Personalized content recommendation and ROI improvement method and system based on multi-dimensional user portrait

Abstract

The present invention discloses a method and system for personalized content recommendation and ROI improvement based on multidimensional user portraits. The method comprises the following steps: collecting user data and generating a multidimensional user portrait through feature extraction and weighted fusion; constructing a correlation graph using the advertising platform content feature vector and the multidimensional portrait as nodes of a graph structure, and the user's interactive behavior data as edges of the graph structure; analyzing the correlation graph and the multidimensional portrait using a causal inference model to identify invalid related variables and key variables that influence user decision-making; making recommendation decisions based on the causal inference results within a multi-objective optimization framework to generate a personalized recommendation list for each user; and calculating the potential conversion value of different user groups based on a time series prediction model to optimize the advertising budget allocation strategy. The present invention can improve the personalization of advertising recommendations and the ROI of advertising delivery.

Description

Technical Field

[0001] The present invention belongs to the technical field of advertising delivery, and in particular relates to a method and system for personalized content recommendation and ROI improvement based on multi-dimensional user portraits. Background Art

[0002] Personalized recommendation technology plays a vital role in a variety of fields, including e-commerce, digital marketing, news push, and short video platforms. By analyzing users' historical behavior and interests, recommendation systems can accurately match users with appropriate content, thereby improving user experience, increasing engagement, and boosting business conversions. With the development of big data and artificial intelligence technologies, personalized recommendations have gradually evolved from traditional methods based on collaborative filtering to intelligent recommendation systems that integrate multidimensional data, deep learning, causal reasoning, and other technologies.

[0003] In advertising and commercial marketing, return on investment (ROI) is a key metric for measuring advertising effectiveness. Improving ROI depends not only on precise, personalized recommendations but also requires comprehensive consideration of factors such as ad content diversity, click-through rate, and long-term user retention. An efficient recommendation system not only increases the probability of users clicking on ads or content but also fosters long-term user engagement, maximizing advertising revenue.

[0004] The Chinese invention patent application with publication number CN117593053A discloses a big data analysis system for advertising delivery in the field of advertising technology, including: a data collection unit: responsible for collecting user data; a data storage unit: used to store and manage massive user data, as well as perform efficient access and query; a data analysis unit: through the application of machine learning, data mining and other technologies, in-depth analysis of user data is performed to extract user characteristics and behavior patterns; an advertising matching unit: matches the advertising information provided by the advertiser with the user data to determine the optimal advertising delivery plan.

[0005] The above solution optimizes advertising delivery based on big data. However, as user behavior patterns become increasingly complex, relying solely on a single data source (such as clickthrough rate) is no longer sufficient to accurately capture user needs. Consequently, recommendation technologies based on multidimensional user profiles have become a research hotspot. Furthermore, traditional advertising budget allocation often relies on historical conversion rates or set rules, making it difficult to dynamically adapt to the potential conversion value of different user groups. This solution is easily misled by data correlation, which can affect the accuracy of advertising delivery. Summary of the Invention

[0006] The present invention provides a personalized content recommendation and ROI improvement method and system based on multi-dimensional user portraits, aiming to solve problems such as insufficient accuracy of existing technologies in the field of advertising delivery.

[0007] To solve the above technical problems, the present invention proposes a personalized content recommendation and ROI improvement method, which includes the following steps:

[0008] Collect user behavior data, social data, contextual data, emotional data, and content consumption trend data, and generate a multi-dimensional profile of the user through feature extraction and weighted fusion;

[0009] The advertising platform content feature vector and the multi-dimensional portrait are used as nodes of the graph structure, and the user's interactive behavior data are used as edges of the graph structure to construct a correlation graph;

[0010] Use causal reasoning models to analyze the association maps and multi-dimensional portraits to identify invalid related variables and key variables that affect user decisions;

[0011] Make recommendation decisions based on causal reasoning results within a multi-objective optimization framework and generate a personalized recommendation list for each user.

[0012] Calculate the potential conversion value of different user groups based on the time series prediction model and optimize advertising budget allocation strategy.

[0013] Preferably, the behavioral data includes browsing history, click history, purchase history, length of stay and collection history; the social data includes friend relationships, interactions and shared content; the contextual data includes device type, geographic location, access time and network environment; the emotional data is user-based text, voice and expression analysis data; the content consumption trend data includes hot topics and data on changes in user interests over time.

