Artificial intelligence-based marketing video generation method

Abstract

The present application relates to the technical field of video automatic generation, and more particularly to a marketing video generation method based on artificial intelligence, comprising the following steps: collecting target product information and associated video materials, splitting the video materials into independent material nodes through image processing, and completing labeling; then, real-time searching of network discussion topics, calculation of flow coefficient, and locking of high-value hotspots; then, screening of adaptive materials through accurate matching of topics and material labels; finally, scene change recognition based on a ResNet model, automatic matching of transitions, completion of material splicing, generation of a video, significant improvement of video production efficiency and propagation accuracy, reduction of production cost, optimization of user viewing experience, and contribution to improvement of video flow data.

Description

Technical Field

[0001] This invention relates to the field of automatic video generation technology, and more particularly to a method for generating marketing videos based on artificial intelligence. Background Technology

[0002] With the advent of the digital age, the e-commerce industry is experiencing rapid growth. Video, as an intuitive, vivid, and easily disseminated format, has become an important medium for e-commerce marketing. The deep integration of artificial intelligence (AI) technology and video production technology is continuously developing in the field of e-commerce marketing.

[0003] Existing technology CN119130502A discloses an AI-based method, apparatus, medium, and product for generating short marketing videos for products. The method includes: constructing a short video industry knowledge base; obtaining user-inputted question information; searching for answers in the short video industry knowledge base based on the question information; obtaining a video selected by the user from the answers as a reference short video; and obtaining product information uploaded by the user. The product information includes product selling points and product materials. Reference text is retrieved from the short video industry knowledge base based on the product selling points, and the reference text is used as supplementary knowledge information. A script is generated based on the reference short video and the supplementary knowledge information, and visual information is generated based on the script and product materials. The visual information is then edited to obtain a short video containing product information.

[0004] However, traditional marketing videos rely on manual planning, material editing, and special effects production. From demand analysis to final output, it usually takes several days or even weeks, which is difficult to adapt to the short-lived nature of trending topics. This causes brands to miss the traffic window. At the same time, the massive amount of video materials accumulated by enterprises are mostly stored in a scattered form, lacking standardized tags and node-based management. It is impossible to quickly locate materials that match trending topics and product selling points, resulting in low material reuse rate and easy waste of resources. Summary of the Invention

[0005] The purpose of this invention is to solve the problems in the background art by proposing an artificial intelligence-based marketing video generation method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: An AI-based marketing video generation method, which includes the following steps: Step 1: Collect product information of the target marketing product, obtain relevant video materials, process all video materials, set sub-nodes, and divide the video materials into several independent materials based on the sub-nodes; Step 2: Identify the information text in the independent materials, match the information text with the product information, and determine the material tags for each video material; Step 3: Search for topics in real-time discussions on the Internet, identify the information matching value between the discussion topics and the target marketing products, determine related topics based on the information matching value, analyze the topic traffic, calculate the traffic coefficient of related topics based on traffic peaks and real-time traffic data, and determine the current hot topics based on the traffic coefficient. Step 4: Based on the matching information between trending topics and target marketing products, select material tags that are the same as the topic's related words, and mark the independent materials corresponding to the selected material tags as suitable materials for this trending topic; Step 5: Based on the selected adaptation materials, use the ResNet model to identify scene changes of adjacent materials, and automatically match the optimal transition to connect the adaptation materials, ultimately automatically generating a marketing video for the target marketing product.

[0007] As a further aspect of the present invention, the acquisition methods for relevant video materials include internal material channels and external material channels. Internal material channels refer to the brand's official product promotional videos, product usage tutorial videos filmed internally by the company, and real-life videos based on user feedback. External material channels include public social media platforms, industry-specific video platforms, user review video sections of e-commerce platforms, professional video material websites, and product review videos. Product information includes basic attribute information and core value information. Basic attribute information refers to product name, model specifications, appearance design, core parameters, price positioning, target audience, usage method, manufacturing process, and brand background. Core value information includes the product's core selling points, the user pain points it solves, application scenarios, competitive advantages, user reputation, and user reviews.

