Precision marketing method and system based on large model

Abstract

The invention discloses a precision marketing method and system based on a large model, and belongs to the technical field of data processing and marketing. Based on user information, historical browsing records and historical search records, short video analysis is carried out by adopting the large model, and meanwhile user similarity analysis is carried out; the final marketing content is generated by combining the two types of analysis, multi-modal data fusion marketing is realized, accurate marketing can be effectively realized under the condition of lacking historical behavior data and advertisement interaction records, the marketing effect and efficiency can be remarkably improved, and the method has wide application prospects and market value.

Description

Technical Field

[0001] The present invention belongs to the technical fields of data processing and marketing technology, and particularly relates to a precise marketing method and system based on a large model. Background Art

[0002] Short videos are a new form of Internet content, usually referring to video content with a duration ranging from a few seconds to several minutes. With the characteristics of being fast, intuitive, and easy to spread, they have quickly become an indispensable part of people's daily lives. Short video content is diverse, covering multiple fields such as entertainment, education, life, and news, meeting the viewing needs of users' fragmented time. Through short video platforms, users can not only watch content but also participate in creation and interaction, forming a powerful social network. The popularity of short videos has also promoted the development of related industries and become a new front for advertising and marketing. With the progress of technology and the continuous change of user needs, the short video industry is still continuously innovating and developing. At the same time, short video advertising not only serves as the main source of commercial revenue for short video platforms but also provides high-quality placement channels for advertisers. Therefore, how to push short video advertisements to interested users to achieve good marketing effects has become an important issue. Short video platforms can effectively improve the marketing performance of advertisements by using recommendation algorithm technology. However, in the existing technology, marketing usually relies on users' personal data and historical behavior data, resulting in a sharp decline in the performance of the recommendation algorithm and difficulty in achieving precise marketing in the absence of historical behavior data and advertising interaction records. Summary of the Invention The present invention provides a precise marketing method and system based on a large model to solve the problem that in the existing technology, marketing cannot be effectively achieved in the absence of historical behavior data and advertising interaction records.

[0003] On the one hand, the present invention provides a precise marketing method based on a large model, including: Obtaining user information, historical browsing records, and historical search records; wherein, both the historical browsing records and historical search records are short videos; Determining the similarity between any two users according to the user information and historical browsing records, and determining the target similar users corresponding to each user according to the similarity between any two users; Extracting first video features from the first historical search records corresponding to the target similar users of the user by using a large model, and generating first marketing content according to the first video features; Extracting second video features from the second historical search records of the user by using a large model, and generating second marketing content according to the second video features; Perform weighted sorting on the first marketing content and the second marketing content to obtain the final marketing content, and push the final marketing content to the user.

[0004] Further, determine the similarity between any two users according to the user information and historical browsing records, including: According to the historical browsing records corresponding to the user, obtain the tags attached when the historical browsing records were published, and obtain the set of browsing tags corresponding to the user; Determine the similarity between any two users according to the user information and the set of browsing tags corresponding to the user.

[0005] Further, determine the similarity between any two users according to the user information and the set of browsing tags corresponding to the user, including: According to the user information corresponding to the user, determine the gender similarity, age similarity, height similarity, and weight similarity between the user and other users; According to the set of browsing tags corresponding to the user, the interest similarity between the user and other users is determined as:

[0006] Wherein, represents the interest similarity between user A and other user B, represents the TD-IDF value corresponding to the h-th browsing tag of user A, H represents the total number of browsing tags corresponding to user A, represents the TD-IDF value of other user B regarding the h-th browsing tag, and when the h-th browsing tag does not exist in the set of browsing tags corresponding to other user B, is set to 0; Determine the similarity between any two users according to the gender similarity, age similarity, height similarity, weight similarity, and interest similarity.

[0007] Further, determine the target similar users corresponding to each user according to the similarity between any two users, including: For any one user, determine other users whose similarity with the user is greater than the preset user similarity threshold, and obtain the target similar users corresponding to the user.

[0008] Further, according to the first historical search records corresponding to the target similar users of the user, use a large model to extract the first video features in the first historical search records, and generate the first marketing content according to the first video features, including: For any one of the first historical search records corresponding to the target similar users corresponding to the user, perform frame extraction on the short video corresponding to the first historical search record to obtain multiple video frames corresponding to the first historical search record; For any one of the video frames corresponding to the first historical search record, use a large model to describe the video frame to generate the first video frame content text; For the first video frame content text corresponding to the first historical search record, use the mean filtering method for filtering to obtain the filtered first video frame content text; For the filtered first video frame content text corresponding to the first historical search record, use BERT to generate text feature vectors to obtain the first video feature corresponding to the first historical search record; Obtain the third video feature of the video to be marketed, and determine the similarity between the first video feature corresponding to the first historical search record and the third video feature of the video to be marketed to obtain the similarity between the first historical search record and the video to be marketed; Determine the first marketing content according to the similarity between the first historical search record and the video to be marketed.

