Intelligent matching short video goods-carrying recommendation method and management system

Abstract

The invention discloses an intelligent matching short video goods-carrying recommendation method and management system, and relates to the technical field of short video goods-carrying recommendation. Comprising the following steps: information collection: obtaining user information, wherein the user information comprises browsing information, a like unit, past purchased commodity information and user basic information; according to the method, the recommendation information set can be obtained through the set information extraction method, derivation method and past purchased commodity information, the relevant goods-carrying video set can be obtained through the cooperation of the recommendation information set and the goods-carrying video library, and the type of video preferred by the user can be judged according to the latest real-time browsing information of the user through the set judgment method, so that the user experience is improved. According to the method, the goods taking video of the user preference type can be obtained according to the user preference type in cooperation with the related goods taking video set, and the goods taking video of the user preference type is recommended to the user, so that the transaction efficiency of the goods can be improved while the use experience of the user is improved, and both the user and a merchant are benefited.

Description

Technical Field

[0001] The present invention relates to the technical field of short video product recommendation, and specifically to an intelligent matching short video product recommendation method and management system. Background Art

[0002] Short video sales refers to a form of product display and sales through short video platforms. In short video sales, anchors or internet celebrities show the usage effects, features and advantages of products through short videos or live broadcasts, and guide users to buy. Short video sales have become a new promotion method in the e-commerce industry, and are favored by more and more merchants and consumers. Through short video sales, merchants can quickly expand product exposure and increase sales, and users can also learn about product information and get shopping inspiration through videos, thereby improving the convenience and fun of shopping.

[0003] Smart matching recommendations can present products that better meet user needs to users, increase users' purchasing intention and conversion rate, and help improve sales and merchants' profitability. Smart matching recommendations can reduce users' time and energy spent on browsing various product information and improve users' shopping efficiency.

[0004] At present, short video sales are usually carried out by recommending live broadcast sales to users. Short video sales also include video sales. Video sales can also be divided into many types, such as live broadcast slicing sales, product evaluation sales, etc. Since live broadcast sales span a long time and often include unnecessary processes, for users who do not have time or do not like to watch live broadcasts, we should cater to their preferences and recommend sales videos of the type that users like, so as to improve the user experience, and improve the sales effect by improving the user experience, so that both users and merchants benefit. Therefore, the present invention is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent matching short video product recommendation method and management system to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above purpose, the present invention provides the following technical solution: a smart matching short video product recommendation method, comprising: Information collection: obtaining user information, including browsing information, like units, previously purchased product information and basic user information. Browsing information includes browsing units and browsing completion, obtaining product promotion video libraries and product information libraries; Demand acquisition: Based on browsing information, multi-level user preference types and reference browsing units are obtained through judgment methods. Based on like units and reference browsing units, user potential needs are obtained through information extraction methods. The parts of user potential needs that are the same as previously purchased product information are eliminated to obtain demand information. Based on previously purchased product information, derived product information is obtained through derivation methods. Demand information and derived product information are integrated to obtain a recommended information set. Video acquisition: traverse the product video library through the recommendation information set to obtain the relevant product video set, and extract the product videos in the relevant product video set that are consistent with the multi-level user preference types to obtain a multi-level recommendation set; Auxiliary information generation: Auxiliary information is obtained through information generation method based on the recommended information set, and the auxiliary information is embedded in the corresponding product promotion video in the multi-level recommendation set to obtain a complete recommendation set; Video recommendation: preset the video recommendation frequency, recommend product-promoting videos based on the user's basic information through the recommendation frequency and the complete recommendation set, obtain the user's behavior information on watching product-promoting videos, and optimize the video recommendation frequency through adjustment methods based on the behavior information; The judgment method includes: presetting a time limit, classifying browsing information based on the time limit to obtain past browsing information and real-time browsing information, the real-time browsing information includes real-time browsing units and their real-time browsing completion, presetting a threshold browsing completion, eliminating real-time browsing units whose real-time browsing completion is less than the threshold browsing completion to obtain reference browsing units, classifying several reference browsing units based on the types of the reference browsing units to obtain several control sets, obtaining user preference types based on the types of the reference browsing units in the control sets, and classifying the control sets through a classification method to obtain multi-level user preference types.