[0008] As a further aspect of the present invention, the method for determining the independent material includes: Obtain product information for the target marketing products, and based on this information, acquire relevant video materials for the target marketing products in a targeted manner; Identify the playback duration Tb of each video clip and compare it with the duration threshold Ty. If Tb ≤ Ty, mark the video clip as an independent clip; if Tb > Ty, mark the corresponding clip as a mixed clip. Acquire all the mixed footage, mark the mixed footage as target analysis footage in sequence, set the unit frame rate, and set a number of image frames for the target analysis footage according to the unit frame rate; Following the time sequence, starting from the initial image frame, adjacent image frames are processed sequentially to obtain image similarity values. If the image similarity value is greater than or equal to the similarity threshold X1, the image frames are merged. Conversely, if the image similarity value is less than the similarity threshold X1, a sub-node is set between these adjacent image frames. Identify the location of sub-nodes, starting from the initial image frame, and divide the target analysis material into several local materials based on the sub-nodes; Next, identify the duration Tf of each local clip, compare the duration Tf of the local clip with the duration threshold Ty again, and mark the local clips with Tf≤Ty as independent clips.

[0009] As a further aspect of the present invention, if Tf > Ty, then all image frames and corresponding screen similarity values ​​in this local material are obtained, the minimum value among the screen similarity values ​​is identified, the two image frames corresponding to the minimum value are identified, and a sub-node is set between these adjacent image frames. Based on the sub-node, the local material is segmented again to obtain new local material. The duration of the new local material is identified and compared with the duration threshold Ty. If the duration of the new local material is greater than the duration threshold Ty, the above method is used again until all independent materials are determined.

[0010] As a further aspect of the present invention, the method for determining the material tags of video materials includes: Select individual materials sequentially from the material database, use visual feature extraction algorithms to identify video frames in the individual materials, and convert the identified frame information into information text; Then, a hybrid deep learning model is used to identify the speech information in the independent material to obtain speech text. The speech text is compared with the prompt text. If the speech text and the prompt text are the same, only the speech text is selected as the information text. If the speech text and the prompt text are different, the speech text and the prompt text are used as information text respectively. The prompt text refers to the subtitle information that appears in the video frame of the independent material. All information texts are acquired and preprocessed. Preprocessing involves removing meaningless words from the information texts. Then, product information of the target marketing product is acquired. A semantic matching method based on deep learning is used to match the preprocessed information texts with the product information. Information words that match the information texts are selected from the product information and these information words are marked as material tags for this independent material.

[0011] As a further aspect of the present invention, the method for determining the related topics includes: Identify the video delivery platforms for the target marketing product and mark them as target platforms in sequence; Using the current time as the time node, identify the real-time discussion topics on the target platform at the current time node, and identify the information matching value between the discussion topics and the target marketing products. The information matching value is obtained after processing and calculating the topic model similarity. The information matching value is compared with the matching threshold: if the information matching value is greater than or equal to the matching threshold, the discussion topic is marked as a relevant topic; otherwise, if the information matching value is less than the matching threshold, the discussion topic is marked as an irrelevant topic.

[0012] As a further aspect of the present invention, the method for determining hot topics includes: Set a pre-observation period and collect topic traffic data for related topics within that period. Set time as the horizontal axis and topic traffic data as the vertical axis to set up a plane coordinate system. Then, visualize the topic traffic data in the previous observation time in the plane coordinate system to obtain the topic traffic curve. Identify peak traffic points in the topic traffic curve and mark the corresponding time period as peak period Tp. Then, identify the real-time traffic data Rs and the real-time rate of change Ks in the topic traffic curve, and then use the calculation formula... Obtain the traffic coefficient Lx for the related topic, where Ts represents the real-time period, Rb is the baseline traffic threshold of the target platform, and b1 is the base coefficient, with 0 < b1 < 1. and These are the proportionality coefficients; The traffic coefficient Lx of the related topic is compared with the threshold change coefficient Ly. If Lx < Ly, the corresponding related topic is marked as a declining topic. Conversely, if Lx ≥ Ly, the corresponding related topic is marked as a hot topic.