[0009] Further, for the first video frame content text corresponding to the first historical search record, using the mean filtering method for filtering to obtain the filtered first video frame content text, including: For any one of the first video frame content texts corresponding to the first historical search record, obtain the cosine similarity score between the first video frame content text and other first video frame content texts; For any one of the first video frame content texts corresponding to the first historical search record, obtain the average value of all cosine similarity scores corresponding to the first video frame content text to obtain the similarity score mean corresponding to the first video frame content text; For the first video frame content text corresponding to the first historical search record, filter out the first video frame content text whose similarity score mean is less than the preset text similarity threshold, and use the unfiltered first video frame content text as the filtered first video frame content text.

[0010] Further, according to the second historical search record corresponding to the user, use a large model to extract the second video feature in the second historical search record, and generate the second marketing content according to the second video feature, including: For any one of the second historical search records corresponding to the user, perform frame extraction on the short video corresponding to the second historical search record to obtain multiple video frames corresponding to the second historical search record; For any one of the video frames corresponding to the second historical search record, use a large model to describe the video frame to generate the second video frame content text; For the text of the second video frame content corresponding to the second historical search record, the mean filtering method is used for filtering to obtain the text of the second video frame content after filtering; For the text of the second video frame content after filtering corresponding to the second historical search record, BERT is used to generate text feature vectors to obtain the second video features corresponding to the second historical search record; Obtain the third video features of the video to be marketed, and determine the similarity between the second video features corresponding to the second historical search record and the third video features of the video to be marketed, so as to obtain the similarity between the second historical search record and the video to be marketed; Determine the second marketing content according to the similarity between the second historical search record and the video to be marketed.

[0011] Further, for the text of the second video frame content corresponding to the second historical search record, the mean filtering method is used for filtering to obtain the text of the second video frame content after filtering, including: For any text of the second video frame content corresponding to the second historical search record, obtain the cosine similarity score between the text of the second video frame content and the text of other second video frame contents; For any text of the second video frame content corresponding to the second historical search record, obtain the average value of all cosine similarity scores corresponding to the text of the second video frame content to obtain the similarity score mean corresponding to the text of the second video frame content; For the text of the second video frame content corresponding to the second historical search record, filter out the text of the second video frame content whose similarity score mean is less than the preset text similarity threshold, and use the unfiltered text of the second video frame content as the text of the second video frame content after filtering.

[0012] Further, perform weighted sorting on the first marketing content and the second marketing content to obtain the final marketing content, and push the final marketing content to the user, including: Assign the same recommendation score to both the first marketing content and the second marketing content to obtain the recommendation score of the first marketing content and the recommendation score of the second marketing content; When the labels between the first marketing content and the second marketing content are the same, add the preset recommendation score increment to both the recommendation score of the first marketing content and the recommendation score of the second marketing content to obtain the recommendation score of the first marketing content after one update and the recommendation score of the second marketing content after one update; For any second marketing content, obtain the similarity between the second marketing content and the first marketing content to obtain the marketing similarity between the second marketing content and the first marketing content; For any second marketing content, when the marketing similarity between the second marketing content and the first marketing content is greater than a preset marketing similarity threshold, the corresponding first marketing content is used as the target marketing content, and the target marketing content corresponding to the second marketing content is obtained; According to the recommendation score of the first marketing content after the first update and the recommendation score of the second marketing content after the first update, transfer the recommendation score of the target marketing content to the corresponding second marketing content to obtain the recommendation score of the first marketing content after the second update and the recommendation score of the second marketing content after the second update; wherein, after the recommendation score of the target marketing content is transferred, the recommendation score of the target marketing content is set to zero; According to the recommendation score of the first marketing content after the second update and the recommendation score of the second marketing content after the second update, sort the first marketing content and the second marketing content in descending order of the recommendation score to obtain the sorted marketing content; Based on the sorted marketing content, take the top K marketing contents as the final marketing contents and push the final marketing contents to the user.

