User analysis method and device based on large model, equipment and medium

Abstract

The invention discloses a user analysis method and device based on a large model, equipment and a medium, and the method comprises the steps: obtaining user data from a plurality of data sources, and carrying out the preprocessing of the user data, so as to form a structured data set; inputting the structured data set into a pre-trained large model based on deep learning, and outputting a current user portrait after performing feature extraction on the user data through the large model; according to the user portrait, analyzing the user by adopting clustering analysis, association rule mining or a classification algorithm, and outputting a user classification result; and according to the user group characteristics corresponding to the user classification result, in combination with the user portrait, performing user behavior prediction so as to push personalized recommendation content to the user. According to the method, the users are analyzed and classified through the user portraits, and the pushing requirements of personalized contents can be accurately and quickly met.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a user analysis method, device, equipment and medium based on a large model. Background Art

[0002] In the era of digital content push, understanding users can adaptively and dynamically launch digital content with different needs for users. User data is the core asset of digital push. Analyzing user data can help us better understand user needs, behaviors and preferences, thereby formulating more effective content push strategies.

[0003] User data includes user portraits. As a labeled description of users, user data is an effective tool that can deeply analyze user needs, preferences, behaviors and values, thereby improving the effect and efficiency of content push. However, with the increase in the amount and complexity of user data, the traditional method of simply labeling user portraits with tags can no longer meet the needs of accurate content push. Summary of the invention

[0004] The embodiments of the present invention provide a user analysis method, device, equipment and medium based on a large model to solve the problem that the traditional method of simply marking user portraits with tags can no longer meet the demand for accurate content push.

[0005] A user analysis method based on a large model, comprising: Obtain user data from multiple data sources, including user behavior data, user attribute data, user social data, and consumption habit data; Preprocess user data, including data cleaning, data standardization, and data normalization, to form a structured data set; Input the structured data set into the pre-trained deep learning-based big model, which is used to extract features from the user data through the big model and output the current user profile; According to the user portrait, cluster analysis, association rule mining or classification algorithm is used to analyze the user and output the user classification results; Based on the user group characteristics corresponding to the user classification results and combined with the user portrait, user behavior prediction is performed to push personalized recommended content to users.

[0006] A user analysis device based on a large model, comprising: The data acquisition module is used to obtain user data from multiple data sources, including user behavior data, user attribute data, user social data and consumption habit data; The data preprocessing module is used to preprocess user data, including data cleaning, data standardization, and data normalization, to form a structured data set; The user portrait output module is used to input the structured data set into the pre-trained large model based on deep learning, and to output the current user portrait after extracting features from the user data through the large model; The classification result output module is used to analyze users based on user portraits using cluster analysis, association rule mining or classification algorithms, and output user classification results; The recommended content push module is used to predict user behavior based on the user group characteristics corresponding to the user classification results and combined with the user portrait, so as to push personalized recommended content to users.

[0007] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned large model-based user analysis method when executing the computer program.

[0008] A computer-readable medium stores a computer program, and when the computer program is executed by a processor, the user analysis method based on a large model is implemented.

[0009] The above-mentioned user analysis method, device, equipment and medium based on a large model obtain user data from multiple data sources, pre-process the user data to form a structured data set; input the structured data set into a pre-trained large model based on deep learning, which is used to extract features from the user data through the large model and output the current user portrait; according to the user portrait, cluster analysis, association rule mining or classification algorithm are used to analyze the user and output the user classification result; according to the user group characteristics corresponding to the user classification result, combined with the user portrait, user behavior prediction is performed, so as to push personalized recommended content to the user. This method is based on large-scale data and deep learning artificial intelligence models to improve the accuracy and real-time performance of user portraits. By analyzing and classifying users through user portraits, the needs of pushing personalized content can be accurately and quickly met. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0011] Figure 1 A schematic diagram of an application environment of a user analysis method based on a large model in one embodiment of the present invention is depicted; Figure 2 A first flow chart of a user analysis method based on a large model in a first embodiment of the present invention is shown; Figure 3 A second flow chart of the user analysis method based on a large model in the second embodiment of the present invention is shown; Figure 4 A third flow chart of a user analysis method based on a large model in a third embodiment of the present invention is shown; Figure 5 A fourth flow chart of a user analysis method based on a large model in a fourth embodiment of the present invention is shown; Figure 6 A fifth flow chart of a user analysis method based on a large model in a fifth embodiment of the present invention is shown; Figure 7 A sixth flow chart of the user analysis method based on a large model in the fifth embodiment of the present invention is shown; Figure 8 A seventh flow chart of the user analysis method based on a large model in the fifth embodiment of the present invention is shown; Fig. 9 A schematic diagram of a user analysis device based on a large model according to an embodiment of the present invention is shown; Fig.10 A schematic diagram of an electronic device according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0012] 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 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.

[0013] The user analysis method based on a large model provided by the embodiment of the present invention can be applied in the following aspects: Figure 1 In the application environment, the user analysis method based on the big model is applied in the user analysis system based on the big model, and the user analysis system based on the big model includes a client and a server, wherein the client communicates with the server through a network. The client, also known as the user end, refers to a program corresponding to the server that provides local services to the client. Furthermore, the client is a computer-side program, an APP program for a smart device, or a third-party applet embedded in other APPs. The client can be installed on, but not limited to, various electronic devices such as personal computers, laptops, smart phones, tablet computers, and portable wearable devices. The server can be implemented with an independent server or a server cluster consisting of multiple servers.

[0014] In one embodiment, if Figure 2 As shown, a user analysis method based on a large model is provided, and the method is applied in Figure 1The server in the example is used as an example, and the specific steps include the following: S10. Obtain user data from multiple data sources, including user behavior data, user attribute data, user social data, and consumption habit data.

