning on a large industry corpus, this model can automatically identify and categorize the specific industry to which the merchant belongs. For example, it can identify Mapo Tofu as Sichuan cuisine, and nail salons as beauty and hair salons, and further subdivide into more specific subcategories. Keyword extraction algorithms based on the Transformer architecture, such as KeyBERT or self-trained sequence labeling models, can also be used to extract core keywords related to the store, products, services, and advantages from the merchant's description. These keywords include not only nouns but may also contain adjectives and verb phrases, ensuring the comprehensiveness and accuracy of the keywords. For example, from "Our store's Mapo Tofu is both spicy and numbing, a 20-year-old signature dish," we can extract "Mapo Tofu," "spicy," "20-year-old brand," and "signature dish." Simultaneously, the model can also extract potential long-tail keywords to expand the scope of subsequent data collection.

[0080] In a preferred embodiment of this application, an industry category recognition network based on multimodal semantic perception and graph reasoning is constructed to identify the industry category to which the merchant belongs. The construction process of the industry category recognition network is as follows:

[0081] The textual information about businesses is represented as vectors using a text encoder (such as a BERT-like model).

[0082] ;

[0083] Among them, v t Let T be a vector of text information;

[0084] Merchant information in image / video format is represented as a vector using a visual encoder (such as ViT or CLIP's Image Encoder):

[0085] ;

[0086] Among them, v I Let I be the image / video information vector;

[0087] Cross-modal alignment is achieved through weighted fusion:

[0088] ;

[0089] Where v is the cross-modal information vector, and α and β are the corresponding weights;

[0090] Construct a knowledge graph GraphBuilder(S) of merchant information S using the GraphBuilder tool;

[0091] Key entities and concepts are extracted from merchant information and used as nodes. Graph Convolutional Networks (GCNs) are then used to perform graph neural network inference on the knowledge graph.

[0092] ;

[0093] Where H is the node feature matrix obtained by graph neural network inference;

[0094] The node feature matrix H is aggregated into a graph-level vector using the READOUT function:

[0095] ;

[0096] Perform multimodal decision fusion:

[0097] ;

[0098] Where MLP_fusion is a multilayer perceptron used for feature fusion, and [;] represents vector concatenation operation;

[0099] Industry category identification:

[0100] ;

[0101] Where y is the industry category probability vector (representing the probability that a merchant belongs to each preset industry category), Softmax is the normalization function, W is the weight of the classification layer, and b is the bias of the classification layer.

[0102] Finally, an industry category recognition network based on multimodal semantic perception and graph reasoning was obtained:

[0103] ;

[0104] In this embodiment, industry category identification is achieved through an industry category recognition network based on multimodal semantic perception and graph reasoning. This network possesses deep semantic understanding and reasoning capabilities, accurately identifying businesses as high-end organic food retailers or boutique coffee shops, rather than general food sellers. This approach is more accurate than simply relying on word frequency to determine industry category. Text and image information complement each other, providing a more comprehensive business profile and enabling cross-verification of multimodal information, thereby enhancing the representativeness and credibility of single-modal information.

[0105] In a preferred embodiment of this application, short video data is collected based on keywords, and content patterns with high customer acquisition potential are learned and extracted from massive amounts of short video data. S200 specifically includes:

[0106] S201. Collect short video data that meets the set conditions from short video platforms based on keywords;

[0107] S202. Generate a structured short video database based on short video data.

[0108] This multi-platform web crawler system, employing high concurrency, high stability, and the ability to circumvent anti-crawling mechanisms, continuously monitors and acquires publicly available data from mainstream short video data platforms (such as Chanmama and Feigua) via APIs. It accurately searches for keywords, including core and long-tail keywords, on target short video data platforms to obtain relevant short video content. It filters popular short videos based on preset filtering conditions, such as those published within the past month, with over X million likes, and over Y thousand comments, identifying highly interactive and popular videos. It extracts structured data from the selected popular short videos, automatically extracting information such as video titles, video text, publishing account type, video tags, and publication time. The extraction of the video text utilizes Automatic Speech Recognition (ASR) technology to transcribe spoken content into text.

[0109] In a preferred embodiment of this application, step S300 involves identifying the marketing intent of the short video data, extracting multi-dimensional features from short video data with different marketing intents, and generating short video copy based on the multi-dimensional features, including:

[0110] S301. Identify the marketing intent of short video data. Marketing intent includes driving traffic to stores, guiding group purchases, launching new products, and promoting brands.

[0111] S302. Extract multi-dimensional features from short video data with different marketing intentions. The multi-dimensional features include intention features, cross-type features, and industry style features.