[0007] Furthermore, the classification method includes: presetting a graded numerical interval, comparing the number of reference browsing units in several control sets with the graded numerical interval to obtain correlation, obtaining multi-level user preference types based on the correlation and the type of reference browsing units in the control set, feedback information of the correlation is that when the number of reference browsing units in the control set is greater than the graded numerical interval, the type of the control set is a first-level user preference type, feedback information of the correlation is that when the number of reference browsing units in the control set is within the graded numerical interval, the type of the control set is a second-level user preference type, feedback information of the correlation is that when the number of reference browsing units in the control set is less than the graded numerical interval, the type of the control set is a third-level user preference type, and integrating the first-level user preference type, the second-level user preference type and the third-level user preference type in sequence to obtain a multi-level user preference type.

[0008] Furthermore, the information extraction method includes: constructing a user portrait based on like units and reference browsing units, obtaining the text information and image information in the like units and reference browsing units to obtain a data set, extracting product information in the data set, the product information is the user's potential demand, constructing future user portraits based on subsequent like units and subsequent reference browsing units, and processing the future user portraits through a comparison method to obtain the user's potential demand.

[0009] Furthermore, the comparison method includes: merging user portraits to obtain a user portrait set, merging product information corresponding to the user portraits to obtain a product information set, establishing a correlation relationship between the user portrait set and the product information set, presetting a threshold overlap, judging the overlap between future user portraits and the user portrait set, comparing the threshold overlap with the overlap to obtain correlation, when the feedback information of the correlation is correlated, extracting the product information in the product information set based on the correlation relationship to obtain the user's potential demand, when the feedback information of the correlation is uncorrelated, obtaining the text information and image information in the future like units and the future reference browsing units to obtain the future data set, extracting the future product information in the future data set to obtain the user's potential demand, and correspondingly integrating the future user portraits and the future product information into the user portrait set and the product information set.

[0010] Furthermore, the derivation method includes: extracting product features in a product information library to obtain a feature library, splitting previously purchased product information to obtain a number of products, extracting product features to obtain feature points, combining feature points to obtain feature combinations, integrating feature combinations and feature points to obtain feature sets, matching features associated with the feature set in the feature library to obtain a number of target features, performing data cleaning on the several target features to obtain a target feature set, and extracting product information from the product information library based on the target feature set to obtain derived product information.

[0011] Furthermore, the information generation method includes: splitting the recommendation information set to obtain product information, searching the evaluation of the product information online to obtain product evaluation information, screening the product evaluation information to obtain high-quality positive review information and high-quality negative review information, and integrating the high-quality positive review information and high-quality negative review information to obtain auxiliary information.

[0012] Furthermore, the adjustment method includes: judging whether the user likes the recommended product promotion video based on the user's behavior information, the behavior information including browsing completion and like behavior, when the browsing completion exceeds a threshold browsing completion and there is like behavior in the behavior information, the video recommendation frequency is increased, when the browsing completion does not exceed the threshold browsing completion and there is no like behavior in the behavior information, the video recommendation frequency is reduced, and when the behavior information is other situations, the video recommendation frequency is not changed.

[0013] An intelligent matching short video product recommendation management system uses the above-mentioned intelligent matching short video product recommendation method.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The intelligent matching short video recommendation method and management system can obtain a recommended information set through the set information extraction method, derivation method and previous purchase information of goods, and obtain a related set of videos with goods through the combination of the recommended information set and the video library with goods, and can determine what type of video the user likes based on the user's latest real-time browsing information through the set judgment method, and can obtain videos with goods of the type that the user likes based on the user's preference type and the related video collection with goods, and by recommending videos with goods of the type that the user likes to the user, the transaction efficiency of goods can be improved while improving the user experience, thereby benefiting both the user and the merchant.