[0015] Specifically, user behavior data refers to a series of operation records generated by users in the process of using products or services, including but not limited to user click behavior, purchase behavior, browsing behavior and other behavior logs. These records can reflect how users interact with products, including browsing content, clicking links, searching for keywords, and staying on the page. For example, on video websites, behavior data is reflected in the record of users watching videos. For example, a user quits after watching the first 30 minutes of a movie, or repeatedly watches a short video, as well as the user's likes, comments, and sharing of the video.

[0016] User attribute data mainly describes the basic characteristics of users, which are relatively stable and do not change with a single user behavior, including but not limited to the user's age, gender, occupation, and educational background. Specifically, it includes demographic information, such as age, gender, geographic location, and device information, such as device model, operating system, etc. For example, the user attribute data of a social platform may show that the user is 25 years old, female, lives in XX city, uses a certain model of mobile phone, and uses a certain operating system.

[0017] User social data refers to data about the user's social behavior and social relationships on social networks or platforms with social functions, including the user's friend relationships, follow lists, number of fans, and interactive information, such as likes, comments, private messages, etc. For example, on professional social platforms, users' social data includes interpersonal network, such as the user's connection with peers, former colleagues, and partners, as well as interactive content within the interpersonal circle, such as sharing and commenting on industry trends.

[0018] Consumption habit data mainly refers to the user's preferences and behavior patterns in the process of purchasing products or services, including purchase frequency, purchase time, purchase channel, consumption amount, brand preference, etc. For example, in the catering industry, consumption habit data may show that a user consumes in restaurants 3-4 times a week, usually at weekend dinner time, with an average consumption of 100-150 yuan per person each time, prefers Chinese cuisine, and often places orders through mobile food delivery platforms.

[0019] S20. Preprocess the user data, including data cleaning, data standardization, and data normalization, to form a structured data set.

[0020] Specifically, this embodiment can clean, denoise, remove duplicates, process outliers, and standardize the collected data, remove invalid data and duplicate data, and normalize the data and convert it into a format suitable for large model input to form a structured data set, such as: converting text data into word vectors, normalizing numerical data, and extracting features and encoding image data. Among them, data normalization is a data preprocessing technology that converts the feature values ​​of data into a specific interval, usually [0,1] or [-1,1]. Data normalization can eliminate the influence of different features due to factors such as dimension and order of magnitude, so that each feature plays a relatively balanced role in the clustering process, thereby improving the accuracy of clustering; the normalized data can more intuitively show the relative importance of each indicator in the overall evaluation.

[0021] S30. Input the structured data set into a pre-trained large model based on deep learning, so as to extract features from the user data through the large model and output the current user portrait.

[0022] Specifically, this embodiment can input the pre-processed user data into a pre-trained large model, wherein the large model can be a neural large model based on deep learning, including but not limited to convolutional neural network, recurrent neural network, Transformer model, decision tree model and random forest model, etc., and optimize the parameters of the model during the training process to improve the model's learning ability for user features. At the same time, in the model construction and training steps, this embodiment can use a distributed training architecture to train the large model, using multiple computing nodes to process training tasks in parallel, so as to shorten the training time and improve the training efficiency.

[0023] The large model uses the attention mechanism to extract features from user data and focus on key information in the user data, which can improve the accuracy and effectiveness of feature extraction. For example, feature extraction based on text content, feature extraction based on time series data, and feature extraction based on image data can obtain user portraits. User portraits include multi-dimensional features of users: a comprehensive description of user interests, consumption habits, social circles, behavioral pattern characteristics, and other information.

[0024] S40. Based on the user portrait, use cluster analysis, association rule mining or classification algorithm to analyze the user and output the user classification result.

[0025] Specifically, the user classification results are used to make personalized recommendations or predict user behavior based on the user group classification results. The clustering algorithm is used to group users according to similarities, such as interests, hobbies, consumption habits, etc. The clustering algorithm can use a density-based clustering algorithm to divide users into different clusters according to the density distribution of user features, so as to find high-density areas and abnormal points in the user group.

[0026] The association rule mining algorithm is used to discover the associations between different items in the data set. It uses the Apriori algorithm or FP-Growth algorithm to mine the associations between user features, thereby discovering the potential connections between different behaviors or attributes of users, and providing support for precision marketing. It is mainly used to analyze frequent patterns in large amounts of data, and to find interesting associations or correlations between items from a large amount of transaction data, shopping basket data, web browsing records, and other data. For example, an association rule such as "70% of customers who buy bread also buy milk" is found in the shopping basket data of the platform.

[0027] Classification algorithms are used to classify users according to their characteristics, for example, dividing users into categories such as high-value users, potential churn users, and new users, so as to develop personalized marketing strategies for different types of users.

[0028] S50. Based on the user group characteristics corresponding to the user classification results and combined with the user portrait, user behavior prediction is performed to push personalized recommendation content to the user. Based on the user's interests, hobbies, consumption habits and other information, user behavior prediction or user needs identification, relevant products or services are recommended to the user.

[0029] Specifically, the recommendation algorithm is based on collaborative filtering, content recommendation or hybrid recommendation mechanism. Furthermore, the recommendation system can be dynamically updated to respond to real-time behavior changes of users.

[0030] Specifically, by analyzing the user's operating behavior in each functional module of the product, we can understand the user's fluency and confusion in using the product. For example, if it is found that many users stay on a certain function page for too short a time or frequently return to the previous page, it may mean that there is a problem with the design or content display of the function and it needs to be optimized; based on user behavior, such as users frequently browsing a certain type of product but not buying it, relevant promotional information or product recommendations can be pushed to increase the possibility of user purchase; user groups with different attributes often have different needs and behavior patterns. For example, based on users classified by age and gender, sports equipment can be promoted to young male user groups, and beauty and skin care products can be recommended to middle-aged female user groups.