[0112] S303. For the multi-dimensional features of short video data for each marketing intent, extract N feature information from each feature dimension to generate a multi-dimensional feature matrix I. M×N M is the feature dimension;

[0113] S304. Generate short video scripts based on multi-dimensional feature matrices.

[0114] Multi-label text classification models, such as those based on RoBERTa, are used to perform deep semantic analysis on the extracted short video data. This intelligently determines the marketing intent of each piece of text, categorizing it into four types: driving traffic to physical stores, guiding group buying, new product launches, and brand promotion. Driving traffic to physical stores focuses on guiding users to make purchases at physical stores, including information such as address, opening hours, and in-store coupons. Guiding group buying focuses on guiding users to purchase group buying packages or coupons online, including information such as group buying links and package details. New product launches announce and promote new products to the market, stimulating user interest and desire to try new things, including information such as new product launches, major releases, and brand-new offerings. Brand promotion focuses on establishing and strengthening the brand's image, values, and trust in the minds of users, cultivating long-term goodwill and loyalty, including information such as commitment to handmade, natural, and organic products, and dedication to providing a better experience.

[0115] We utilized a content feature learning model combining deep neural networks (such as Transformer and LSTM) and attention mechanisms to perform unsupervised and supervised learning on a large number of short video copy, extracting multi-dimensional features from short video data with different marketing intentions. These multi-dimensional features include intent features, cross-type features, and industry style features. Intent features include features for copy that drives traffic to stores, features for copy that guides group buying, features for copy that launches new products, and features for copy that promotes the brand. For the learning of features for copy that drives traffic to stores, we analyzed the commonalities of successful copy, including but not limited to: location-based keywords such as "specific business district" or "specific intersection," descriptions of in-store benefits such as "free drinks upon arrival" or "first 10 customers get a free meal," phrases that create a sense of atmosphere, vocabulary emphasizing the experience, and CTA (Call-to-Action) wording such as "come to the store and try it!" or "navigate to the store!" For learning the characteristics of copywriting guiding group buying, the commonalities of successful group buying copywriting are analyzed, including but not limited to: highlighting group buying packages, designing price anchors, scarcity marketing such as limited time and quantity, emphasizing discounts, and guiding the purchase path such as clicking the link below to buy. For cross-type feature learning, the characteristics of different types of copywriting in extracting general marketing structures such as opening attractiveness, pain point identification, trust endorsement, and closing guidance are learned simultaneously. For industry style feature learning, the specific style, common phrases, and emotional expression methods of popular short video copywriting in the industry category of the merchant are learned. Personalized and diversified short video copywriting is generated based on multi-dimensional feature vectors. Based on the information input by the merchant, the industry characteristics analyzed by AI, and the copywriting features learned from different marketing intentions, N high-quality short video copywritings with high similarity in marketing intent are automatically generated. N is configurable, for example, 5-10. The generated short video copy structure includes an engaging opening to quickly grab the user's attention; a pain point / need introduction, highlighting the pain points addressed by the merchant's products / services; product / service highlights showcasing the merchant's advantages; trust endorsements, such as local recognition and a high number of repeat customers; a clear CTA guiding users to the store or group buying; embedded promotional information, combining promotional details provided by the merchant; and multiple version generation, generating several different styles and emphases of copy for the same marketing objective, allowing merchants to choose the appropriate version.

[0116] In a preferred embodiment of this application, short video text generation is achieved through the following text generation model:

[0117] ;

[0118] Where T is the text vector, F I F is the intention feature vector. T For cross-type feature vectors, F S Let W be the industry style feature vector, W be the adaptive feature weight matrix, and Decode be the decoding function. For Hadama product, Here, is the semantic fusion operator, and tanh is the nonlinear activation function.

[0119] This copy generation model can deeply integrate three core features, accurately match four major marketing intentions, reuse feature matrices to improve efficiency, adapt to multiple industry scenarios, and generate copy that is suitable for the characteristics of short videos, taking into account both professionalism and communicability.

[0120] In a preferred embodiment of this application, in step S400, the digital human model is driven to broadcast based on the short video script to generate a short video, including:

[0121] S401. Obtain the digital human model;

[0122] S402. Convert short video scripts into digital human voice;

[0123] S403. Based on the digital human's voice, drive the digital human model to broadcast lip movements, facial expressions, and body movements to generate the first short video;

[0124] S404. Add audiovisual auxiliary information to the first short video to generate a second short video. The audiovisual auxiliary information includes background information, background music information, subtitle information, transition effect information, and special effects information.