[0014] Preferably, the feature extraction includes the following methods:

[0015] After selecting the characteristic dimensions of behavioral data and social data, the data of the corresponding characteristic dimensions are normalized or standardized, and then spliced into behavioral vectors and social vectors respectively;

[0016] The context data is one-hot encoded and the encoded data is concatenated into a context vector;

[0017] Sentiment data is analyzed using natural language processing, and sentiment analysis scores are converted into sentiment vectors;

[0018] Content consumption trend data is converted into content consumption trend vectors through keyword extraction and one-hot encoding.

[0019] Preferably, the specific method for identifying invalid related variables and key variables that affect user decision-making is:

[0020] Construct a causal graph, using a directed acyclic graph to represent the causal relationship between variables;

[0021] For each variable, calculate its impact on the user's decision:

[0022]

[0023] Where, Representing variables The average causal effect on user decisions; In the intervention group, the variable The expected value of the overall user decision after the intervention is applied; In the control group, the variable The expected value of the overall user decision when no intervention is applied;

[0024] According to the average causal effect, the variables are divided into invalid related variables and key variables according to the set rules.

[0025] Preferably, the specific method of identifying invalid related variables and key variables that affect user decision-making further includes calculating individual causal effects for different user groups:

[0026]

[0027] Where, For a single user In terms of variables Individual causal effects that influence user decisions, used for decision-making recommendations; Indicates that when a single user The expected value of the user's decision when the intervention is applied; Indicates that no single user The expected value of the user's decision when the intervention is applied;

[0028] According to the average causal effect and individual causal effect, the variables are divided into invalid related variables and key variables according to the set rules.

[0029] Preferably, the recommendation decision is made based on the causal reasoning results under the multi-objective optimization framework, and the specific method is:

[0030] Define a multi-objective optimization function:

[0031]

[0032] Where, To maximize the optimization objective, is the click-through rate, which indicates the user’s interest in the recommended content. Return on investment, which indicates the purchase conversion rate of advertisements or products. To increase the diversity of advertising content, For the long-term benefits of advertising, is a hyperparameter used to adjust the weight of each target;

[0033] Optimize multi-objective optimization functions using Pareto frontiers;

[0034] Calculate the final recommendation score of each user for different content based on the Pareto front optimization results;

[0035] The final recommendation scores are sorted, and the first one or more items are selected to generate a personalized recommendation list, which is then displayed to the corresponding user.

[0036] Preferably, the personalized recommendation list dynamically adjusts the recommendation strategy through a multi-armed bandit algorithm to generate an adjusted personalized recommendation list, and displays the adjusted personalized recommendation list to the user; the multi-armed bandit adopts the Thompson sampling algorithm to optimize the reward function.

[0037] Preferably, the optimization method of the advertising budget allocation strategy is specifically as follows:

[0038] Divide user groups, calculate the multi-dimensional portraits of all individual users in the group, and calculate the corresponding group feature vectors;

[0039] Build and train time series forecasting models based on ARIMA or LSTM;

[0040] Use the trained time series model to predict the conversion rate and click-through rate of ads for different user groups in different time periods in the future;

[0041] Calculate the potential conversion value of each segment based on predicted conversion and click-through rates;

[0042] Calculate the budget ratio for each group based on the potential conversion value weights of different groups, and deliver ads according to the budget ratio.

[0043] Preferably, the division of user groups includes the following division methods:

[0044] Segment users into groups based on their demographic characteristics;

[0045] Divide user groups based on their historical behavior data;

[0046] Use the K-means clustering algorithm, density-based clustering algorithm, or Gaussian mixture model clustering algorithm to cluster users and divide them into high-value users, potential conversion users, and low-value users.