[0031] Personalized recommendation content can provide users with adaptive services. For example, for users using mobile devices, the page layout can be optimized to make it more suitable for operation on the mobile screen; through the user's social network, word-of-mouth marketing and viral marketing can be carried out. For example, brands can promote products or services to more potential users to build a social graph; analyzing users' social data to build a social graph and understand the user's position and influence in the social network will help discover potential opinion leaders and potential cooperation opportunities between users.

[0032] This embodiment can push relevant product information at the user's preferred purchase time or recommend brands and products that the user may like through precision marketing and personalized recommendations. For example, at the time when the user often buys skin care products, push new skin care product series and recommend them according to the user's brand preferences; at the same time, by analyzing consumption habit data, identify high-value and high-loyalty users, provide them with exclusive services and discounts, and encourage them to continue consumption; this method analyzes and classifies users through user portraits, and can accurately and quickly meet the needs of pushing personalized content.

[0033] In a specific embodiment, if Figure 3 In step S10, i.e., obtaining user data from multiple data sources, the following steps are specifically included: S210. Track the user's online behavior through the portal (website and application) and record the user's online trajectory data.

[0034] Specifically, in terms of websites, web analytics tools (such as Google Analytics, etc.) can be used to embed specific tracking scripts in the web page code to record information such as the URL of the page visited by the user, the time spent on each page, the scroll depth of the page, the links clicked, etc. For example, for a news information website, it is possible to know what types of news articles the user browsed, whether they were sports, entertainment or finance, how long they stayed on each article page, whether they clicked on related recommended links, etc.

[0035] On the application side, behavior tracking is achieved with the help of a software development kit (SDK). For example, a shopping app can record the frequency of users opening the app, browsing on each product category page, search keywords, etc. For example, if a user frequently opens the app to view a certain type of electronic products or searches for a specific brand of mobile phone, these can be recorded as online trajectory data.

[0036] S220. Collect users’ social interaction data through social media terminals.

[0037] Specifically, for mainstream social media platforms (such as Weibo, WeChat Moments, Facebook, Twitter, etc.), data can be obtained through the application programming interface (API) provided by the platform. The collected data includes the user's friend network, such as who are the user's frequently interacting friends, which social groups the user has joined, etc.; there are also interactions with the user's posted content, the number of likes, comments, and reposts received for the posted dynamics, as well as user operations such as liking, commenting, and sharing other people's content. For example, if an Internet celebrity posts a promotional video on social media, by collecting interactive data, it can be known how well the video is disseminated and popular among its fan base and a wider range of users.

[0038] S230. Collect the user's e-commerce interaction data through the e-commerce terminal.

[0039] Specifically, e-commerce interaction data may include the following: Historical purchase data: Contains the details of the products purchased by users, such as product name, brand, price, purchase time, purchase quantity, etc. For example, from the e-commerce platform, we can know that a user has purchased clothing five times in the past year, three of which are dresses of a certain brand, with an average purchase price of 200-300 yuan each time, and the purchase time is concentrated in the season change season.

[0040] Collection data: records the items that users have added to their favorites, which can reflect items that users are interested in but may not have purchased at the time. For example, a user may have collected a high-end camera lens, but although they have not yet purchased it, it reflects their potential purchase intention. In the future, relevant discount information can be pushed to promote purchases.

[0041] E-commerce search data: collect keywords entered by users in the search bar of e-commerce platforms to understand the types of products or services users are looking for. For example, if a user searches for "sports shoes for running" many times, it means that he or she has a need to buy running shoes. The system can optimize product display and recommendation strategies based on this type of search data and give priority to recommending running shoes that meet the requirements to the user.

[0042] S240. Obtain device-side usage data through user login behavior.

[0043] Specifically, the core purpose of obtaining device-side usage data through user login behavior is to fully understand how users interact with related systems or platforms on specific devices. Login behavior is often the key starting point for users to start using services. Collecting device-side usage data through this behavior can accurately capture multi-dimensional information such as the operating habits, usage frequency, and usage time of different users on their respective devices. For example, for a mobile application, by recording data such as the mobile phone model, system version, and login time when the user logs in, it is possible to analyze the popularity of the application on different types of devices and different time periods, thereby providing a basis for subsequent product optimization, function iteration, and operational strategy formulation.

[0044] S250. Establish a real-time connection with the data source through the data acquisition interface. When new user data is generated in the data source, transmit the new user data to the data acquisition system in real time for subsequent processing.

[0045] Specifically, a real-time connection is established with the data source through the data collection interface to transmit newly generated user data, aiming to ensure the timeliness and integrity of the data. In many business scenarios, user data is constantly changing and generated. Only by acquiring these new data in real time and participating in subsequent processing can the analysis and decisions made based on these data be based on the latest status. For example, in an e-commerce platform, if the user's real-time browsing, ordering and other new behavior data can be transmitted to the data collection system in a timely manner, the platform can adjust the product recommendation strategy based on these data in real time, accurately push products that the user may be interested in at the moment to the user, and improve the user's shopping convenience and the platform's sales conversion rate.

[0046] In a specific embodiment, if Figure 4 In step S50, i.e., pushing personalized recommended content to the user, the steps specifically include: S310. Based on user behavior prediction and user portrait, select personalized recommendation content from multiple dimensions and generate a dynamic product recommendation list.