[0013] On the other hand, the present invention provides a precise marketing system based on a large model, including: an information acquisition module, a similar user acquisition module, a first marketing module, a second marketing module, and a comprehensive marketing module; The information acquisition module is used to acquire user information, historical browsing records, and historical search records; wherein, the historical browsing records are N latest short videos with the user viewing progress exceeding a preset viewing threshold; the historical search records are M latest short videos that the user has searched and the viewing progress exceeds the preset viewing threshold; The similar user acquisition module is used to determine the similarity between any two users according to the user information and the historical browsing records, and determine the target similar users corresponding to each user according to the similarity between any two users; The first marketing module is used to extract the first video features in the first historical search records by using a large model according to the first historical search records corresponding to the target similar users corresponding to the user, and generate the first marketing content according to the first video features; The second marketing module is used to extract the second video features in the second historical search records by using a large model according to the second historical search records corresponding to the user, and generate the second marketing content according to the second video features; The comprehensive marketing module is used to perform weighted sorting on the first marketing content and the second marketing content to obtain the final marketing content, and push the final marketing content to the user.

[0014] A precise marketing method and system based on a large model provided by the present invention, by using user information, historical browsing records, and historical search records as the basis, adopts a large model for short video analysis, and at the same time conducts user similarity analysis, combines the two analyses to generate the final marketing content, realizes multi-modal data fusion marketing, can effectively achieve precise marketing even in the absence of historical behavior data and advertising interaction records, can significantly improve the effect and efficiency of marketing, and has broad application prospects and market value. Brief Description of the Drawings

[0015] The accompanying drawings herein are incorporated into the specification and form a part of the specification, showing embodiments consistent with the present invention, and are used together with the specification to explain the principles of the present invention.

[0016] Figure 1 It is a flowchart of a precise marketing method based on a large model provided by an embodiment of the present invention.

[0017] Figure 2 It is a schematic structural diagram of a precise marketing system based on a large model provided by an embodiment of the present invention.

[0018] In the accompanying drawings, 201 - Information Acquisition Module, 202 - Similar User Acquisition Module, 203 - First Marketing Module, 204 - Second Marketing Module, 205 - Comprehensive Marketing Module.

[0019] Through the above accompanying drawings, specific embodiments of the present invention have been shown, and there will be more detailed descriptions hereinafter. These drawings and textual descriptions are not intended to limit the scope of the inventive concept in any way, but to illustrate the concept of the present invention to those skilled in the art by referring to specific embodiments. Detailed Embodiments

[0020] Here, exemplary embodiments will be described in detail, and the examples are shown in the accompanying drawings. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0021] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] As Figure 1 shown, an embodiment of the present invention provides a precise marketing method based on a large model, including: S101. Obtain user information, historical browsing records, and historical search records; wherein, both the historical browsing records and historical search records are short videos; The historical viewing records are short videos with a viewing progress exceeding a preset viewing progress threshold, and the number of collected historical viewing records is set to N, which are collected based on the current time (i.e., the marketing time); the historical search records are short videos that are searched and have a viewing progress exceeding the preset viewing progress threshold, and the number of collected historical search records is set to M, which are collected based on the current time (i.e., the marketing time).

[0023] In recent years, short video advertisements, as the main source of commercial revenue for short video platforms, also provide high-quality placement channels for advertisers. Therefore, how to push short video advertisements to interested users to achieve good marketing effects has become an important issue. Therefore, the embodiments of the present invention are mainly for marketing on short video platforms, and both historical viewing records and historical search records should be short videos.

[0024] At the same time, both historical viewing records and historical search records should be content that users are more interested in. Therefore, a viewing threshold needs to be set, and when the user's viewing progress exceeds this viewing threshold, it will be involved in the calculation. And the latest short videos refer to the short videos that the user has viewed at historical times based on the marketing time.

[0025] S102. Determine the similarity between any two users according to the user information and historical viewing records, and determine the target similar users corresponding to each user according to the similarity between any two users; Through the user information and historical viewing records, it can be determined whether the behaviors of two users are similar. If they are similar, the target similar users corresponding to the users can be determined, so as to conduct marketing based on the target similar users.

[0026] S103. Extract the first video features in the first historical search record corresponding to the target similar users corresponding to the user by using a large model, and generate the first marketing content according to the first video features; The first marketing content is generated based on the user similarity, and the user similarity is mainly obtained depending on the user information and historical viewing records. Therefore, compared with the traditional similarity algorithm, the embodiments of the present invention can also achieve precise marketing when the user does not have historical behavior data.