[0047] Accordingly, the present invention also proposes a personalized content recommendation and ROI improvement system based on multi-dimensional user portraits. The system is used to implement the above-mentioned personalized content recommendation and ROI improvement method, including:

[0048] User data collection module, used to collect user behavior data, social data, contextual data, emotional data and content consumption trend data, and store and pre-process the data;

[0049] Feature extraction and profile building module, used to extract features from collected data and generate a multi-dimensional profile of the user;

[0050] A graph structure construction module is used to construct a correlation graph by using the advertising platform content feature vector and the multi-dimensional portrait as nodes of the graph structure and the user's interactive behavior data as edges of the graph structure;

[0051] The causal reasoning analysis module is used to analyze correlation maps and multi-dimensional portraits based on causal reasoning models to identify invalid related variables and key variables that affect user decisions;

[0052] The recommendation decision optimization module is used to make recommendation decisions under a multi-objective optimization framework and to implement dynamic recommendation strategy adjustments using a multi-armed bandit machine;

[0053] Ad budget optimization module, which is used to calculate the potential conversion value of different user groups based on the time series prediction model and optimize the ad budget allocation strategy;

[0054] The recommended content ranking and display module is used to calculate the final recommendation score of each user for different content based on the Pareto front optimization results, and generate a personalized recommendation list for display;

[0055] The system storage and management module is used to store user data, model parameters, optimization results and recommendation records, and supports system operation status monitoring and management.

[0056] Compared with the prior art, the present invention has the following technical effects:

[0057] 1. The personalized content recommendation and ROI improvement method proposed in this invention uses multi-dimensional user portraits to improve recommendation accuracy, solving the problem that traditional methods cannot accurately predict user interests, avoiding the recommendation content deviation caused by a single user portrait, enabling the recommendation system to adapt to user needs in different scenarios, and improving the personalization of recommendations.

[0058] 2. The personalized content recommendation and ROI improvement method proposed in this paper uses a causal inference model to analyze user-content relationships, identify key variables that truly influence user decisions, and eliminate invalid correlation factors. This solves the problem of misleading recommendations based on statistical correlation, avoids spurious correlations that influence recommendations, ensures that recommended content is more aligned with users' true interests, and improves conversion rates. It also makes advertising more targeted, reduces invalid exposure, lowers advertising costs, and improves ROI.

[0059] 3. The personalized content recommendation and ROI improvement method proposed in this paper utilizes a multi-armed bandit algorithm to achieve an exploration-exploitation balance, continuously optimizing the recommendation list, dynamically adjusting the recommendation strategy, and increasing user interaction. Incorporating real-time user feedback ensures that recommended content is continuously optimized to meet evolving user interests. This solves the problem of fixed recommendation strategies being unable to adapt to changing user interests, enhances the system's adaptability, effectively improves ad conversion rates and user click-through rates, and improves long-term user engagement.

[0060] 4. The personalized content recommendation and ROI improvement method proposed in this paper uses the ARIMA / LSTM time series prediction model to predict the potential conversion value of different user groups and optimize advertising budget allocation. This solves the problem that traditional advertising budget allocation relies on historical data and lacks dynamic adjustment capabilities. Based on predicted future conversion rates and click-through rates, it accurately adjusts advertising delivery strategies and improves ROI. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flowchart of the personalized content recommendation and ROI improvement method of the present invention. DETAILED DESCRIPTION

[0062] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0063] Example 1

[0064] This embodiment is a personalized content recommendation and ROI improvement method based on multi-dimensional user portraits. Figure 1 As shown, the following steps are included from step 1 to step 5:

[0065] Step 1: Collect user behavior data, social data, situational data, emotional data, and content consumption trend data, and generate a multi-dimensional portrait of the user through feature extraction and weighted fusion.

[0066] The behavioral data includes browsing history, click history, purchase history, length of stay and collection history; the social data includes friend relationships, interactions and shared content; the contextual data includes device type, geographic location, access time and network environment; the emotional data is user-based text, voice and expression analysis data; the content consumption trend data includes hot topics and data on changes in user interests over time.

[0067] Data sources can include user behavior, social networks, environmental factors, emotional states, and content consumption records, including but not limited to the following data:

[0068] Behavioral data comes from users' browsing, clicking, collecting, commenting, sharing, purchasing records and duration of stay. Data features include click-through rate, dwell time, purchase frequency, bounce rate, etc.

[0069] Social data comes from user friendships, interaction records, following / following, social platform content, and likes / reposts. Data features include friend recommendations, group influence, and shared interests.

[0070] Contextual data includes the user's device information, geographic location, time, and network environment. Data characteristics include device type, access time period, location preference, etc.