[0015] At the same time, in the information extraction method, the user's like unit and reference browsing unit are used to construct a user portrait, product information is obtained based on the like unit and reference browsing unit, and the user portrait is recorded to obtain a user portrait set and a product information set. The user portrait set and the product information set will be continuously enriched during the information extraction process, so that when the user's potential needs are subsequently obtained, it is only necessary to build a user portrait based on the like unit and the reference browsing unit and compare it with the user portrait set, and then the corresponding user potential needs can be obtained in the product information set through correlation, which is conducive to improving work efficiency and reducing occupancy. Through the setting of the derivative method, the product information related to the user's previously purchased product information can be analyzed based on the user's previously purchased product information, which is conducive to enriching the information in the recommendation information set, facilitating meeting user needs and improving user experience.

[0016] At the same time, by setting the information generation method, auxiliary information of the product information, that is, positive review information and negative review information, can be obtained according to the user's recommended information set. By allowing users to browse the auxiliary information, users can further understand the product, which is beneficial to improving the user experience. By setting the adjustment method, the frequency of video recommendations can be adjusted according to the behavioral information fed back by users, thereby increasing the time and stickiness of users staying on the platform, which is beneficial to improving the user activity and user retention rate of the platform, and avoiding the loss of users from the platform due to a large number of recommendations of sales videos. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the overall process structure of the upper half of the present invention; Figure 2 It is a schematic diagram of the overall process structure of the lower half of the present invention; Figure 3 It is a schematic diagram of the structure of the derivation method of the present invention; Figure 4It is a schematic diagram of the classification method structure of the present invention. DETAILED DESCRIPTION

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

[0019] Through intelligent matching recommendations, users can get more personalized shopping recommendations on the short video platform, increase the time and stickiness of users on the platform, and improve the user activity and retention rate of the platform. Intelligent matching short video recommendations can improve user experience, increase sales conversion rate, save user time and energy, and improve platform stickiness, which has positive significance for both merchants and users.