[0027] S104. Extract the second video features in the second historical search record corresponding to the user by using a large model, and generate the second marketing content according to the second video features; The second historical search record indicates the content that the user is more interested in at historical moments. By processing and identifying this part of the data through a large model, the second marketing content is generated, enabling precise marketing to be further achieved even without the user purchasing or searching for goods.

[0028] S105. Perform weighted sorting on the first marketing content and the second marketing content to obtain the final marketing content, and push the final marketing content to the user.

[0029] Both the first marketing content and the second marketing content are generated based on user data. By performing weighted sorting on the first marketing content and the second marketing content, more advertising videos that the user is interested in can be marketed, effectively improving the accuracy of marketing.

[0030] A precise marketing method and system based on a large model provided by the present invention, based on user information, historical browsing records, and historical search records, uses a large model to analyze short videos, and at the same time performs user similarity analysis. By combining the two analyses to generate the final marketing content, multi-modal data fusion marketing is achieved. It can effectively achieve precise marketing even in the absence of historical behavior data and advertising interaction records, significantly improving the effect and efficiency of marketing, and having broad application prospects and market value.

[0031] In an embodiment of the present invention, determining the similarity between any two users according to the user information and historical browsing records includes: According to the historical browsing records corresponding to the user, obtain the tags marked when the historical browsing records were published, and obtain the set of browsing tags corresponding to the user; It should be noted that the browsing tags corresponding to the historical browsing records may be the same. For any one browsing tag, the more historical browsing records corresponding to the browsing tag, the more it indicates that the user is more inclined to browse the short video style corresponding to the browsing tag. Therefore, marketing can be carried out based on this.

[0032] Determine the similarity between any two users according to the user information and the set of browsing tags corresponding to the user.

[0033] In an embodiment of the present invention, determining the similarity between any two users according to the user information and the set of browsing tags corresponding to the user includes: According to the user information corresponding to the user, determine the gender similarity, age similarity, height similarity, and weight similarity between the user and other users; For example, the gender similarity can be: when the genders of two users are the same, the gender similarity is considered 100%, otherwise it is 0%.

[0034] The age similarity can be as follows: An age difference threshold can be set in advance. When the age difference between two users is greater than or equal to this age difference threshold, the age similarity between the two users can be considered 0%. Similarly, when the age difference is zero, the age similarity between the two users can be considered 100%. When the age difference varies within this threshold, the changing percentage is taken as the age similarity. For example, if the age difference threshold is set to 10, when the age difference is 10, the age similarity between the two users is 0%; when the age difference is 9, which has changed by 10%, then the age similarity between the two users is 10%.

[0035] The height similarity can be as follows: A height difference threshold can be set in advance. When the height difference between two users is greater than or equal to this height difference threshold, the height similarity between the two users can be considered 0%. Similarly, when the height difference is zero, the height similarity between the two users can be considered 100%. When the height difference varies within this threshold, the changing percentage is taken as the height similarity.

[0036] The weight similarity can be as follows: A weight difference threshold can be set in advance. When the weight difference between two users is greater than or equal to this weight difference threshold, the weight similarity between the two users can be considered 0%. Similarly, when the weight difference is zero, the weight similarity between the two users can be considered 100%. When the weight difference varies within this threshold, the changing percentage is taken as the weight similarity.

[0037] Based on the browsing tag sets corresponding to the users, the interest similarity between a user and other users is determined as follows:

[0038] wherein, represents the interest similarity between user A and other user B, represents the TD-IDF value corresponding to the h-th browsing tag of user A, and H represents the total number of browsing tags corresponding to user A, represents the TD-IDF value of other user B regarding the h-th browsing tag, and when the h-th browsing tag does not exist in the browsing tag set corresponding to other user B, it is set to 0; The TF-IDF value is obtained by multiplying the term frequency (TF) and the inverse document frequency (IDF), and is used to evaluate the importance of a word to a document set or a corpus.

[0039] Based on the gender similarity, age similarity, height similarity, weight similarity, and interest similarity, the similarity between any two users is determined.

[0040] Optionally, the gender similarity, age similarity, height similarity, weight similarity, and interest similarity can be directly added as the similarity between two users. It can also be implemented by setting different weight coefficients for the gender similarity, age similarity, height similarity, weight similarity, and interest similarity respectively. After multiplying the gender similarity, age similarity, height similarity, weight similarity, and interest similarity by their respective corresponding weight coefficients and summing them up, the similarity between two users is obtained. These two ways of obtaining similarity can be selected according to actual needs. The embodiments of the present invention preferably adopt the second way to obtain user similarity.