[0071] Sentiment data can be collected from user comments, posts, comments, and voice messages. The data features are the sentiment tendencies of user comments (such as positive, neutral, and negative).

[0072] Content consumption trends are reflected in users' content consumption preferences. Specific data features include preferred content categories, recent popularity trends, etc.

[0073] The feature extraction method of this embodiment adopts different methods according to different data:

[0074] After selecting feature dimensions for behavioral and social data, the dimensions that can be selected include: Since different features have different value ranges and require normalization, the data of the corresponding feature dimensions can be normalized or standardized, and then concatenated into behavioral and social vectors, respectively. Specifically, Min-Max normalization can be used to normalize data with a relatively stable value range, such as click-through rate; Z-Score normalization can be used to normalize data with a wide distribution, such as the number of purchases; and Log transformation can be used to normalize data with long-tail distribution characteristics, such as the number of product purchases. Converting behavioral data into feature vectors generally involves feature normalization, statistical feature extraction, and time window aggregation. Common methods include one-hot encoding, Min-Max normalization, TF-IDF (Term Frequency–Inverse Document Frequency), and embedding representation.

[0075] Taking behavioral data as an example, the specific process of feature extraction in this embodiment is explained. First, a set of key behavioral data is set in the collected behavioral data, such as: click behavior (number of clicks, click frequency), dwell time (page dwell time, average dwell time), conversion behavior (purchase, collection, sharing, etc.), time window (statistics of time periods such as 7 days and 30 days). Since different features have different value ranges, normalization processing is required. Those skilled in the art can select appropriate normalization / standardization processing methods according to data characteristics, such as Min-Max normalization, Z-Score normalization and Log transformation. Then convert the behavioral data into a feature vector. Suppose a user's behavioral data is as follows:

[0076] User ID: 123456; Number of clicks in the past 7 days: 120; Average stay time in the past 7 days (seconds): 80; Number of purchases in the past 30 days: 5; Number of shares in the past 7 days: 10; Number of collections in the past 7 days: 20.

[0077] Based on global data statistics, the maximum values for the four metrics of clicks in the past seven days, average dwell time in the past seven days, shares in the past seven days, and favorites in the past seven days were 500, 300, 50, and 100, respectively, and the minimum values were all 0. After Min-Max normalization, their values were 0.24, 0.267, 0.2, and 0.2, respectively. For the metric of purchases in the past 30 days, after log normalization, the value was 1.792.

[0078] Concatenate the above data into a behavior vector: X1=(0.24, 0.267, 1.792, 0.2, 0.2).

[0079] Contextual data is one-hot encoded and concatenated into context vectors. Key features such as device type (PC, mobile, tablet), operating system (iOS, Android, Windows), active time periods (morning / noon / evening), and geographic locations (country, province, city) can all be one-hot encoded.

[0080] Sentiment data is analyzed using natural language processing, and the sentiment analysis scores are converted into sentiment vectors. This example uses BERT, LSTM, and SVM to classify sentiment (positive, neutral, and negative) and calculate sentiment scores (in the range [-1, 1]). Sentiment fluctuations are then analyzed, such as the mean and variance of sentiment over the past seven days. Assume that a user's recent comment analysis is as follows: text sentiment score: 0.8; mean: 0.75; variance: 0.1. This is converted to a vector: X² = (0.8, 0.75, 0.1).

[0081] Content consumption trend data is converted into a content consumption trend vector through keyword extraction and one-hot encoding. Specifically, after extracting keywords using TF-IDF, the percentage of each keyword in all categories is calculated as the initial data. Because the percentage data range is [0, 1], this embodiment does not perform normalization or standardization on this percentage data. Content consumption trend data also includes topic popularity data, which is also converted into a numerical value using one-hot encoding. All content consumption trend data is then concatenated into a content consumption trend vector.

[0082] The above-mentioned spliced vectors are all vectors corresponding to a single user. Finally, each user's behavior vector, social vector, situational vector, emotional vector and content consumption trend vector are spliced together to generate a multi-dimensional portrait Xu of each user.

[0083] In some other embodiments of the present invention, principal component analysis (PCA) may be performed on the normalized / standardized, encoded, and converted data to reduce dimensionality in order to improve computational efficiency; or an embedding method may be used to perform low-dimensional vector representation on category features (such as the type of content clicked).