[0020] like Figure 1-Figure 4 As shown, the present invention provides a technical solution: an intelligent matching short video product recommendation method and management system, the method comprising: Information collection: obtaining user information, including browsing information, like units, previously purchased product information and basic user information. Browsing information includes browsing units and browsing completion, obtaining product promotion video libraries and product information libraries; Demand acquisition: Based on browsing information, multi-level user preference types and reference browsing units are obtained through judgment methods. Based on like units and reference browsing units, user potential needs are obtained through information extraction methods. The parts of user potential needs that are the same as previously purchased product information are eliminated to obtain demand information. Based on previously purchased product information, derived product information is obtained through derivation methods. Demand information and derived product information are integrated to obtain a recommended information set. Video acquisition: traverse the product video library through the recommendation information set to obtain the relevant product video set, and extract the product videos in the relevant product video set that are consistent with the multi-level user preference types to obtain a multi-level recommendation set; Auxiliary information generation: Auxiliary information is obtained through information generation method based on the recommended information set, and the auxiliary information is embedded in the corresponding product promotion video in the multi-level recommendation set to obtain a complete recommendation set; Video recommendation: preset the video recommendation frequency, recommend product-promoting videos based on the user's basic information through the recommendation frequency and the complete recommendation set, obtain the user's behavior information on watching product-promoting videos, and optimize the video recommendation frequency through adjustment methods based on the behavior information; It should be noted that in the information collection stage, user information is obtained through the platform. When obtaining information, the privacy of users should be protected to avoid leakage of user information. The video library and product information library are also obtained through the platform. In the demand acquisition stage, multi-level user preference types are obtained by browsing information and judging methods. The video information of the like unit is also in the browsing information. By eliminating the same part as the previously purchased product information in the user's potential needs, the appearance of the video of the purchased product can be avoided, which is conducive to improving the accuracy of the recommendation of the video. By setting the derivative method, more relevant product information is obtained according to the previously purchased product information to obtain the derived product information, which is conducive to increasing the richness of the recommended information set. In the video acquisition stage, the information in the recommended information set is all Product information, traverse the product video library according to the product information, that is, search for the product video corresponding to the product information in the product video library, and collect all the product videos together to obtain a related product video set. The traversal process can be completed by training a dedicated model. Here, the function of the trained dedicated model is to input product information and output product videos. In order to improve the accuracy of the traversal to obtain product videos, manual review of the results output by the dedicated model can be used to extract the product videos in the related product video set that are consistent with the multi-level user preference types to obtain a multi-level recommendation set. This process can also be carried out by training a dedicated model in combination with manual review. The user preference type is the type of video, for example: popular science, funny, emotional, etc. The video type in the related product video set can be obtained through the platform , the sorting order of the sales videos in the multi-level recommendation set is consistent with the multi-level user preference types. In the auxiliary information generation stage, auxiliary information is generated through the information generation method to facilitate users to understand the detailed information of the products and improve the user's usage experience. In the video recommendation process, the sales video is recommended to the user through the user's basic information and the recommendation frequency. When the sales video appears, the auxiliary information corresponding to the product in the sales video is output to the user in the form of documents, pictures, etc. after the sales video is played, so that the user can further understand the information of the product in the sales video. The method provided in this application is embedded in a short video sales platform. The specific short video sales platform is selected according to actual usage needs. This application obtains the user's browsing information, like unit, and past purchase When accessing product information and basic user information, it is obtained according to the short video shopping platform used for access. In the database of the short video shopping platform, the user is determined according to the user's registered mobile phone number, WeChat account, QQ number and other basic information, and the user's browsing information, like unit, previous purchase product information and user basic information are obtained. How to protect the user's privacy in this process? Since the method provided in this application is embedded in the short video shopping platform, the process of implementing the method is to perform data transmission and other operations on the short video shopping platform. Since the data transmission and other operations of the short video shopping platform itself are encrypted, the user's privacy will be protected when obtaining the user's data and transmitting the data. At the same time, the data encryption method can also be customized to encrypt the data to protect the user's privacy.Common encryption methods include symmetric encryption algorithms and asymmetric encryption algorithms. The encryption method used generates a secret key to encrypt and decrypt the transmitted data to protect the privacy of the user. The information extraction method specifically uses the visual recognition technology in the prior art to identify the text information and image information in the like unit and the reference browsing unit to obtain a data set, extract the product information in the data set, and obtain the user's potential needs based on the product information. The specific process of visual recognition technology is: collect complete information including images and texts through network crawling (using crawler technology) or manual collection of relevant data, manually annotate the collected images, identify the like unit, reference browsing unit and related text area, or use annotation tools (such as LabelImg, VGG Image Annotator, etc.) to create annotated datasets, adjust the size of the image to match the input requirements of the recognition model, such as the commonly used 224x224 pixels, perform image denoising, enhancement, rotation correction and other processing to ensure the best image quality, segment the text area in the image, extract the text part that needs to be recognized, standardize the text content, including removing special characters, converting to lowercase, etc., use pre-trained convolutional neural network (CNN) models, such as ResNet, VGG, MobileNet, etc., to extract image features, select appropriate OCR tools (such as Tesseract, EasyOCR, etc.) to recognize text information in the image, input the processed image into the OCR tool, recognize the text information therein, and output the recognition result, clean the recognized text and image data, remove duplicate, erroneous or useless data to obtain the dataset, and after obtaining the dataset, verify the extracted text and image information through manual review or using the verification dataset to ensure the accuracy of the data, extract the product information in the dataset. A process, by obtaining the names and keywords of the goods sold in the short video platform to obtain a comparison set, combining the data set comparison set, obtaining the same parts of the two to obtain product information, eliminating previous purchase information from the product information, that is, obtaining demand information, the demand information is the names or keywords of several goods, and the product information is derived by a derivative method and combined with the original product information to obtain a recommendation information set, the auxiliary information is to split the recommendation information set to obtain several product information, and retrieve the product information in the platform used to obtain the high-quality positive review information and high-quality negative review information of the product information, integrate the high-quality positive review information and high-quality negative review information to obtain auxiliary information, the complete recommendation set is the evaluation information of the short video with goods to be recommended and the goods with goods in the short video with goods, the multi-level user preference type is obtained by judging the type of the user's favorite video, that is, the likes and browsing time, and classifying the type of the user's favorite video, the recommendation frequency and time limit are formulated according to the actual usage, the recommendation frequency can be defaulted to recommending 3 short videos with goods in 1 hour, and the time limit is a time node, which can be specifically half a year before the real time,Specialized models refer to deep learning models, such as convolutional neural networks, etc. The correlation is obtained by comparing the number of reference browsing units in several control sets with the graded numerical intervals, that is, the quantitative relationship.