[0047] Specifically, this step can accurately create personalized recommendation content that fits users' personal interests, needs and preferences by integrating two key elements: user behavior prediction and user portrait, and generate a dynamic product recommendation list. Personalized recommendations can greatly improve the user experience, help users more efficiently discover products or services that they may be interested in, and also help improve business conversion rates, user retention rates and other indicators. For example, on video platforms, based on the user's past viewing behavior (such as favorite video types, viewing time, etc.) combined with their age, region, interests and hobbies and other portrait information, it is possible to recommend videos that suit the user's taste, making the user more willing to stay on the platform and continue browsing.

[0048] S320. Push corresponding target recommended content to the user based on the dynamic product recommendation list.

[0049] Specifically, this embodiment can push the corresponding target recommendation content to the user based on the generated dynamic product recommendation list, pass the personalized recommendation information screened and prepared in advance to the user, and enhance the interaction and stickiness between the user and the platform. For example, the e-commerce platform will display the recommended products for the user in the recommended position on the homepage or inform the user through message push, guiding the user to view the product details and place an order.

[0050] S330. When pushing target recommended content, user data is captured in real time, and pre-processing of the user data is repeated. The structured data set is input into a pre-trained large model based on deep learning, which is used to extract features from the user data through the large model and then output the current user portrait. According to the user portrait, cluster analysis, association rule mining or classification algorithm is used to analyze the user to obtain a user classification result. According to the user group characteristics corresponding to the user classification result and combined with the user portrait, user behavior prediction is performed, thereby pushing personalized recommended content to the user, so as to update the dynamic product recommendation list in real time.

[0051] Specifically, this embodiment can capture user data in real time to update the dynamic product recommendation list in real time, so that the recommended content can always fit the user's latest status and needs. As users continue to interact with the platform, their interests, preferences, etc. may change at any time. By updating the recommendation list in real time, the accuracy and timeliness of the recommendations can be guaranteed, and high-quality personalized recommendation services can be continuously provided to users, so that users can always get valuable recommendations.

[0052] Real-time capture of user data can utilize data collection interfaces, point-of-use technologies, etc., to timely collect corresponding data when users perform various operations, such as clicking, browsing, searching, etc., and then convert it into a structured data set according to the established data preprocessing process, including data cleaning, normalization, encoding, and other operations, and then input it into the pre-trained deep learning large model. For example, the large model of the Transformer architecture can extract features from the data very well and output an updated user portrait. Cluster analysis is then used to cluster similar users into one category, and the commonalities of the group are analyzed to guide recommendations and association rule mining. The association between different user behaviors and products is found for recommendation, and classification algorithms are used to divide users into different categories. Targeted recommendations and other data analysis methods are used to obtain user classification results, and then combined with the updated user portrait to predict user behavior, and finally the dynamic recommendation list is updated in real time.

[0053] In a specific embodiment, if Figure 5In step S310, i.e., based on the user behavior prediction and the user portrait, personalized recommendation content is selected from multiple dimensions to generate a dynamic product recommendation list, which specifically includes the following steps: S410. Obtain a list of potential user behaviors based on user behavior prediction.

[0054] Specifically, this embodiment can obtain a list of potential user behaviors through user behavior prediction, aiming to dig out a series of behaviors that users may have in the future, dig out potential needs and trends of users in advance, and help the system to be more forward-looking in resource allocation, content preparation, and optimization of personalized services, so as to better meet the needs of users that have not yet been expressed but may appear. For example, on an online reading platform, if it can be predicted that a user may be interested in a certain type of book next, even if the user has not actively searched or browsed related books yet, the corresponding recommended content can be prepared in advance.

[0055] Machine learning and deep learning algorithms are usually used to predict user behavior. For example, historical data on user behavior can be collected, such as browsing history, purchase history, operation duration, etc., and these data can be organized into a time series data format. Then, time series prediction models, such as traditional statistical models such as the ARIMA model, or recurrent neural networks based on deep learning, such as RNN and its improved versions LSTM, GRU, etc., can be used to predict future behaviors and output a series of possible situations of potential behaviors, thereby forming a list of potential user behaviors. In addition, probabilistic graph models such as Bayesian networks can be combined to infer the user's future potential behavior by analyzing the conditional probability relationship between different behaviors. At the same time, in order to improve the accuracy of the prediction, feature engineering can also be performed on the data to extract and filter out key features that have an important impact on behavior prediction, such as the correlation between features such as the user's age, region, and recent activity level and behavior.

[0056] S420. Obtain the behavior ratio of each user's potential behavior based on the user portrait and each user's potential behavior, thereby confirming the content recommendation frequency of each personalized recommended content based on the behavior ratio.

[0057] Specifically, this embodiment can combine the user portrait with each user's potential behavior to obtain the behavior ratio, and then determine the recommendation frequency of each personalized recommendation content, so that the personalized recommendation is more scientific, reasonable and in line with the user's actual situation. Different potential behaviors have different importance and probability of occurrence for different users. By clarifying the behavior ratio, the frequency of pushing each recommended content can be accurately controlled according to the characteristics of each user, avoiding excessive or insufficient push, and improving the effectiveness of the recommendation and user acceptance. For example, for some users who have both shopping and browsing information needs, the appropriate push frequency for shopping recommendations and information recommendations can be determined based on their portraits and potential behavior ratios, so that users can receive recommended content that matches their own interest ratios in a more orderly manner.

[0058] S430. Generate a dynamic product recommendation list based on content recommendation frequency.

[0059] Specifically, this embodiment can generate a dynamic product recommendation list based on the content recommendation frequency, that is, based on the recommendation frequency determined in the previous steps, integrate a dynamic product recommendation list that is updated in a timely manner and meets the personalized needs of users, so as to display suitable and likely to be accepted by users products or services to users at an appropriate frequency, further optimize the efficiency of users in obtaining information, guide users to find more content of interest, and help the system achieve precision marketing and improve business conversion rate. For example, on a music playback platform, a dynamic recommendation list is generated according to the recommendation frequency corresponding to different song types, so that users can hear music recommendations that meet their current preferences and have an appropriate frequency at different time periods.