[0125] By employing digital human-driven and synthesizer technologies for digital human asset management, businesses can create their own digital human avatars or choose from a database of digital humans provided by the system. AI-generated text is transformed into digital human speech using high-quality, natural-sounding TTS technology, supporting various timbres, speaking speeds, and emotions.

[0126] The AI-powered intelligent editor automatically matches and arranges materials based on the text content: it intelligently filters relevant visual content, such as store environment, product close-ups, and service scenarios, from the image / video material library provided by the merchant or the system's built-in copyright library. The system can automatically arrange the playback order and duration of the materials according to the rhythm and semantics of the text, achieving intelligent editing. Simultaneously, the AI-powered intelligent editor can automatically add smooth transition effects (such as fade-in / fade-out, push-pull, etc.) and appropriate video effects (including subtitle animation, local highlights, zoom in / out, etc.) to enhance visual appeal. It can also intelligently match background music (BGM) from the copyright music library based on the video's theme, emotion, and industry attributes, and automatically adjust the volume to ensure consistency with the spoken audio. Furthermore, the AI-powered intelligent editor can automatically generate recognizable and customizable subtitles and recommend attractive video titles based on the text's highlights. It supports outputting in multiple formats and resolutions (such as MP4, 1080P) that meet the requirements of mainstream short video platforms and provides compression optimization functions for easy uploading and publishing.

[0127] In a preferred embodiment of this application, step S500, determining the publishing information and publishing the short video, includes:

[0128] S501. For each marketing intent, obtain the set of account types, short video tags, and posting times for short video data;

[0129] S502. Extract the publishing account types from the set of publishing account types;

[0130] S502. Extract short video tags from the short video tag set;

[0131] S503. Count the number of release times falling into different time periods in the release time set, and extract the release time from the time period with the most release times.

[0132] S504. Publish short videos based on the account type, short video tags, and publication time.

[0133] For PI / SDK integration and authorization, merchants bind their short video platform accounts (such as Douyin Open Platform and Kuaishou Open Platform) to this system through OAuth or other authorization mechanisms. The authorized system can publish and obtain public data of the published content.

[0134] In a preferred embodiment of this application, step S600 involves acquiring recaptured data after the short video is published and optimizing the short video based on the recaptured data, including:

[0135] S601. Obtain the data collected after the short video is published. The collected data includes exposure data, interaction data, conversion data, and user behavior data.

[0136] S602. Based on the collected data, the short videos are comprehensively evaluated from multiple evaluation dimensions;

[0137] Specifically:

[0138] Calculate the overall evaluation score:

[0139] ;

[0140] Where F is the overall evaluation score, E is the exposure dimension score, D is the interaction dimension score, C is the conversion dimension score, B is the user behavior score, and ω1, ω2, ω3, and ω4 are the weights of different evaluation dimensions.

[0141] Exposure dimension score:

[0142] ;

[0143] Where V represents the number of views, CTR represents the number of clicks, CR represents the number of completed plays, and λ1, λ2, and λ3 represent the weights of different exposure dimensions.

[0144] The interaction dimension score is:

[0145] ;

[0146] Where L is the number of likes, Cm is the number of comments, Sh is the number of shares, Sv is the number of favorites, F is the number of followers, and μ1, μ2, μ3, μ4, and μ5 are the weights of different interaction dimensions.

[0147] Conversion dimension score:

[0148] ;

[0149] ;

[0150] ;

[0151] ;

[0152] Wherein, ν1, ν2, and ν3 are the weights of different transformation dimensions;

[0153] User behavior score:

[0154] ;

[0155] ;

[0156] ;

[0157] ;

[0158] Among them, ξ1, ξ2, and ξ3 are the weights of different user behavior dimensions;

[0159] S603. Optimize the short video copy, digital human model, posting account type, short video tags and / or posting time based on the comprehensive evaluation results.