[0084] Step 2: Use the content feature vector of the advertising platform and the multi-dimensional portrait as nodes of the graph structure, and the user's interactive behavior data as edges of the graph structure to construct a correlation graph.

[0085] The user's portrait vector is only a feature of a single node, but to form a complete user-item graph, the user's interactive behavior data is also needed as edges to connect users and content. In the graph structure of this embodiment, it includes user nodes and content nodes. In the user node, each user u is characterized by its multidimensional portrait Xu; in the content node, each content c is characterized by its content feature Xc (such as content category, keywords, release time, etc.). The edge (u, c) connecting user u and content c represents the user's interactive behavior, such as clicks, likes, purchases, etc. The weight of the edge can be calculated based on the interaction intensity, for example: the number of clicks (the more times, the greater the edge weight); the length of stay (the longer the stay, the greater the edge weight); the conversion rate (purchase, collection, etc.).

[0086] Assume the following interaction data:

[0087]

[0088] Edge weights can be defined:

[0089]

[0090] in, represents the weight of the edge between users and content, is a hyperparameter.

[0091] The source of content nodes depends mainly on the specific scenario of the recommendation system, such as e-commerce, social media, news platforms, etc. Generally speaking, content nodes come from the recommended content on the platform, including:

[0092] E-commerce platforms (Taobao, Amazon, etc.): product (SKU), product category, brand;

[0093] Short video platforms (TikTok, Kuaishou, Bilibili, etc.): video content, author, and tags;

[0094] News platforms (such as Toutiao): news articles, topics, and keywords;

[0095] Social media (Xiaohongshu, Weibo, etc.): posts, topics, tags.

[0096] The features of the content nodes need to be converted into vectors Xc in the same way as the conversion of multi-dimensional portraits.

[0097] Step three: Use a causal reasoning model to analyze the association map and multi-dimensional portrait to identify invalid related variables and key variables that affect user decisions.

[0098] Through causal reasoning, this embodiment achieves the following goals:

[0099] Identify causal variables: features that have a direct causal impact on user behavior (e.g., clicks, conversions, purchases, etc.);

[0100] Eliminate spurious correlations: variables that are only correlated but not causally related to improve the generalization ability and recommendation accuracy of subsequent steps.

[0101] The specific method for identifying invalid related variables and key variables that affect user decisions is:

[0102] Construct a causal graph, using a directed acyclic graph to represent the causal relationship between variables;

[0103] For each variable, calculate its impact on the user's decision:

[0104]

[0105] Where, Representing variables The average causal effect on user decisions; In the intervention group, the variable The expected value of the overall user decision after the intervention is applied; In the control group, the variable The expected value of the overall user decision when no intervention is applied;

[0106] According to the average causal effect, the variables are divided into invalid related variables and key variables according to the set rules.

[0107] In some other embodiments of the present invention, the specific method of identifying invalid related variables and key variables that affect user decision-making further includes calculating individual causal effects for different user groups:

[0108]

[0109] Where, For a single user In terms of variables Individual causal effects that influence user decisions, used for decision-making recommendations; Indicates that when a single user The expected value of the user's decision when the intervention is applied; Indicates that no single user The expected value of the user's decision when the intervention is applied;

[0110] According to the average causal effect and individual causal effect, the variables are divided into invalid related variables and key variables according to the set rules. Specifically, based on the above analysis, we can classify the variables as follows:

[0111]

[0112] Regarding the high, medium, and low ATE classifications, in this embodiment, ATE represents the increase in user purchase rates due to advertising, with ATE ≥ 10% being high, 2% ≤ ATE < 10% being medium, and 0 ≤ ATE < 2% being low. Regarding ITE, in this embodiment, IT represents the purchase rate increase of a specific user due to advertising, with ITE > 5% being high, 1% ≤ ITE ≤ 5% being medium, and ITE < 1% being low. Variables are labeled as conclusion variables and their weights are reassigned during the recommendation process. For example, key variables have their weights increased by 60% from their original weights, medium variables retain their original weights, and insignificant related variables have their weights decreased by 75% from their original weights. The specific adjustment range can be set as needed during implementation.

[0113] Step 4: Make recommendation decisions based on causal reasoning results under the multi-objective optimization framework and generate a personalized recommendation list for each user.