[0041] In the embodiments of the present invention, according to the similarity between any two users, determining the target similar users corresponding to each user includes: For any one user, determining other users whose similarity with the user is greater than a preset user similarity threshold to obtain the target similar users corresponding to the user.

[0042] Optionally, for any one user, the L other users with the largest similarity can also be selected as the target similar users.

[0043] In the embodiments of the present invention, according to the first historical search records corresponding to the target similar users corresponding to the user, using a large model to extract the first video features in the first historical search records and generating the first marketing content based on the first video features includes: For any one of the first historical search records corresponding to the target similar users corresponding to the user, performing frame extraction on the short video corresponding to the first historical search record to obtain multiple video frames corresponding to the first historical search record; For example, based on a preset data acquisition frequency, video frames of the short video corresponding to the first historical search record can be collected, so that multiple video frames corresponding to the first historical search record can be obtained.

[0044] For any one video frame corresponding to the first historical search record, using a large model to describe the video frame to generate the first video frame content text; Optionally, in the embodiments of the present invention, the large model can be set to LLaVA (Large Language and Vision Assistant) or LLM (Large Language Model). The embodiments of the present invention preferably use LLaVA. However, it should be noted that the above large model is only an example of the embodiments of the present invention. Under the condition of ensuring the same or approximate functions, other large models can also be used to describe the video frames, utilize the ability of the large model to analyze and understand the image scene content, and enrich the understanding of the image content through the text content, thereby improving the overall understanding of the video content by the model and finally realizing marketing.

[0045] For the text of the first video frame content corresponding to the first historical search record, the mean filtering method is used for filtering to obtain the text of the first video frame content after filtering. In the process of recognizing the text of the first video frame content, there may be a problem that the text of the first video frame content does not match the video frame scene information. Therefore, the embodiments of the present invention use the mean filtering method for filtering to filter out the text of the first video frame content that does not match the video frame scene information, thereby improving the marketing accuracy.

[0046] For the text of the filtered first video frame content corresponding to the first historical search record, BERT (pre-trained language model) is used to generate text feature vectors to obtain the first video feature corresponding to the first historical search record. Optionally, all the text feature vectors of the first historical search record can be concatenated in chronological order to determine the first video feature corresponding to the first historical search record.

[0047] Obtain the third video feature of the video to be marketed, and determine the similarity between the first video feature corresponding to the first historical search record and the third video feature of the video to be marketed to obtain the similarity between the first historical search record and the video to be marketed. The method for obtaining the third video feature is the same as the methods for obtaining the first video feature and the second video feature, which will not be elaborated here. The cosine similarity between the first video feature and the third video feature can be determined to obtain the similarity between the first historical search record and the video to be marketed.

[0048] Determine the first marketing content according to the similarity between the first historical search record and the video to be marketed.

[0049] For example, according to the similarity between the first historical search record and the video to be marketed, Q videos to be marketed with the largest similarity can be determined to obtain Q first marketing contents.

[0050] In an embodiment of the present invention, for the text of the first video frame content corresponding to the first historical search record, the mean filtering method is used for filtering to obtain the text of the first video frame content after filtering, including: For any text of the first video frame content corresponding to the first historical search record, obtain the cosine similarity score between the text of the first video frame content and other texts of the first video frame content; For any text of the first video frame content corresponding to the first historical search record, obtain the average value of all cosine similarity scores corresponding to the text of the first video frame content to obtain the similarity score mean corresponding to the text of the first video frame content; For the text of the first video frame content corresponding to the first historical search record, filter out the text of the first video frame content whose similarity score mean is less than the preset text similarity threshold, and use the unfiltered text of the first video frame content as the text of the first video frame content after filtering.

[0051] The embodiment of the present invention uses the mean filtering method for filtering to filter out the text of the first video frame content that does not conform to the video frame scene information, thereby improving the marketing accuracy.

[0052] In an embodiment of the present invention, according to the second historical search record corresponding to the user, a large model is used to extract the second video feature in the second historical search record, and the second marketing content is generated according to the second video feature, including: For any second historical search record corresponding to the user, perform frame extraction on the short video corresponding to the second historical search record to obtain multiple video frames corresponding to the second historical search record; For any video frame corresponding to the second historical search record, use a large model to describe the video frame to generate the text of the second video frame content; For the text of the second video frame content corresponding to the second historical search record, use the mean filtering method for filtering to obtain the text of the second video frame content after filtering; For the text of the second video frame content after filtering corresponding to the second historical search record, use BERT to generate a text feature vector to obtain the second video feature corresponding to the second historical search record; Obtain the third video feature of the video to be marketed, and determine the similarity between the second video feature corresponding to the second historical search record and the third video feature of the video to be marketed to obtain the similarity between the second historical search record and the video to be marketed; Determine the second marketing content according to the similarity between the second historical search record and the video to be marketed.