[0160] The system uses a data crawler module to capture and update key data from merchants' published short videos in real-time or periodically. Based on these key data metrics, the system comprehensively evaluates the short videos and conducts multi-dimensional, in-depth cross-analysis, including A / B testing: if a merchant publishes multiple versions of a video, the system automatically performs A / B testing to compare the effects and identify the best version. User profile analysis: combining comment content and interaction data, the system analyzes the target user's preferences for different content. Content type effect comparison: comparing the conversion rates of videos that drive traffic to stores versus those that guide group buying. Traffic source analysis: analyzing video traffic sources such as recommendation feeds, search, and followers. Finally, the system generates detailed and actionable content optimization reports and recommendations. Copywriting optimization suggestions: for example, adding more impactful questions at the beginning of the copy, reducing the pace of spoken text, highlighting promotional information, trying to add localized keywords, and making the CTA more explicit, such as "click the homepage link to claim a coupon." Video presentation optimization suggestions: for example, adding more product close-ups, changing background music that doesn't match the copy's rhythm, and compressing excessively long videos to under 15 seconds. Optimization suggestions for posting strategies: For example, suggesting posting between 8-9 PM when user activity is higher, and trying to add interactive topics to the copy to encourage user comments. Based on these suggestions, the AI ​​copywriter optimizes the short video copy, digital human model, posting account type, short video tags, and / or posting time, using these as key parameters for the next short video copy generation. For example, if a certain type of copy is ineffective, the AI ​​copywriter will adjust its generation strategy, reducing the generation of that type of copy or automatically avoiding elements that contribute to poor performance. Personalized content recommendations are implemented, automatically recommending new content templates, copy structures, or digital human performance styles most likely to bring high conversion rates based on historical data and optimization results. New short video content is automatically generated based on optimization suggestions and can be automatically uploaded to the merchant's linked short video account, achieving a completely automated intelligent marketing loop.

[0161] The method for generating short videos for physical stores in this application has the following beneficial effects:

[0162] 1. Significantly lowers the barrier to entry for brick-and-mortar businesses to engage in short video marketing: No professional content planning, shooting, or editing skills are required; any business can easily get started and achieve automated content production.

[0163] 2. Significantly improve content production efficiency and reduce costs: Say goodbye to cumbersome manual production processes and high agency operation fees, and achieve massive, high-frequency content updates with extremely low marginal costs.

[0164] 3. Achieve precise marketing and high conversion rates: Through AI deep learning of industry and user preferences, generate customized content that is highly matched to the target audience and has high customer acquisition potential, thereby effectively improving store visit rates or group purchase redemption rates.

[0165] 4. Provides continuous and iterative marketing capabilities: The unique data feedback and intelligent optimization loop enables marketing content to be continuously adjusted and improved based on actual results, ensuring a sustained return on marketing investment.

[0166] 5. Build end-to-end, unattended marketing automation: From content creation, production, and distribution to performance analysis and optimization, the system achieves a high degree of automation, truly realizing intelligent marketing.

[0167] Based on the above-described method for generating short videos for physical stores, a second aspect of this application provides a device for generating short videos for physical stores, such as... Figure 2-4 As shown, it includes:

[0168] User information acquisition module 100 is used to acquire merchant information and generate keywords based on the merchant information;

[0169] The short video data acquisition module 200 is used to acquire short video data based on keywords.

[0170] The short video copy generation module 300 is used to identify the marketing intent of short video data, extract multi-dimensional features from short video data with different marketing intents, and generate short video copy based on the multi-dimensional features.

[0171] The short video generation module 400 is used to drive the digital human model to broadcast based on the short video script and generate short videos.

[0172] The short video publishing module 500 is used to determine the publishing information and publish the short video;

[0173] The short video optimization module 600 is used to obtain the data collected after the short video is published and to optimize the short video based on the data collected.

[0174] The physical merchant short video generation device of this application achieves an automated and intelligent closed loop of short video content generation, publishing, data feedback and optimization by deeply integrating AI (including natural language processing, machine learning, computer vision, etc.), digital human technology, big data analysis, web crawling and other technologies.

[0175] In another aspect of this application, a computer device is provided, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described method for generating short videos for physical merchants.

[0176] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer device 700 suitable for implementing the embodiments of this application. Figure 5 The computer device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments described in this application.

[0177] like Figure 5 As shown, the computer device 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 702 or programs loaded from storage section 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the device 700. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0178] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0179] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit (CPU) 701, it performs the functions defined in the methods of this application. It should be noted that the computer storage medium of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0180] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0181] The modules or units described in the embodiments of this application can be implemented in software or hardware. The described modules or units can also be located in a processor, and the names of these modules or units do not necessarily constitute a limitation on the module or unit itself.

[0182] In another aspect, this application also provides a computer-readable storage medium, which may be included in the apparatus described in the above embodiments; or it may exist independently and not assembled into the apparatus. The computer-readable storage medium carries one or more programs that, when executed by the apparatus, process data according to the aforementioned method for generating short videos for physical merchants.