[0013] As a further aspect of the present invention, the method for determining the adaptable material includes: The process involves acquiring trending topics, identifying their matching information with target marketing products, and marking these matching information as topic-related terms. Then, all independent creative materials and their corresponding material tags are obtained. Next, using TF-IDF combined with cosine similarity algorithm, the similarity between topic-related terms and material tags is calculated to obtain a cosine value. Material tags with cosine values ​​greater than or equal to a threshold Y1 are selected, and the independent creative materials corresponding to these selected material tags are marked as suitable creative materials for this trending topic.

[0014] Compared with existing technologies, the advantages of this invention are: This invention achieves intelligent generation of marketing videos through a full-process AI-driven approach: it includes collecting target product information and related video materials, processing images to break them down into independent segments and tagging them; then, it searches online discussion topics in real time, calculates traffic coefficients to identify high-value hot topics; next, it selects suitable materials through precise matching of topics and material tags; finally, it identifies scene changes based on a ResNet model and automatically matches transitions to complete the splicing of materials and generate a video, significantly improving video production efficiency and dissemination accuracy, reducing production costs, optimizing user viewing experience, and helping to increase video playback traffic. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method flow structure of the present invention. Detailed Implementation

[0016] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0017] Reference Figure 1 An AI-based marketing video generation method, which specifically includes the following steps: Step 1: Identify the target marketing product, collect product information for the target marketing product, and based on this information, acquire video materials related to the target marketing product. Preprocess all video materials to determine the material database. Specific methods for determining the material database include: Obtain product information for the target marketing products, and based on the product information, acquire relevant video materials for the target marketing products. The acquisition methods for relevant video materials include internal material channels and external material channels. Internal material channels refer to the brand's official product promotional videos, product usage tutorial videos filmed internally by the company, and real-life videos based on user feedback. External material channels include public social media platforms, industry vertical video platforms, user review video sections of e-commerce platforms, professional video material websites, and product review videos, etc. Furthermore, product information includes basic attribute information and core value information. Basic attribute information refers to product name, model specifications, appearance design, core parameters (performance indicators, functional configuration, battery life, service life, etc.), price positioning, target audience, usage method, manufacturing process, brand background, etc.; core value information includes the product's core selling points (differentiated advantages, such as technological innovation, material upgrades, unique functions, high cost performance, novel design, etc.), solutions to user pain points, application scenarios, competitive advantages, user reputation and evaluation highlights, etc. The playback duration Tb of each video clip is identified and compared with the duration threshold Ty. If Tb ≤ Ty, the video clip is marked as an independent clip. If Tb > Ty, the corresponding clip is marked as a mixed clip. The specific duration of the duration threshold Ty is set by those skilled in the art based on big data experience. Obtain all the mixed materials, arbitrarily select one mixed material, and use this mixed material as an example to mark it as the target analysis material; Set a unit frame rate and set a number of image frames for the target analysis material according to the unit frame rate. The specific value of the unit frame rate is set by those skilled in the art based on big data experience. In this embodiment, the unit frame rate is set to 30fps. Following the chronological order, starting from the initial image frame, adjacent image frames are sequentially processed for similarity to obtain image similarity values. If the image similarity value is greater than or equal to the similarity threshold X1, the image frames are merged. Conversely, if the image similarity value is less than the similarity threshold X1, a sub-node is set between adjacent image frames. The specific value of the similarity threshold X1 is obtained by those skilled in the art based on big data calculations. The similarity processing method uses a CNN feature extraction algorithm, and the specific processing procedure is existing technology, which will not be elaborated here. Identify the location of sub-nodes, starting from the initial image frame, and divide the target analysis material into several local materials based on the sub-nodes; Next, identify the duration Tf of each local clip and compare the duration Tf of the local clip with the duration threshold Ty again. If Tf ≤ Ty, mark this local clip as an independent clip. If Tf > Ty, obtain all image frames in this local clip and their corresponding screen similarity values, identify the minimum value among the screen similarity values, identify the two image frames corresponding to the minimum value, and set a sub-node between these adjacent image frames. Based on the sub-node, segment the local clip again to obtain new local clips. Identify the duration of the new local clip and compare it with the duration threshold Ty. If the duration of the new local clip is greater