[0021] The judgment method includes: presetting a time limit, classifying browsing information based on the time limit to obtain past browsing information and real-time browsing information, the real-time browsing information includes real-time browsing units and their real-time browsing completion, presetting a threshold browsing completion, eliminating real-time browsing units whose real-time browsing completion is less than the threshold browsing completion to obtain reference browsing units, classifying several reference browsing units based on the types of the reference browsing units to obtain several control sets, obtaining user preference types based on the types of the reference browsing units in the control sets, and classifying the control sets through a classification method to obtain multi-level user preference types.

[0022] It should be noted that the user's preference types will change at different times. The time limit is determined according to the actual situation. The closer the time limit is to the real time, the better, while ensuring that there are sufficient browsing information samples. The threshold browsing completion is the completion of the user's viewing of this browsing unit, that is, the progress of the user in viewing this unit. The browsing unit is a video. The threshold browsing completion is determined according to the actual user usage. Under normal circumstances, the threshold browsing completion can be set to 80%. The type of reference browsing unit can be obtained through the video tag of the reference browsing unit on the platform. By setting the judgment method, it determines what type of video the user likes based on the user's latest real-time browsing information, avoiding information asymmetry and recommending videos of the type that the user liked a long time ago, which is conducive to improving the user's usage experience. At the same time, it classifies the user's preference types. When recommending videos, it can give priority to recommending the best user-preferred type of goods-carrying videos, which is conducive to improving the transaction efficiency of goods when recommending videos, and also to improving the user's usage experience.

[0023] like Figure 4 As shown, the classification method includes: presetting a graded numerical interval, comparing the number of reference browsing units in several control sets with the graded numerical interval to obtain correlation, obtaining a multi-level user preference type based on the correlation and the type of the reference browsing units in the control set, the feedback information of the correlation is that when the number of reference browsing units in the control set is greater than the graded numerical interval, the type of the control set is a first-level user preference type, the feedback information of the correlation is that when the number of reference browsing units in the control set is within the graded numerical interval, the type of the control set is a second-level user preference type, the feedback information of the correlation is that when the number of reference browsing units in the control set is less than the graded numerical interval, the type of the control set is a third-level user preference type, and the first-level user preference type, the second-level user preference type and the third-level user preference type are sequentially integrated to obtain a multi-level user preference type.

[0024] It should be noted that the grading numerical interval is determined according to the actual situation. The meaning expressed by the grading numerical interval is the information range of the number of reference browsing units in the control set. In the multi-level user preference types, the preference types are sorted into primary user preference types, secondary user preference types, and tertiary user preference types. For example, the grading numerical interval is 4-6. At this time, if the number of reference browsing units in the control set exceeds 6, the type of the control set is the primary user preference type. If the number of reference browsing units in the control set is 4 or 5 or 6, the type of the control set is the secondary user preference type. If the number of reference browsing units in the control set is less than 4, the type of the control set is the tertiary user preference type. The user preference types are graded through the classification method, which is convenient for subsequent video recommendations to recommend goods-carrying videos according to the order, so as to improve the user experience.