[0114] The method for making a recommendation decision based on the causal reasoning result under the multi-objective optimization framework includes the following steps S41-S44:

[0115] S41: Define a multi-objective optimization function:

[0116]

[0117] Where, To maximize the optimization objective, is the click-through rate, which indicates the user’s interest in the recommended content. Return on investment, which indicates the purchase conversion rate of advertisements or products. To increase the diversity of advertising content, For the long-term benefits of advertising, is a hyperparameter used to adjust the weight of each target. The hyperparameter here is reassigned according to the variable label of step S3.

[0118] S42: Optimize the multi-objective optimization function using the Pareto front. Because recommendation tasks involve multiple conflicting optimization objectives (e.g., CTR and ROI may affect each other), this embodiment uses Pareto front optimization to find the optimal recommendation strategy. A set of optimal content is selected from the Pareto front for recommendation score calculation.

[0119] S43: Calculate the final recommendation score of each user for different contents based on the Pareto front optimization results. In this step, the multi-objective optimization results and user personalized weights are combined to calculate the final recommendation score of each user u for a single content in a set of optimal contents. Recommended score:

[0120]

[0121] in, For content Relative recommendation score of user u, Respectively CTR, ROI, ad content diversity, and long-term benefits, It is determined by the user's multi-dimensional portrait. For example, users with a higher purchasing tendency may increase the ROI weight, and users who like novel content may increase the diversity weight.

[0122] S44: Sort the final recommendation scores, select the top item or items to generate a personalized recommendation list, and display it to the corresponding user. In a specific implementation, only one advertisement may be displayed in the end, that is, the number of elements in the personalized recommendation list is 1, or the number of elements in the personalized recommendation list is greater than 1, but only the top-ranked advertisement in the list is displayed to the user; or multiple advertisements may be displayed to the user in a carousel manner, and the multiple advertisements are obtained from the personalized recommendation list according to their ranking.

[0123] Sort the calculated recommendation scores according to The recommended contents of user u are sorted in descending order based on the value of , the number of recommended contents is set to k, the first k recommended contents are selected as the final personalized recommendation list, and the generated personalized recommendation list is fed back to the recommendation system front end.

[0124] In some embodiments of the present invention, the personalized recommendation list dynamically adjusts the recommendation strategy using a multi-armed bandit algorithm to generate an adjusted personalized recommendation list, which is then displayed to the user. The multi-armed bandit optimizes the reward function using the Thompson sampling algorithm. The multi-armed bandit optimizes the reward function using the Thompson sampling algorithm. Specifically, the data used by the multi-armed bandit includes the user's personalized recommendation list, historical click data, historical conversion data, advertising budget constraints, and user feedback (click / purchase behavior). The recommendation score used in step S44 is added to the exploration gain score calculated by the multi-armed bandit to serve as the new ranking basis.

[0125] In some other embodiments of the present invention, the ε-Greedy algorithm or the UCB (Upper Confidence Bound) algorithm may be used to optimize the reward function.

[0126] Step 5: Calculate the potential conversion value of different user groups based on the time series prediction model and optimize the advertising budget allocation strategy.

[0127] The optimization method of the advertising budget allocation strategy specifically includes the following steps S51-S55:

[0128] S51: Divide the user groups, calculate the multi-dimensional portraits of all individual users in the group, and calculate the corresponding group feature vectors.

[0129] In an embodiment of the present invention, dividing user groups includes the following division methods:

[0130] Segment users into groups based on their demographic characteristics;

[0131] Divide user groups based on their historical behavior data;

[0132] Use the K-means clustering algorithm, density-based clustering algorithm, or Gaussian mixture model clustering algorithm to cluster users and divide them into high-value users, potential conversion users, and low-value users.

[0133] S52: Build and train a time series forecasting model based on ARIMA (Auto Regression Moving Average) or LSTM (Long Short-Term Memory).

[0134] S53: Use the trained time series model to predict the advertising conversion rate and click-through rate of different user groups in different time periods in the future.