[0060] This embodiment sorts and groups all the products or services available for recommendation according to their corresponding content recommendation frequencies, and selects appropriate product combinations to form a dynamic recommendation list based on the display rules of the platform, such as how many items can be displayed in the recommendation column of the mobile terminal interface, how to allocate recommended content in different sections of the web page, etc. This embodiment can use a database management system to store and query product information and recommendation frequency data, implement data screening, sorting and integration operations through algorithm code, and dynamically adjust the recommendation list in combination with real-time monitoring of user feedback (for example, user clicks on recommended content, browsing time, whether to collect, and other interactive situations) to adapt to the latest user needs and behavior changes.

[0061] In a specific embodiment, if Figure 6 In step S420, the behavior proportion of each user's potential behavior is obtained according to the user portrait and each user's potential behavior, which specifically includes the following steps: S510. Obtain the advertising participation data in the user portrait, the advertising participation data includes the user's advertising click frequency, advertising conversion rate and advertising participation timing.

[0062] Specifically, the user's ad click frequency refers to the number of times a user clicks on an ad within a certain period of time, which directly reflects the user's attention to the presented ad. For example, if a user clicks on various product ads pushed by an e-commerce platform ten times in a month, while another user only clicks twice, it means that the former is more interested in the ad content and pays more attention to it than the latter.

[0063] A higher click frequency means that users may have potential needs for related products or services, which can help identify which users are more willing to actively learn about the information carried by the advertisement, thereby providing a basis for subsequent precision marketing. For example, users who frequently click on fitness equipment ads should have fitness-related needs, and more fitness-related product ads can be pushed to them in the future.

[0064] The user's ad conversion rate refers to the proportion of users who eventually complete an actual transaction after clicking on an ad. For example, if an electronic product ad is run 100 times, ten users click on the ad, and only two of them complete the purchase, then the ad conversion rate is 20%. A high ad conversion rate means that the ad successfully guides users to complete consumption decisions. A low conversion rate indicates that there may be problems in the presentation of the selling points of the ad, the purchase process, and other links, which need to be optimized.

[0065] The timing of user engagement with ads refers to the specific time points or time periods during a day, week, or specific period when users are more inclined to click and participate in ad interactions. For example, data analysis found that ads for a certain type of workplace learning course were clicked significantly more frequently between 8 and 10 p.m. on weekdays than at other times, indicating that this time period is a high-engagement opportunity for the target user group. Understanding the timing of ad engagement allows advertisers to choose to push ads at the most appropriate time, improve the effectiveness of ad exposure, and increase user engagement. For example, for takeaway platform ads targeting office workers, if they are placed during the lunch and dinner time periods, they are more likely to attract users to place orders.

[0066] S520. Based on the advertising participation data, adjust the advertising delivery strategy in the user's potential behavior to improve the relevance and conversion rate of the advertisement, and the advertising delivery strategy includes the proportion of advertising behavior.

[0067] Specifically, if advertising is not considered based on actual user engagement, it is often easy to blindly place ads, such as pushing a large number of baby formula ads to young single users who have no interest in maternal and infant products, which not only wastes advertising resources but also easily causes user loss. By analyzing data such as ad click frequency, transaction rate, and engagement timing, we can accurately understand user preferences and behavior patterns, and thus accurately adjust and optimize advertising strategies.

[0068] Advertising behavior ratio refers to the proportion of different types of advertising activities, such as search advertising, information flow advertising, pop-up advertising, etc., or advertising of different products and categories, in the overall advertising plan. For example, an e-commerce platform will use 50% of its advertising budget for clothing advertising, 30% for electronic product advertising, and 20% for household goods advertising.

[0069] If it is found that users have a high click frequency and transaction rate for electronic product ads, and that their participation is better during certain time periods or at certain times, then the proportion of electronic product ads can be appropriately increased, such as increasing its budget share to 40%, and correspondingly reducing the proportion of other categories of ads. At the same time, combined with the participation data of different advertising formats, if it is found that information flow ads bring higher user participation, the proportion of information flow ads in the overall delivery can also be increased, thereby improving the overall relevance and conversion rate of ads, making advertising more in line with users' potential behaviors and needs.

[0070] This embodiment can optimize and improve the advertising effect by accurately obtaining key advertising participation data in user portraits, and then reasonably and flexibly adjusting the advertising delivery strategy based on these data.

[0071] In a specific embodiment, if Figure 7 In this embodiment, the following steps are also specifically included: S610. Regularly collect user feedback data on recommended content, advertisements or services.

[0072] or, S620. When the amount of new user data reaches a preset threshold, the new user data is added to the training data set to ensure the accuracy and timeliness of the model's user analysis.

[0073] Specifically, user feedback data is a direct source of understanding the user's true feelings when using the system, and can reflect key information such as whether the recommended content meets the user's interests, whether the advertisement is attractive, and whether the service meets the needs. For example, users' ratings and comments on songs recommended by a music recommendation platform, or inquiries about products displayed in advertisements on e-commerce platforms, are all feedback data.

[0074] The preset threshold is set to ensure that the amount of data added to the training data set is of a certain scale and representativeness, to avoid overfitting of the model or poor training results due to too frequent or small amounts of data being added. For example, every 1,000 new valid user data is set as a threshold, and data integration is performed when this number is reached. This allows the model to learn and update on a relatively stable basis with sufficient incremental information.