[0183] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for generating short videos for physical stores, characterized in that, include: S100: Obtain merchant information and generate keywords based on the merchant information; S200. Obtain popular short video data based on the keywords; S300. Identify the marketing intent of the short video data, extract multi-dimensional features from the short video data with different marketing intents, and generate short video copy based on the multi-dimensional features, including: S301. Identify the marketing intent of the short video data, including the marketing intent of driving traffic to stores, guiding group purchases, launching new products, and promoting brands. S302. Extract multi-dimensional features from the short video data with different marketing intentions, wherein the multi-dimensional features include intention features, cross-type features and industry style features; S303. For the multi-dimensional features of the short video data for each marketing intent, extract N feature information from each feature dimension to generate a multi-dimensional feature matrix I. M×N M is the feature dimension; S304. Generate short video script based on the multi-dimensional feature matrix; S400: The digital human model is driven to broadcast according to the short video script, generating a short video; S500: Determine the publishing information and publish the short video; S600. Obtain the recollection data after the short video is published, and optimize the short video based on the recollection data.

2. The method for generating short videos for physical stores according to claim 1, characterized in that, In step S100, merchant information is obtained, and keywords are generated based on the merchant information, including: S101. Obtain merchant information, which includes store name, main products / services, store features and advantages, target customer groups, and marketing materials; S102. Construct an industry category identification network to identify the industry category to which the merchant belongs based on the merchant information. The industry category identification network is as follows: ； Where y is the industry category probability vector, Softmax is the normalization function, W is the weight of the classification layer, and b is the bias of the classification layer. T For text information, I For image / video information, α and β are the corresponding weights, MLP_fusion is a multilayer perceptron used for feature fusion, and GraphBuilder( S (This refers to merchant information) S Knowledge graph; S103. Generate keywords based on the industry category.

3. The method for generating short videos for physical stores according to claim 2, characterized in that, In S200, popular short video data is obtained based on the keywords, including: S201. Collect short video data that meets the set conditions from the short video platform based on the keywords; S202. Generate a structured short video database based on the short video data.

4. The method for generating short videos for physical stores according to claim 3, characterized in that, In S400, the digital human model is driven to broadcast based on the short video script, generating a short video, including: S401. Obtain the digital human model; S402. Convert the short video script into a digital human voice; S403. Based on the digital human's voice, drive the digital human model's lip movements, facial expressions, and body movements to broadcast, generating a first short video; S404. Add audiovisual auxiliary information to the first short video to generate a second short video. The audiovisual auxiliary information includes background information, background music information, subtitle information, transition effect information, and special effects information.

5. The method for generating short videos for physical stores according to claim 4, characterized in that, In S500, the publishing information is determined, and the short video is published, including: S501. For each marketing intent, obtain the set of publishing account types, the set of short video tags, and the set of publishing times for the short video data; S502. Extract the publishing account types from the set of publishing account types; S502. Extract short video tags from the short video tag set; S503. Count the number of release times falling into different time periods in the release time set, and extract the release time from the time period with the most release times; S504. Publish the short video according to the published account type, the short video tag, and the published time.

6. The method for generating short videos for physical stores according to claim 5, characterized in that, In S600, the recapture data after the short video is published is obtained, and the short video is optimized based on the recapture data, including: S601. Obtain the data collected after the short video is published, including exposure data, interaction data, conversion data, and user behavior data. S602. The short video is comprehensively evaluated from multiple evaluation dimensions based on the collected data; S603. Optimize the short video text, the digital human model, the publishing account type, the short video tags, and / or the publishing time based on the comprehensive evaluation results.

7. A short video generation device for physical stores, characterized in that, include: The merchant information acquisition module is used to acquire merchant information and generate keywords based on the merchant information; The short video data acquisition module is used to acquire popular short video data based on the keywords. The short video copy generation module is used to identify the marketing intent of the short video data, extract multi-dimensional features from the short video data with different marketing intents, and generate short video copy based on the multi-dimensional features, including: Identify the marketing intent of the short video data, including the intent to drive traffic to stores, guide group buying, launch new products, and promote brands. Multi-dimensional features are extracted from the short video data with different marketing intentions. These multi-dimensional features include intent features, cross-type features, and industry style features. For each marketing intent, the short video data has multi-dimensional features. N feature information from each feature dimension is extracted to generate a multi-dimensional feature matrix I. M×N M is the feature dimension; Generate short video scripts based on the multi-dimensional feature matrix; The short video generation module is used to drive the digital human model to broadcast based on the short video script and generate a short video. The short video publishing module is used to determine the publishing information and publish the short video; The short video optimization module is used to obtain the data collected after the short video is published, and to optimize the short video based on the data collected.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the computer program, it implements the method for generating short videos for physical merchants as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it can implement the method for generating short videos for physical merchants as described in any one of claims 1 to 6.

Images