[0025] like Figure 1 As shown, the information extraction method includes: constructing a user portrait based on like units and reference browsing units, obtaining the text information and image information in the like units and reference browsing units to obtain a data set, extracting the product information in the data set, the product information is the user's potential demand, constructing a subsequent user portrait based on subsequent like units and subsequent reference browsing units, and processing the subsequent user portrait by a comparison method to obtain the user's potential demand; the comparison method includes: merging user portraits to obtain a user portrait set, merging the product information corresponding to the user portrait to obtain a product information set, and establishing an association relationship between the user portrait set and the product information set. , preset a threshold overlap, determine the overlap between future user portraits and the user portrait set, compare the threshold overlap with the overlap to get the correlation, when the feedback information of the correlation is correlated, then extract the product information in the product information set based on the correlation relationship to get the user's potential demand, when the feedback information of the correlation is uncorrelated, then obtain the text information and image information in the future likes unit and the future reference browsing unit to get the future data set, extract the future product information in the future data set to get the user's potential demand, and integrate the future user portrait and the future product information into the user portrait set and the product information set accordingly.

[0026] It should be noted that the copy information and image information in the like units and reference browsing units can be obtained through the information acquisition technology and visual image technology in the existing technology. The product information in the extracted data set can be extracted by training a professional extraction model. The reference effect is achieved by constructing user portraits. The subsequent like units and subsequent reference browsing units refer to the like units and reference browsing units that are processed later. The threshold overlap is determined according to the actual situation. For example: the first time the user portrait is constructed and the product information is obtained, the first user portrait is a user portrait set, and the first product information is a product information set. When the user portrait is constructed for the second time, the user portrait constructed for the second time is compared with the user portrait in the user portrait set, and the overlap and threshold overlap are judged to obtain the correlation. When the overlap exceeds the threshold overlap, the feedback information representing the correlation is related. The user portrait constructed for the second time uses the product information of the user portrait constructed for the first time. When the threshold overlap is not reached, the feedback information representing the correlation is unrelated. The text information and image information in the like unit and the reference browsing unit are obtained to obtain a data set, and the product information in the data set is extracted. The user portrait constructed for the second time is integrated into the user portrait set, and the product information of the second user portrait is integrated into the product information set. When the user portrait is constructed for the third time, and so on, through the setting of the information extraction method, the user portrait set and the product information set will be continuously enriched in the working process, so that when the user's potential needs are obtained in the subsequent process, it is only necessary to build the user portrait based on the like unit and the reference browsing unit and compare it with the user portrait set, and then the corresponding user's potential needs can be obtained in the product information set through the correlation, which is conducive to improving work efficiency and reducing occupancy.

[0027] like Figure 2 and Figure 3 As shown, the derivation method includes: extracting product features in a product information library to obtain a feature library, splitting previously purchased product information to obtain a number of products, extracting product features to obtain feature points, combining feature points to obtain feature combinations, integrating feature combinations and feature points to obtain feature sets, matching features associated with the feature sets in the feature library to obtain a number of target features, processing a number of target features by a data processing method to obtain a target feature set, and extracting product information from the product information library based on the target feature set to obtain derived product information.