[0075] As time goes by and business develops, user behaviors and preferences will change. Timely addition of new user data can allow the model to continuously adapt to new situations. For example, the dating preferences of users of a social software may change from initially dating in the same city to dating based on interests and hobbies. The addition of new data can enable the model to accurately capture this change, making subsequent analysis of user matching more realistic and ensuring the accuracy of the analysis; at the same time, it can also allow the model to keep up with the latest characteristics of current users and reflect timeliness.

[0076] S630. Analyze feedback data or new user data to measure the current prediction accuracy of the large model.

[0077] Specifically, this embodiment can use a variety of statistical analysis methods and comparison means. For example, the recommended content that the model predicts the user will be interested in is matched and compared with the content that the user actually expressed interest in in feedback, and the matching degree is calculated; or the error analysis is performed based on the user's estimated probability of clicking on the advertisement and the actual click situation. Through these analyses, it is possible to quantitatively know how accurate the model is in predicting user behavior and preferences at the current stage.

[0078] Clarifying the current prediction accuracy is the prerequisite for subsequent targeted optimization. Only by clearly knowing the advantages and disadvantages of the model prediction can we take improvement measures to optimize the model.

[0079] S640. Adjust the model parameters of the large model according to the current prediction accuracy, train and optimize the large model to improve the future prediction accuracy of the large model.

[0080] Specifically, based on the measured prediction accuracy, if the recommendation accuracy for a certain user group is low, it may be necessary to adjust the model parameters related to the feature identification and interest matching of the group. For example, for the news information recommendation model, if the data analysis shows that the accuracy of recommending science and technology information to young user groups is not high, the parameters related to the weight of science and technology keywords and the weight of young users' interests can be adjusted.

[0081] This embodiment continuously adjusts parameters to enable the model to better learn the rules and patterns in user data, reduce prediction errors, and improve the ability to make more accurate predictions when facing similar user scenarios and needs in the future. Continuous training and optimization can enable the large model to continuously evolve and adapt to the ever-changing user behavior environment.

[0082] S650. Analyze the characteristics of each user group to optimize product recommendations, advertising placement, or service provision.

[0083] This embodiment can divide users according to multiple dimensions, such as age, gender, region, consumption habits, frequency of use, etc. For example, e-commerce platform users can be divided into young high-consumption groups, middle-aged stable consumption groups, elderly affordable consumption groups, etc. Data analysis is performed based on the behavioral characteristics of different groups shown in historical data, such as purchased product categories, advertising preferences, and service response time requirements.

[0084] This embodiment can formulate differentiated strategies based on the characteristics of different groups. For young high-spending groups, fashionable and high-end products can be recommended, and creative and trendy advertisements can be placed; for the elderly affordable consumer group, the focus is on recommending cost-effective and practical products, using simple and straightforward advertisements, and improving the acceptance of product recommendations and the conversion rate of advertising through precise matching.

[0085] S660. Adjust parameters of large models based on user feedback to improve the accuracy of analysis.

[0086] For example, if a user reports that a certain video recommendation model always recommends repeated video types, then the relevant parameters for controlling recommendation diversity will be adjusted for the user to increase the probability of different types of videos being recommended.

[0087] This embodiment uses parameter adjustment guided by user feedback to flexibly adjust the large model to better meet the actual expectations of users, making it more accurate in analyzing user interests, behaviors, etc., so that it can be better applied to business application scenarios such as product recommendations and advertising.

[0088] The solution provided in this embodiment can form a complete closed loop that continuously optimizes the large model and related business strategies based on user data and feedback, which helps to continuously improve the system's adaptability to users and the efficiency of content recommendation.

[0089] In a specific embodiment, if Figure 8 In step S40, cluster analysis, association rule mining or classification algorithm is used to analyze the user and output the user classification result, which specifically includes the following steps: S710. Using a density-based clustering algorithm, users are divided into different clusters according to the density distribution of user characteristics, so as to discover high-density areas and abnormal points in the user group.

[0090] Specifically, the objects in the data set in the density-based clustering algorithm, in this embodiment, are users, characterized by their various features, and their distribution in space is dense or sparse. Areas with high density mean that there are more users with similar features gathered together, while areas with low density may have outliers, that is, users with large differences in characteristics from the surrounding groups. For example, in the user data of the e-commerce platform, the feature space is constructed based on features such as purchase frequency, consumption amount, and browsed product categories. Users who frequently purchase maternal and child products with high consumption amounts and browse categories concentrated on baby products may form a high-density area, while users who occasionally purchase products that have nothing to do with maternal and child products are relatively likely to be outliers.

[0091] This embodiment can also divide users into different clusters, which can clearly sort out user groups with similar behaviors and characteristics. This cluster can more accurately discover naturally formed user groups from the perspective of feature density, which is helpful for subsequent in-depth exploration of the commonalities and differences between different groups. For example, for a social platform, users with high activity, frequent participation in topic discussions and many friends can be clustered into one cluster, and users who are occasionally online and rarely interact can be clustered into another cluster, so as to design different function guidance cases in a targeted manner.

[0092] S720. Form a user classification result according to the user characteristics corresponding to each high-density area.

[0093] Specifically, after determining the high-density areas corresponding to the users, it is necessary to analyze the common and representative features of the users in the area. For example, in an online education platform, the user characteristics of a certain high-density area are frequent browsing of programming courses, participating in free trials related to programming, and purchasing elementary programming teaching materials, etc. The users in this area can be classified as "programming learning intention users." By refining and summarizing the features of users in different high-density areas, a multi-dimensional user classification result based on actual behavior and interest characteristics can be formed. Compared with simple demographic classification, this classification is more in line with the actual status of users in specific business scenarios and can better guide subsequent actions.