[0053] The method for obtaining the second marketing content is similar to the method for obtaining the first marketing content, and the embodiment of the present invention will not elaborate further.

[0054] In an embodiment of the present invention, for the text of the second video frame content corresponding to the second historical search record, a mean filtering method is adopted for filtering to obtain the text of the second video frame content after filtering, including: For any text of the second video frame content corresponding to the second historical search record, obtain the cosine similarity score between the text of the second video frame content and other texts of the second video frame content; For any text of the second video frame content corresponding to the second historical search record, obtain the average value of all cosine similarity scores corresponding to the text of the second video frame content to obtain the mean similarity score corresponding to the text of the second video frame content; For the text of the second video frame content corresponding to the second historical search record, filter out the text of the second video frame content whose mean similarity score is less than the preset text similarity threshold, and use the unfiltered text of the second video frame content as the text of the second video frame content after filtering.

[0055] In an embodiment of the present invention, weighted sorting is performed on the first marketing content and the second marketing content to obtain the final marketing content, and the final marketing content is pushed to the user, including: Assign the same recommendation score to both the first marketing content and the second marketing content to obtain the recommendation score of the first marketing content and the recommendation score of the second marketing content; When the labels between the first marketing content and the second marketing content are the same, add the preset recommendation score increment to both the recommendation score of the first marketing content and the recommendation score of the second marketing content to obtain the recommendation score of the first marketing content after one update and the recommendation score of the second marketing content after one update; The first marketing content is the short video content concerned by the target similar users, and the second marketing content is the short video content concerned by the user himself. Therefore, when the two types of content are similar or close, they are more worthy of being recommended.

[0056] It should be noted that when there are multiple first marketing contents and second marketing contents corresponding to the same label, the recommendation score is added with the preset recommendation score increment at the same time.

[0057] For any second marketing content, obtain the similarity between the second marketing content and the first marketing content to obtain the marketing similarity between the second marketing content and the first marketing content; Optionally, for the marketing similarity between the second marketing content and the first marketing content, a video feature extraction method based on a large model can be used to determine the video features, and then the marketing similarity is determined according to the video features of the second marketing content and the video features of the first marketing content, that is, the cosine similarity between the video features.

[0058] For any second marketing content, when the marketing similarity between the second marketing content and the first marketing content is greater than a preset marketing similarity threshold, the corresponding first marketing content is used as the target marketing content, and the target marketing content corresponding to the second marketing content is obtained; According to the recommendation score of the first marketing content after the first update and the recommendation score of the second marketing content after the first update, transfer the recommendation score of the target marketing content to the corresponding second marketing content to obtain the recommendation score of the first marketing content after the second update and the recommendation score of the second marketing content after the second update; among them, after the recommendation score of the target marketing content is transferred, the recommendation score of the target marketing content is set to zero; Compared with the first marketing content, the second marketing content can better reflect the content that users are interested in. Therefore, when the first marketing content is similar to the second marketing content, the recommendation score of the second marketing content can be increased, thereby improving the marketing accuracy.

[0059] According to the recommendation score of the first marketing content after the second update and the recommendation score of the second marketing content after the second update, sort the first marketing content and the second marketing content in descending order of the recommendation score to obtain the sorted marketing content; Based on the sorted marketing content, take the top K marketing content as the final marketing content and push the final marketing content to the user.

[0060] It should be noted that in the process of pushing the final marketing content to the user, it can be pushed in order, and the pushing method is defined by the staff in actual applications.

[0061] As Figure 2 shown, an embodiment of the present invention provides a precise marketing system based on a large model, including: an information acquisition module 201, a similar user acquisition module 202, a first marketing module 203, a second marketing module 204, and a comprehensive marketing module 205; The information acquisition module 201 is used to acquire user information, historical browsing records, and historical search records; among them, the historical browsing records are N latest short videos whose viewing progress of the user exceeds a preset viewing threshold; the historical search records are M latest short videos that the user searches and whose viewing progress exceeds a preset viewing threshold; The similar user acquisition module 202 is used to determine the similarity between any two users according to the user information and the historical browsing records, and determine the target similar users corresponding to each user according to the similarity between the any two users; The first marketing module 203 is configured to extract first video features from the first historical search records corresponding to the target similar users corresponding to the user by using a large model, and generate first marketing content according to the first video features; The second marketing module 204 is configured to extract second video features from the second historical search records corresponding to the user by using a large model, and generate second marketing content according to the second video features; The comprehensive marketing module 205 is configured to perform weighted sorting on the first marketing content and the second marketing content to obtain final marketing content, and push the final marketing content to the user.