than the duration threshold Ty, process it according to the above method until all independent clips are determined. Then, all the independent materials are collected and integrated to obtain a material database for the target marketing product; Step Two: Tag identification is performed on the individual clips in the media database to determine the media tags for each video clip. Specific methods for determining media tags include: Select any independent material from the material database. Using this independent material as an example, first use a visual feature extraction algorithm to identify the video frames in the independent material, and then convert the identified frame information into information text. Then, the semantic recognition algorithm is used to identify the speech information in the independent material to obtain speech text. The speech text is compared with the prompt text. If the speech text and the prompt text are the same, only the speech text is selected as the information text. If the speech text and the prompt text are different, the speech text and the prompt text are used as information text respectively. The prompt text refers to the subtitle information that appears in the video frame of the independent material. Furthermore, in this embodiment, the visual feature extraction algorithm uses a combination mechanism of recurrent neural network and natural language processing, and the semantic recognition algorithm uses a hybrid deep learning model. All information texts are acquired and preprocessed, including removing meaningless words. Then, product information of the target marketing product is acquired, and the preprocessed information texts are matched with the product information. Information words that match the information texts are selected from the product information and tagged as material tags for this independent material. The matching of information texts with product information adopts a semantic matching method based on deep learning, such as pre-trained word vectors + similarity calculation method or cross-modal pre-trained model algorithm. Process the remaining independent materials in the same way as described above, and determine the material tags for each independent material; Step 3: Search for topics in real-time discussion on the internet and analyze the traffic to these topics to identify current trending topics. Specific methods for identifying trending topics include: Identify the video delivery platforms for the target marketing product. Select any video delivery platform and mark it as the target platform. Using the current time as the time node, identify the real-time discussion topics on the target platform at the current time node, and identify the information matching value between the discussion topics and the target marketing products. The information matching value is obtained by topic model (LDA) similarity processing and calculation. The topic model (LDA) similarity is an existing technology and will not be elaborated here. The information matching value is compared with the matching threshold: if the information matching value is greater than or equal to the matching threshold, the discussion topic is marked as a relevant topic; otherwise, if the information matching value is less than the matching threshold, the discussion topic is marked as an irrelevant topic. Set a pre-observation period and collect topic traffic data for related topics within that period. The specific value of the pre-observation time is set by those skilled in the art based on big data experience. In this embodiment, the pre-observation time is set to one week. Set time as the horizontal axis and topic traffic data as the vertical axis to set up a plane coordinate system. Then, visualize the topic traffic data in the previous observation time in the plane coordinate system to obtain the topic traffic curve. Identify peak traffic points in the topic traffic curve and mark the corresponding time period as peak period Tp. Then, identify the real-time traffic data Rs and the real-time rate of change Ks in the topic traffic curve, and then use the calculation formula... Obtain the traffic coefficient Lx for the related topic, where Ts represents the real-time period, Rb is the baseline traffic threshold of the target platform, and b1 is the base coefficient, with 0 < b1 < 1. and These are the proportionality coefficients; The traffic coefficient Lx of the related topic is compared with the threshold change coefficient Ly. If Lx < Ly, the corresponding related topic is marked as a declining topic. Conversely, if Lx ≥ Ly, the corresponding related topic is marked as a hot topic. The specific value of the threshold change coefficient Ly was obtained by those skilled in the art based on big data calculations. Step 4: Obtain trending topics, identify the matching information between trending topics and target marketing products, and mark the matching information as topic-related words. Then, obtain all independent materials and their corresponding material tags, and use the text similarity algorithm to select material tags that are the same as the topic-related words. Mark the independent materials corresponding to the selected material tags as suitable materials for this trending topic. The text similarity algorithm used here is a combination of TF-IDF and cosine similarity algorithm. That is, the text is converted into TF-IDF vectors and the cosine value between the vectors is calculated. Material tags with cosine values ​​greater than or equal to the threshold Y1 are selected, and the independent materials corresponding to the selected material tags are marked as suitable materials for this hot topic. The specific value of the threshold Y1 is obtained by those skilled in the art based on big data calculations. Step 5: Based on the selected adaptation materials, use the ResNet model to identify scene changes of adjacent materials, and automatically match the optimal transition to connect the adaptation materials, ultimately automatically generating a marketing video for the target marketing product.