[0028] It should be noted that the features of the product can be extracted by training a professional model. The feature of the product is the name of the product. The process of combining feature points to obtain feature combinations is as follows: there are three feature points, namely 1, 2, and 3. The combined feature points are 1 and 2, 2 and 3, 1 and 3, and 1, 2 and 3. The process of matching the feature set with the features in the feature library is to select the feature combination or feature points in the feature set in turn to traverse the feature library to obtain the associated target features. The matching process can be performed by training a professional model, and the effect achieved is as follows: for example, the feature combination is a dishwasher and tableware. The target features obtained after traversing the feature library are detergents, chopsticks, etc. associated with the dishwasher and tableware. When the feature combination cannot match the result in the feature library, the next feature combination or feature point matching is performed in turn. The target feature is cleaned by a data processing method to ensure the quality of the target feature. By setting the derivative method, the product information related to the previously purchased product information can be obtained based on the user's previously purchased product information, which is conducive to enriching the information of the recommendation information set, meeting user needs and improving user experience.

[0029] like Figure 2 As shown, the information generation method includes: splitting the recommendation information set to obtain product information, searching the evaluation of the product information online to obtain product evaluation information, screening the product evaluation information to obtain high-quality positive review information and high-quality negative review information, and integrating the high-quality positive review information and high-quality negative review information to obtain auxiliary information.

[0030] It should be noted that, in the process of obtaining product evaluation information by retrieving product information online, the product evaluation information can be obtained by manually retrieving product information on shopping platforms or search engines. Screening the good and bad reviews in the evaluation information is an existing technology. The good and bad reviews in the product evaluation information can be screened according to the word count standard to obtain high-quality good and bad reviews. The screening process can be performed by training professional models and manually by staff. The review process can adopt sampling review to improve efficiency, and can also adopt a deep learning model specially trained to review the output results to improve efficiency. By setting the information generation method, the good and bad review information of the product information can be obtained, which is convenient for users to browse and helps to improve the user experience.

[0031] like Figure 2 As shown, the adjustment method includes: judging whether the user likes the recommended sales video based on the user's behavior information, the behavior information includes browsing completion and like behavior, when the browsing completion exceeds the threshold browsing completion and there is like behavior in the behavior information, the video recommendation frequency is increased, when the browsing completion does not exceed the threshold browsing completion and there is no like behavior in the behavior information, the video recommendation frequency is reduced, and when the behavior information is other situations, the video recommendation frequency is not changed.

[0032] It should be noted that by setting the adjustment method, the frequency of video recommendations can be adjusted according to the behavioral information fed back by users, thereby increasing the time and stickiness of users on the platform, which is conducive to improving the user activity and user retention rate of the platform, and avoiding the loss of users from the platform due to a large number of product-promoting video recommendations.

[0033] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is limited by the attached embodiments and their equivalents.

Claims

1. An intelligent matching short video product recommendation method, characterized by: The method comprises: Information collection: obtaining user information, including browsing information, like units, previously purchased product information and basic user information. Browsing information includes browsing units and browsing completion, obtaining product promotion video libraries and product information libraries; Demand acquisition: Based on browsing information, multi-level user preference types and reference browsing units are obtained through judgment methods. Based on like units and reference browsing units, user potential needs are obtained through information extraction methods. The parts of user potential needs that are the same as previously purchased product information are eliminated to obtain demand information. Based on previously purchased product information, derived product information is obtained through derivation methods. Demand information and derived product information are integrated to obtain a recommended information set. Video acquisition: traverse the product video library through the recommendation information set to obtain the relevant product video set, and extract the product videos in the relevant product video set that are consistent with the multi-level user preference types to obtain a multi-level recommendation set; Auxiliary information generation: Auxiliary information is obtained through information generation method based on the recommended information set, and the auxiliary information is embedded in the corresponding product promotion video in the multi-level recommendation set to obtain a complete recommendation set; Video recommendation: preset the video recommendation frequency, recommend product-promoting videos based on the user's basic information through the recommendation frequency and the complete recommendation set, obtain the user's behavior information on watching product-promoting videos, and optimize the video recommendation frequency through adjustment methods based on the behavior information; The judgment method includes: presetting a time limit, classifying browsing information based on the time limit to obtain past browsing information and real-time browsing information, the real-time browsing information includes real-time browsing units and their real-time browsing completion, presetting a threshold browsing completion, eliminating real-time browsing units whose real-time browsing completion is less than the threshold browsing completion to obtain reference browsing units, classifying several reference browsing units based on the types of the reference browsing units to obtain several control sets, obtaining user preference types based on the types of the reference browsing units in the control sets, and classifying the control sets through a classification method to obtain multi-level user preference types.