[0062] The precise marketing system based on a large model provided by an embodiment of the present invention can execute the above precise marketing method, and its principle and beneficial effects are similar, which will not be elaborated here.

[0063] Those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention, which follow the general principles of the present invention and include known common knowledge or conventional technical means in the technical field not disclosed by the present invention. It should be understood that the present invention is not limited to the precise structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present invention is only limited by the appended claims.

Claims

1. A precise marketing method based on a large model, characterized in that Including: Obtain user information, historical browsing records, and historical search records; among them, both the historical browsing records and historical search records are short videos. Determine the similarity between any two users according to the user information and historical browsing records, and determine the target similar users corresponding to each user according to the similarity between any two users. Extract the first video features in the first historical search record by using a large model according to the first historical search record corresponding to the target similar user of the user, and generate the first marketing content according to the first video features. Extract the second video features in the second historical search record by using a large model according to the second historical search record of the user, and generate the second marketing content according to the second video features. Perform weighted sorting on the first marketing content and the second marketing content to obtain the final marketing content, and push the final marketing content to the user.

2. The precise marketing method based on a large model according to claim 1, characterized in that Determine the similarity between any two users according to the user information and historical browsing records, including: Obtain the tags attached when the historical browsing record was published according to the historical browsing record corresponding to the user, and obtain the set of browsing tags corresponding to the user. Determine the similarity between any two users according to the user information corresponding to the user and the set of browsing tags.

3. The precise marketing method based on a large model according to claim 2, wherein Determine the similarity between any two users according to the user information corresponding to the user and the set of browsing tags, including: Determine the gender similarity, age similarity, height similarity, and weight similarity between the user and other users according to the user information corresponding to the user. Determine the interest similarity between the user and other users according to the set of browsing tags corresponding to the user as: Among them, represents the interest similarity between user A and other user B, represents the TD-IDF value corresponding to the h-th browsing tag of user A, where H represents the total number of browsing tags corresponding to user A, represents the TD-IDF value of other user B regarding the h-th browsing tag, and when the h-th browsing tag does not exist in the browsing tag set corresponding to other user B, it is set to 0; Determine the similarity between any two users according to the gender similarity, age similarity, height similarity, weight similarity, and interest similarity.

4. The precise marketing method based on a large model according to claim 3, characterized in that Determine the target similar users corresponding to each user according to the similarity between any two users, including: For any one user, determine other users whose similarity with the user is greater than the preset user similarity threshold to obtain the target similar users corresponding to the user.

5. The precise marketing method based on a large model according to claim 3, wherein Extract the first video features in the first historical search record by using a large model according to the first historical search record corresponding to the target similar user of the user, and generate the first marketing content according to the first video features, including: For any one first historical search record corresponding to the target similar user of the user, perform frame extraction on the short video corresponding to the first historical search record to obtain multiple video frames corresponding to the first historical search record. For any one video frame corresponding to the first historical search record, use a large model to describe the video frame and generate the first video frame content text. Filter the first video frame content text corresponding to the first historical search record by using the mean filtering method to obtain the filtered first video frame content text. Use BERT to generate text feature vectors for the filtered first video frame content text corresponding to the first historical search record to obtain the first video features corresponding to the first historical search record. Obtain the third video feature of the video to be marketed, and determine the similarity between the first video feature corresponding to the first historical search record and the third video feature of the video to be marketed, so as to obtain the similarity between the first historical search record and the video to be marketed; Determine the first marketing content according to the similarity between the first historical search record and the video to be marketed.

6. The precise marketing method based on a large model according to claim 5, wherein For the text of the first video frame content corresponding to the first historical search record, use the mean filtering method for filtering to obtain the text of the first video frame content after filtering, including: For any text of the first video frame content corresponding to the first historical search record, obtain the cosine similarity score between the text of the first video frame content and the text of other first video frame content; For any text of the first video frame content corresponding to the first historical search record, obtain the average value of all cosine similarity scores corresponding to the text of the first video frame content to obtain the average similarity score corresponding to the text of the first video frame content; For the text of the first video frame content corresponding to the first historical search record, filter out the text of the first video frame content whose average similarity score is less than the preset text similarity threshold, and use the unfiltered text of the first video frame content as the text of the first video frame content after filtering.