[0018] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for generating a marketing video based on artificial intelligence, characterized by, The method specifically includes the following steps: Step 1: Collect product information of the target marketing product, obtain relevant video materials, process all video materials, set sub-nodes, and divide the video materials into several independent materials based on the sub-nodes; Step 2: Identify the information text in the independent materials, match the information text with the product information, and determine the material tags for each video material; Step 3: Search for topics in real-time discussions on the Internet, identify the information matching value between the discussion topics and the target marketing products, determine related topics based on the information matching value, analyze the topic traffic, calculate the traffic coefficient of related topics based on traffic peaks and real-time traffic data, and determine the current hot topics based on the traffic coefficient. Step 4: Based on the matching information between trending topics and target marketing products, select material tags that are the same as the topic's related words, and mark the independent materials corresponding to the selected material tags as suitable materials for this trending topic; Step 5: Based on the selected adaptation materials, use the ResNet model to identify scene changes of adjacent materials, and automatically match the optimal transition to connect the adaptation materials, ultimately automatically generating a marketing video for the target marketing product. 2.The AI-based marketing video generation method of claim 1, wherein, The acquisition of relevant video materials includes internal and external material channels. Internal material channels refer to the brand's official product promotional videos, product usage tutorial videos filmed internally by the company, and real-life videos based on user feedback. External material channels include public social media platforms, industry-specific video platforms, user review video sections of e-commerce platforms, professional video material websites, and product review videos. Product information includes basic attribute information and core value information. Basic attribute information refers to product name, model specifications, appearance design, core parameters, price positioning, target audience, usage method, manufacturing process, and brand background. Core value information includes the product's core selling points, the user pain points it solves, application scenarios, competitive advantages, user reputation, and user reviews. 3.The AI-based marketing video generation method of claim 1, wherein, Methods for determining independent source material include: Obtain product information for the target marketing products, and based on this information, acquire relevant video materials for the target marketing products in a targeted manner; Identify the playback duration Tb of each video clip and compare it with the duration threshold Ty. If Tb ≤ Ty, mark the video clip as an independent clip; if Tb > Ty, mark the corresponding clip as a mixed clip. Acquire all the mixed footage, mark the mixed footage as target analysis footage in sequence, set the unit frame rate, and set a number of image frames for the target analysis footage according to the unit frame rate; Following the time sequence, starting from the initial image frame, adjacent image frames are processed sequentially to obtain image similarity values. If the image similarity value is greater than or equal to the similarity threshold X1, the image frames are merged. Conversely, if the image similarity value is less than the similarity threshold X1, a sub-node is set between these adjacent image frames. Identify the location of sub-nodes, starting from the initial image frame, and divide the target analysis material into several local materials based on the sub-nodes; Next, identify the duration Tf of each local clip, compare the duration Tf of the local clip with the duration threshold Ty again, and mark the local clips with Tf≤Ty as independent clips. 4.The AI-based marketing video generation method of claim 3, wherein, If Tf > Ty, then obtain all image frames and their corresponding screen similarity values ​​in this local material, identify the minimum value among the screen similarity values, identify the two image frames corresponding to the minimum value, and set a sub-node between these adjacent image frames. Based on the sub-node, the local material is segmented again to obtain new local material. The duration of the new local material is identified and compared with the duration threshold Ty. If the duration of the new local material is greater than the duration threshold Ty, then the above method is used again until all independent materials are determined.