[0094] Accurate user classification results can allow the system to clearly understand its own user composition, know the size and core characteristics of different types of users, and facilitate "teaching students in accordance with their aptitude" for each type of user. For example, for the "fashion trend followers" user classification of the e-commerce platform, the latest clothing, accessories and other products can be recommended to them; for the "cost-effective pursuers" classification, the focus is on recommending practical products with large discounts to improve the accuracy and effectiveness of marketing.

[0095] S730. Analyze the abnormal situation of the abnormal point.

[0096] Specifically, anomalies represent individual users or small groups with obvious differences in characteristics from the mainstream user group. There may be many reasons for their occurrence. For example, it may be that new users have just registered and have not yet formed a stable behavior pattern, or it may be that users have special needs and unique usage habits. Taking the video platform as an example, most users watch videos for about 1-3 hours a day, and some users watch for more than 8 hours a day, which are anomalies. Analyzing anomalies can, on the one hand, explore potential new needs. Perhaps these abnormal users represent a niche but potential application scenario; on the other hand, it can also check whether there are problems such as data errors to ensure the accuracy of the overall analysis.

[0097] This embodiment can analyze abnormal points from multiple perspectives, check the degree of difference between various indicators of its user characteristics and those of the normal group, and trace its usage history, operation trajectory, etc. For example, for e-commerce users with abnormally high consumption amounts, analyze the categories of goods purchased, the frequency of orders, and whether there are bulk purchases, etc., to determine whether it is a collective purchase behavior or an individual special consumption behavior, etc., and then decide whether it is necessary to formulate a separate service strategy for such special users or adjust some rules in the data analysis model.

[0098] After outputting the user classification results, it also includes: S740. Generate visualization reports, data reports or system interface call data based on the user classification results to provide a basis for the system's marketing, product recommendations, and customer service strategy formulation. The visualization report includes: user group distribution pie charts, user behavior trend line charts, and user feature correlation heat maps to intuitively display user analysis results.

[0099] Specifically, the visualization report includes user group distribution pie charts, user behavior trend line charts, and user feature correlation heat maps. Among them, the user group distribution pie chart can clearly show the proportion of different user categories in the total number of users, and understand the general situation of the system user structure. For example, in a mobile application, a pie chart can be used to intuitively show the percentage of various user groups such as "young game enthusiasts", "workplace learning needs" or "life service inquirers" in the total number of users, so as to understand the core user groups and confirm the focus of resource allocation.

[0100] User behavior trend line chart: It can show the behavior characteristics of a certain user group over time. For example, the "high-spending loyal users" of an e-commerce platform, their monthly purchase frequency, consumption amount and other indicators are displayed through a line chart. It can help the system analyze the activity of this type of users and whether the consumption trend is rising or falling, so as to take timely corresponding content push to stimulate consumption and improve loyalty.

[0101] User feature correlation heat map: It can intuitively reveal the degree of correlation between different user features. For example, on a travel booking platform, the heat map can show the correlation between user age, income level, and the type of travel destination, accommodation grade, and other features, helping the system to better understand the factors that affect user decisions and optimize product recommendations and marketing strategies, such as focusing on recommending high-end overseas travel products to high-income middle-aged users.

[0102] Data reports list in detail the data information related to each user classification in the form of tables, such as the number of each type of users, behavioral data statistics in different time periods, etc., to facilitate subsequent specific data query, comparative analysis, and archiving records, and provide a detailed data basis for in-depth analysis of user situations.

[0103] The system interface calls data to facilitate data interaction and collaboration between different systems within the system, such as the marketing system and customer service system. For example, the marketing system can call user classification data to accurately screen target users for advertising, and the customer service system can understand the common problems that different types of users may encounter in advance based on user classification, make corresponding service plans, and improve overall operational efficiency and service quality.

[0104] This embodiment forms a complete user analysis and decision-making support system from user data mining and classification to result display and application, ensuring accurate push of content that meets user needs.

[0105] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0106] In one embodiment, a user analysis device based on a large model is provided, and the user analysis device based on a large model corresponds one-to-one to the user analysis method based on a large model in the above embodiment. Fig. 9 As shown, the user analysis device based on the large model includes a data acquisition module 10, a data preprocessing module 20, a user portrait output module 30, a classification result output module 40 and a recommended content push module 50. The functional modules are described in detail as follows: The data acquisition module 10 is used to acquire user data from multiple data sources, including user behavior data, user attribute data, user social data and consumption habit data.

[0107] The data preprocessing module 20 is used to preprocess the user data, including data cleaning, data standardization, and data normalization, so as to form a structured data set.

[0108] The user portrait output module 30 is used to input the structured data set into a pre-trained large model based on deep learning, and to output the current user portrait after extracting features from the user data through the large model.

[0109] The classification result output module 40 is used to analyze the user according to the user portrait by using cluster analysis, association rule mining or classification algorithm, and output the user classification result.

[0110] The recommended content push module 50 is used to predict user behavior based on the user group characteristics corresponding to the user classification results in combination with the user portrait, so as to push personalized recommended content to the user.

[0111] For the specific definition of the user analysis device based on the big model, please refer to the definition of the user analysis method based on the big model above, which will not be repeated here. Each module in the above-mentioned user analysis device based on the big model can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the electronic device in the form of hardware, or can be stored in the memory of the electronic device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0112] In one embodiment, an electronic device is provided. The electronic device may be a server, and its internal structure diagram may be as shown in FIG. Fig.10 As shown. The electronic device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile medium and an internal memory. The non-volatile medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile medium. The database of the electronic device is used for data related to the user analysis method based on a large model. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a user analysis method based on a large model is implemented.

[0113] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the user analysis method based on the large model in the above embodiment is implemented, for example Figure 2 Alternatively, when the processor executes the computer program, the functions of each module / unit of the user analysis device based on the large model in the above embodiment are realized, for example Fig. 9 The functions of the modules 10 to 50 are shown in Figure 1. To avoid repetition, they will not be described in detail here.