7. The precise marketing method based on the large model according to claim 6, wherein According to the second historical search record corresponding to the user, use a large model to extract the second video feature in the second historical search record, and generate the second marketing content according to the second video feature, including: For any second historical search record corresponding to the user, perform frame extraction on the short video corresponding to the second historical search record to obtain multiple video frames corresponding to the second historical search record; For any video frame corresponding to the second historical search record, use a large model to describe the video frame to generate the text of the second video frame content; For the text of the second video frame content corresponding to the second historical search record, use the mean filtering method for filtering to obtain the text of the second video frame content after filtering; For the text of the second video frame content after filtering corresponding to the second historical search record, use BERT to generate a text feature vector to obtain the second video feature corresponding to the second historical search record; Obtain the third video feature of the video to be marketed, and determine the similarity between the second video feature corresponding to the second historical search record and the third video feature of the video to be marketed, so as to obtain the similarity between the second historical search record and the video to be marketed; Determine the second marketing content according to the similarity between the second historical search record and the video to be marketed.

8. The precise marketing method based on a large model according to claim 7, wherein For the text of the second video frame content corresponding to the second historical search record, use the mean filtering method for filtering to obtain the text of the second video frame content after filtering, including: For any text of the second video frame content corresponding to the second historical search record, obtain the cosine similarity score between the text of the second video frame content and the text of other second video frame content; For any text of the second video frame content corresponding to the second historical search record, obtain the average value of all cosine similarity scores corresponding to the text of the second video frame content to obtain the average similarity score corresponding to the text of the second video frame content; For the text of the second video frame content corresponding to the second historical search record, filter out the text of the second video frame content whose average similarity score is less than the preset text similarity threshold, and use the unfiltered text of the second video frame content as the text of the second video frame content after filtering.

9. The precise marketing method based on a large model according to claim 7, wherein Weighted sorting is performed on the first marketing content and the second marketing content to obtain the final marketing content, and the final marketing content is pushed to the user, including: Assign the same recommendation score to both the first marketing content and the second marketing content to obtain the recommendation score of the first marketing content and the recommendation score of the second marketing content; When the tags between the first marketing content and the second marketing content are the same, add the preset recommendation score increment to both the recommendation score of the first marketing content and the recommendation score of the second marketing content to obtain the recommendation score of the first marketing content after the first update and the recommendation score of the second marketing content after the first update; For any second marketing content, obtain the similarity between the second marketing content and the first marketing content to obtain the marketing similarity between the second marketing content and the first marketing content; For any second marketing content, when the marketing similarity between the second marketing content and the first marketing content is greater than the preset marketing similarity threshold, use the corresponding first marketing content as the target marketing content to obtain the target marketing content corresponding to the second marketing content; According to the recommendation score of the first marketing content after the first update and the recommendation score of the second marketing content after the first update, transfer the recommendation score of the target marketing content to the corresponding second marketing content to obtain the recommendation score of the first marketing content after the second update and the recommendation score of the second marketing content after the second update; among them, after the recommendation score of the target marketing content is transferred, set the recommendation score of the target marketing content to zero; According to the recommendation score of the first marketing content after the second update and the recommendation score of the second marketing content after the second update, sort the first marketing content and the second marketing content in descending order of the recommendation score to obtain the sorted marketing content; Based on the sorted marketing content, take the first K marketing contents as the final marketing content and push the final marketing content to the user.

10. A precise marketing system based on a large model, characterized in that, Including: An information acquisition module, a similar user acquisition module, a first marketing module, a second marketing module, and a comprehensive marketing module; The information acquisition module is used to acquire user information, historical browsing records, and historical search records; among them, the historical browsing records are N latest short videos with the user's viewing progress exceeding the preset viewing threshold; the historical search records are M latest short videos that the user has searched and whose viewing progress exceeds the preset viewing threshold; The similar user acquisition module is used to determine the similarity between any two users according to the user information and the historical browsing records, and determine the target similar user corresponding to each user according to the similarity between any two users; The first marketing module is used to extract first video features from the first historical search records corresponding to the target similar users corresponding to the user by using a large model, and generate first marketing content according to the first video features; The second marketing module is used to extract second video features from the second historical search records corresponding to the user by using a large model, and generate second marketing content according to the second video features; The comprehensive marketing module is used to perform weighted sorting on the first marketing content and the second marketing content to obtain the final marketing content, and push the final marketing content to the user.

Images