5. The marketing video generation method based on artificial intelligence according to claim 1, characterized in that, Methods for determining the tags for video footage include: Select individual materials sequentially from the material database, use visual feature extraction algorithms to identify video frames in the individual materials, and convert the identified frame information into information text; Then, a hybrid deep learning model is used to identify the speech information in the independent material to obtain speech text. The speech text is compared with the prompt text. If the speech text and the prompt text are the same, only the speech text is selected as the information text. If the speech text and the prompt text are different, the speech text and the prompt text are used as information text respectively. The prompt text refers to the subtitle information that appears in the video frame of the independent material. All information texts are acquired and preprocessed. Preprocessing involves removing meaningless words from the information texts. Then, product information of the target marketing product is acquired. A semantic matching method based on deep learning is used to match the preprocessed information texts with the product information. Information words that match the information texts are selected from the product information and these information words are marked as material tags for this independent material.

6. The marketing video generation method based on artificial intelligence according to claim 1, characterized in that, Methods for determining related topics include: Identify the video delivery platforms for the target marketing product and mark them as target platforms in sequence; Using the current time as a time node, identify the real-time discussion topics on the target platform at the current time node, and identify the information matching value between the discussion topics and the target marketing products. The information matching value is obtained after processing and calculating the topic model similarity. The information matching value is compared with the matching threshold: if the information matching value is greater than or equal to the matching threshold, the discussion topic is marked as a relevant topic; otherwise, if the information matching value is less than the matching threshold, the discussion topic is marked as an irrelevant topic.

7. The marketing video generation method based on artificial intelligence according to claim 6, characterized in that, Methods for identifying trending topics include: Set a pre-observation period and collect topic traffic data for related topics within that period. Set time as the horizontal axis and topic traffic data as the vertical axis to set up a plane coordinate system. Then, visualize the topic traffic data in the previous observation time in the plane coordinate system to obtain the topic traffic curve. Identify peak traffic points in the topic traffic curve and mark the corresponding time period as peak period Tp. Then, identify the real-time traffic data Rs and the real-time rate of change Ks in the topic traffic curve, and then use the calculation formula... Obtain the traffic coefficient Lx for the related topic, where Ts represents the real-time period, Rb is the baseline traffic threshold of the target platform, and b1 is the base coefficient, with 0 < b1 < 1. and These are the proportionality coefficients; The traffic coefficient Lx of the related topic is compared with the threshold change coefficient Ly. If Lx < Ly, the corresponding related topic is marked as a declining topic. Conversely, if Lx ≥ Ly, the corresponding related topic is marked as a hot topic.

8. The marketing video generation method based on artificial intelligence according to claim 1, characterized in that, Methods for determining suitable materials include: The process involves acquiring trending topics, identifying their matching information with target marketing products, and marking these matching information as topic-related terms. Then, it involves acquiring all independent creative materials and their corresponding material tags. Next, using TF-IDF combined with cosine similarity algorithm, it calculates the similarity between topic-related terms and material tags to obtain cosine values. Material tags with cosine values ​​greater than or equal to a threshold Y1 are selected, and the independent creative materials corresponding to these selected material tags are marked as suitable creative materials for this trending topic.

Images