[0114] In one embodiment, a computer readable medium is provided on which a computer program is stored. When the computer program is executed by a processor, the user analysis method based on a large model in the above embodiment is implemented, for example Figure 2 Alternatively, when the computer program is executed by the processor, the functions of each module / unit in the user analysis device based on the large model in the above device embodiment are realized, for example Fig. 9 The functions of the modules 10 to 50 are shown in Figure 1. To avoid repetition, they will not be described in detail here.

[0115] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments of the present application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0116] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0117] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A user analysis method based on a large model, characterized in that: include: Obtain user data from multiple data sources, including user behavior data, user attribute data, user social data, and consumption habit data; Preprocessing the user data, including data cleaning, data standardization, and data normalization, to form a structured data set; Inputting the structured data set into a pre-trained large model based on deep learning, so as to extract features from the user data through the large model and output the current user portrait; According to the user portrait, cluster analysis, association rule mining or classification algorithm is used to analyze the user and output the user classification result; Based on the user group characteristics corresponding to the user classification results and combined with the user portrait, user behavior prediction is performed to push personalized recommended content to users.

2. The user analysis method based on a large model as claimed in claim 1, characterized in that: The acquiring of user data from multiple data sources includes: Track users' online behavior through the portal and record users' online trajectory data; Collect users’ social interaction data through social media terminals; Collect users’ e-commerce interaction data through the e-commerce terminal; Obtain device usage data through user login behavior; A real-time connection is established with the data source through the data acquisition interface. When new user data is generated in the data source, the new user data is transmitted to the data acquisition system in real time for subsequent processing.

3. The user analysis method based on a large model as claimed in claim 1, characterized in that: The pushing of personalized recommended content to the user includes: Based on user behavior prediction and user portrait, personalized recommendation content is selected from multiple dimensions to generate a dynamic product recommendation list; According to the dynamic product recommendation list, push corresponding target recommendation content to the user; When pushing the target recommended content, user data is captured regularly, and the preprocessing of the user data is repeatedly performed, and the structured data set is input into a pre-trained large model based on deep learning, which is used to extract features from the user data through the large model and output the current user portrait. According to the user portrait, cluster analysis, association rule mining or classification algorithm is used to analyze the user to obtain a user classification result. According to the user group characteristics corresponding to the user classification result and in combination with the user portrait, user behavior prediction is performed, thereby pushing personalized recommended content to the user, so as to update the dynamic product recommendation list in real time.

4. The user analysis method based on a large model as claimed in claim 3, characterized in that: The method of selecting personalized recommendation content from multiple dimensions based on user behavior prediction and user portrait to generate a dynamic product recommendation list includes: Based on user behavior prediction, obtain a list of potential user behaviors; According to the user portrait and each of the user's potential behaviors, the behavior ratio of each of the user's potential behaviors is obtained, so as to determine the content recommendation frequency of each of the personalized recommended contents according to the behavior ratio; Based on the content recommendation frequency, the dynamic product recommendation list is generated.

5. The user analysis method based on a large model as claimed in claim 4, characterized in that: The obtaining the behavior proportion of each potential behavior of the user according to the user portrait and each potential behavior of the user includes: Obtaining advertising participation data in the user portrait, wherein the advertising participation data includes the user's advertising click frequency, advertising conversion rate, and advertising participation timing; According to the advertisement participation data, an advertisement delivery strategy in the user's potential behavior is adjusted, and the advertisement delivery strategy includes a proportion of advertisement behavior.

6. The user analysis method based on a large model as claimed in claim 1, characterized in that: Also includes: Regularly collect user feedback data on recommended content, advertisements or services; or, When the amount of new user data reaches a preset threshold, the new user data is added to the training data set; Analyzing the feedback data or new user data to measure the current prediction accuracy of the large model; Adjusting the model parameters of the large model according to the current prediction accuracy, and training and optimizing the large model; Analyze the characteristics of each user group to optimize product recommendations, advertising, or service provision; Adjust parameters of large models based on user feedback to improve the accuracy of analysis.

7. The user analysis method based on a large model as claimed in claim 1, characterized in that: The method of analyzing users by cluster analysis, association rule mining or classification algorithm and outputting user classification results includes: A density-based clustering algorithm is used to divide users into different clusters according to the density distribution of user characteristics, which is used to discover high-density areas and abnormal points in the user group; Forming a user classification result according to the user characteristics corresponding to each of the high-density areas; Performing abnormal situation analysis on the abnormal point; After outputting the user classification result, the method further includes: A visualization report, a data report or system interface call data is generated according to the user classification result, wherein the visualization report includes: a user group distribution pie chart, a user behavior trend line chart and a user feature correlation heat map.

8. A user analysis device based on a large model, characterized in that: include: The data acquisition module is used to obtain user data from multiple data sources, including user behavior data, user attribute data, user social data and consumption habit data; A data preprocessing module, used to preprocess the user data, including data cleaning, data standardization, and data normalization, to form a structured data set; A user portrait output module, used to input the structured data set into a pre-trained large model based on deep learning, and to output the current user portrait after extracting features from the user data through the large model; A classification result output module is used to analyze the user according to the user portrait by using cluster analysis, association rule mining or classification algorithm, and output the user classification result; The recommended content push module is used to predict user behavior based on the user group characteristics corresponding to the user classification results and combined with the user portrait, so as to push personalized recommended content to users.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the large model-based user analysis method according to any one of claims 1 to 7 is implemented.

10. A computer-readable medium storing a computer program, characterized in that: When the computer program is executed by a processor, the large model-based user analysis method as described in any one of claims 1 to 7 is implemented.

Images