2. According to claim 1, a smart matching short video recommendation method is characterized by: The classification method includes: presetting a graded numerical interval, comparing the number of reference browsing units in a plurality of control sets with the graded numerical interval to obtain correlation, obtaining a multi-level user preference type based on the correlation and the type of the reference browsing unit in the control set, feedback information of the correlation is that when the number of reference browsing units in the control set is greater than the graded numerical interval, the type of the control set is a primary user preference type, feedback information of the correlation is that when the number of reference browsing units in the control set is within the graded numerical interval, the type of the control set is a secondary user preference type, feedback information of the correlation is that when the number of reference browsing units in the control set is less than the graded numerical interval, the type of the control set is a tertiary user preference type, and sequentially integrating the primary user preference type, the secondary user preference type and the tertiary user preference type to obtain a multi-level user preference type.

3. According to claim 1, a smart matching short video recommendation method is characterized by: The information extraction method includes: constructing a user portrait based on like units and reference browsing units, obtaining text information and image information in the like units and reference browsing units to obtain a data set, extracting product information in the data set, the product information is the user's potential demand, constructing a future user portrait based on subsequent like units and subsequent reference browsing units, and processing the future user portrait by a comparison method to obtain the user's potential demand.

4. According to claim 3, a smart matching short video recommendation method is characterized by: The comparison method includes: merging user portraits to obtain a user portrait set, merging product information corresponding to the user portraits to obtain a product information set, establishing an association relationship between the user portrait set and the product information set, presetting a threshold overlap, judging the overlap between future user portraits and the user portrait set, comparing the threshold overlap with the overlap to obtain correlation, when the feedback information of the correlation is that there is correlation, then extracting the product information in the product information set based on the correlation relationship to obtain the user's potential demand, when the feedback information of the correlation is that there is no correlation, then obtaining the text information and image information in the future like units and the future reference browsing units to obtain the future data set, extracting the future product information in the future data set to obtain the user's potential demand, and correspondingly integrating the future user portraits and the future product information into the user portrait set and the product information set.

5. According to claim 1, a smart matching short video recommendation method is characterized by: The derivation method includes: extracting commodity features in a commodity information library to obtain a feature library, splitting previously purchased commodity information to obtain a plurality of commodities, extracting commodity features to obtain feature points, combining feature points to obtain feature combinations, integrating feature combinations and feature points to obtain feature sets, matching features associated with the feature sets in the feature library to obtain a plurality of target features, performing data cleaning on the plurality of target features to obtain a target feature set, and extracting commodity information in the commodity information library based on the target feature set to obtain derived commodity information.

6. According to claim 1, a smart matching short video recommendation method is characterized by: The information generation method includes: splitting the recommendation information set to obtain product information, searching the evaluation of the product information online to obtain product evaluation information, screening the product evaluation information to obtain high-quality positive evaluation information and high-quality negative evaluation information, and integrating the high-quality positive evaluation information and high-quality negative evaluation information to obtain auxiliary information.

7. According to claim 1, a smart matching short video recommendation method is characterized by: The adjustment method includes: judging whether the user likes the recommended sales video based on the user's behavior information, the behavior information including browsing completion and like behavior, when the browsing completion exceeds a threshold browsing completion and there is like behavior in the behavior information, the video recommendation frequency is increased, when the browsing completion does not exceed the threshold browsing completion and there is no like behavior in the behavior information, the video recommendation frequency is reduced, and when the behavior information is other situations, the video recommendation frequency is not changed.

8. An intelligent matching short video product recommendation management system, characterized by: An intelligent matching short video product recommendation method as described in any one of claims 1-7 is used.

Images