[0135] The goal of this step is to predict the CTR and CVR for different user groups over the next T time periods. The CTR and CVR data from the last n days / hours are fed into the trained ARIMA / LSTM model to predict the CTR and CVR for future time windows (e.g., 7 or 30 days). The output predictions are: For high-value users, the CTR is expected to be ≥10% and the CVR is expected to be ≥5%; for potential conversion users, the CTR is expected to be 2% or less and the CVR is expected to be less than 10% and the CVR is expected to be less than 1%; for low-value users, the CTR is expected to be less than 2% and the CVR is expected to be less than 1%.

[0136] S54: Calculate the potential conversion value of each group based on the predicted conversion rate and click-through rate. The specific calculation method is as follows:

[0137]

[0138] Where, The potential conversion value of the user group, is the predicted value of group click rate, is the predicted value of group conversion rate, The average revenue per user in the group is obtained from historical data.

[0139] S55: Calculate the budget ratio for each group based on the potential conversion value weights of different groups, and deliver advertisements according to the budget ratio.

[0140] Example 2

[0141] This embodiment is a personalized content recommendation and ROI improvement system based on multi-dimensional user portraits. The system is used to implement the personalized content recommendation and ROI improvement method as described in Example 1, including a user data collection module, a feature extraction and portrait construction module, a graph structure construction module, a causal reasoning analysis module, a recommendation decision optimization module, an advertising budget optimization module, a recommended content sorting and display module, and a system storage and management module.

[0142] The user data collection module is used to collect user behavior data, social data, situational data, emotional data, and content consumption trend data, and store and pre-process the data;

[0143] The feature extraction and profile building module is used to extract features from the collected data and generate a multi-dimensional profile of the user;

[0144] The graph structure construction module is used to construct a correlation graph by using the advertising platform content feature vector and the multi-dimensional portrait as nodes of the graph structure and the user's interactive behavior data as edges of the graph structure;

[0145] The causal reasoning analysis module is used to analyze correlation maps and multi-dimensional portraits based on causal reasoning models, and identify invalid related variables and key variables that affect user decisions;

[0146] The recommendation decision optimization module is used to make recommendation decisions under a multi-objective optimization framework and to implement dynamic recommendation strategy adjustments using a multi-armed bandit machine;

[0147] The advertising budget optimization module is used to calculate the potential conversion value of different user groups based on the time series prediction model and optimize the advertising budget allocation strategy;

[0148] The recommended content ranking and display module is used to calculate the final recommendation score of each user for different content based on the Pareto front optimization results, and generate a personalized recommendation list for display;

[0149] The system storage and management module is used to store user data, model parameters, optimization results and recommendation records, and supports system operation status monitoring and management.

[0150] The above description is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this field, several variations and improvements can be made without departing from the creative concept of the present invention, which all fall within the scope of protection of the present invention.

Claims

1. A personalized content recommendation and ROI improvement method based on multi-dimensional user portraits, characterized by: The following steps are involved: Collect user behavior data, social data, contextual data, emotional data, and content consumption trend data, and generate a multi-dimensional profile of the user through feature extraction and weighted fusion; The advertising platform content feature vector and the multi-dimensional portrait are used as nodes of the graph structure, and the user's interactive behavior data are used as edges of the graph structure to construct a correlation graph; Use a causal reasoning model to analyze the association map and multi-dimensional portrait to identify invalid related variables and key variables that affect user decisions. The specific method is to build a causal graph and use a directed acyclic graph to represent the causal relationship between variables; calculate the impact of each variable on user decision-making: Where, Representing variables The average causal effect on user decisions; In the intervention group, the variable The expected value of the overall user decision after the intervention is applied; In the control group, the variable The expected value of the overall user decision when no intervention is applied; based on the average causal effect, the variables are divided into invalid related variables and key variables according to the set rules; Make recommendation decisions based on causal reasoning results within a multi-objective optimization framework and generate a personalized recommendation list for each user. Calculate the potential conversion value of different user groups based on the time series prediction model and optimize advertising budget allocation strategy.

2. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 1 is characterized in that: The specific method of identifying invalid related variables and key variables that affect user decisions also includes calculating individual causal effects for different user groups: Where, For a single user In terms of variables Individual causal effects that influence user decisions, used for decision-making recommendations; Indicates that when a single user The expected value of the user's decision when the intervention is applied; Indicates that no single user The expected value of the user's decision when the intervention is applied; According to the average causal effect and individual causal effect, the variables are divided into invalid related variables and key variables according to the set rules.

3. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 1 or 2, characterized in that: The behavioral data includes browsing history, click history, purchase history, length of stay and collection history; the social data includes friend relationships, interactions and shared content; the contextual data includes device type, geographic location, access time and network environment; the emotional data is user-based text, voice and expression analysis data; the content consumption trend data includes hot topics and data on changes in user interests over time.

4. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 1 or 2, characterized in that: The feature extraction includes the following methods: After selecting the characteristic dimensions of behavioral data and social data, the data of the corresponding characteristic dimensions are normalized or standardized, and then spliced into behavioral vectors and social vectors respectively; The context data is one-hot encoded and the encoded data is concatenated into a context vector; Sentiment data is analyzed using natural language processing, and sentiment analysis scores are converted into sentiment vectors; Content consumption trend data is converted into content consumption trend vectors through keyword extraction and one-hot encoding.

5. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 1 or 2, characterized in that: The recommendation decision is made based on the causal reasoning results in the multi-objective optimization framework. The specific method is as follows: Define a multi-objective optimization function: Where, To maximize the optimization objective, is the click-through rate, which indicates the user’s interest in the recommended content. Return on investment, which indicates the purchase conversion rate of advertisements or products. To increase the diversity of advertising content, For the long-term benefits of advertising, is a hyperparameter used to adjust the weight of each target; Optimize multi-objective optimization functions using Pareto frontiers; Calculate the final recommendation score of each user for different content based on the Pareto front optimization results; The final recommendation scores are sorted, and the first one or more items are selected to generate a personalized recommendation list, which is then displayed to the corresponding user.

6. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 1 or 2, characterized in that: The personalized recommendation list dynamically adjusts the recommendation strategy through a multi-armed bandit algorithm to generate an adjusted personalized recommendation list, and displays the adjusted personalized recommendation list to the user; the multi-armed bandit adopts the Thompson sampling algorithm to optimize the reward function.

7. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 1 or 2, characterized in that: The optimization method of the advertising budget allocation strategy is specifically as follows: Divide user groups, calculate the multi-dimensional portraits of all individual users in the group, and calculate the corresponding group feature vectors; Build and train time series forecasting models based on ARIMA or LSTM; Use the trained time series model to predict the conversion rate and click-through rate of ads for different user groups in different time periods in the future; Calculate the potential conversion value of each segment based on predicted conversion and click-through rates; Calculate the budget ratio for each group based on the potential conversion value weights of different groups, and deliver ads according to the budget ratio.

8. The personalized content recommendation and ROI improvement method based on multi-dimensional user portraits according to claim 7 is characterized in that: The division of user groups includes the following division methods: Segment users into groups based on their demographic characteristics; Divide user groups based on their historical behavior data; Use the K-means clustering algorithm, density-based clustering algorithm, or Gaussian mixture model clustering algorithm to cluster users and divide them into high-value users, potential conversion users, and low-value users.

9. Personalized content recommendation and ROI improvement system based on multi-dimensional user portraits, characterized by: The system is used to implement the personalized content recommendation and ROI improvement method according to claim 1 or 2, comprising: User data collection module, used to collect user behavior data, social data, contextual data, emotional data and content consumption trend data, and store and pre-process the data; Feature extraction and profile building module, used to extract features from collected data and generate a multi-dimensional profile of the user; A graph structure construction module is used to construct a correlation graph by using the advertising platform content feature vector and the multi-dimensional portrait as nodes of the graph structure and the user's interactive behavior data as edges of the graph structure; The causal reasoning analysis module is used to analyze correlation maps and multi-dimensional portraits based on causal reasoning models to identify invalid related variables and key variables that affect user decisions; The recommendation decision optimization module is used to make recommendation decisions under a multi-objective optimization framework and to implement dynamic recommendation strategy adjustments using a multi-armed bandit machine; Ad budget optimization module, which is used to calculate the potential conversion value of different user groups based on the time series prediction model and optimize the ad budget allocation strategy; The recommended content ranking and display module is used to calculate the final recommendation score of each user for different content based on the Pareto front optimization results, and generate a personalized recommendation list for display; The system storage and management module is used to store user data, model parameters, optimization results and recommendation records, and supports system operation status monitoring and management.

Images