Message processing method, device and system based on user characteristics and medium

Abstract

The invention discloses a message processing method, device and system based on user characteristics and a medium. The method comprises the following steps: acquiring user characteristic information and service rule information; calculating a preference service type of the user according to the user feature information; calculating a recommendation service weight according to the preference service type and the business rule information; and generating and outputting a service message according to the recommended service weight. According to the technical scheme, prediction modeling can be carried out on the preference of the user by utilizing the user characteristics, the click rate of message pushing is improved, the labor cost of operation is reduced, accurate prediction is carried out on the preference of the user by effectively utilizing the user characteristic information, meanwhile, strategy rules set by services are considered, and the user experience is improved. The click rate of the user to the pushed message is effectively improved, manpower resources are saved, the service content can be accurately pushed to the user, and the satisfaction of the user is improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a message processing method, device, system and medium based on user characteristics. Background Art

[0002] In the message recommendation project, accurate and personalized message content is pushed to users to improve user experience, increase click-through rate and thus increase user stickiness to the app.

[0003] Traditional message push is configured by the operation side through manual selection of groups of people. Different organizations need to configure operation groups. In order to recommend personalized message content to users, a recommendation model based on thousands of people is required.

[0004] In previous projects, only operations personnel were used to configure the message content pushed to users. Due to constraints such as time, region, and personnel costs, the push accuracy was low, and it was easy to push content that users were not interested in or even disgusted with, which not only wasted the operations personnel's time but also failed to achieve business results. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a message processing method, device, system and medium based on user characteristics that can be applied to data processing or other related fields. Its main purpose is to use user characteristics to predict user preferences, improve the click rate of message push and reduce operating labor costs.

[0006] The technical solution of the present invention is as follows:

[0007] A first aspect of the present invention provides a message processing method based on user characteristics, comprising:

[0008] Obtain user feature information and business rule information;

[0009] Calculating the user's preferred service type based on the user characteristic information;

[0010] Calculating a recommended service weight according to the preferred service type and the business rule information;

[0011] A service message is generated and outputted according to the recommended service weight.

[0012] In one embodiment, the step of obtaining user feature information and business rule information includes:

[0013] Get the identification ID entered by the user;

[0014] The user feature information and corresponding business rule information are obtained in the feature table according to the identification ID.

[0015] In one embodiment, the step of calculating the user's preferred service type based on the user characteristic information further includes:

[0016] Get the initialization constant model;

[0017] The initialized constant model is iteratively trained to obtain an updated constant model; and the updated constant model is weighted summed to obtain a final constant model.

[0018] In one embodiment, the step of calculating the recommended service weight according to the preferred service type and the business rule information includes:

[0019] The preference service type is combined with the business rule information. When the user is identified as a forced push group based on the combination result, the service weight of the corresponding service preference type is set to 1, and the service weights of other service preference types are 0; if the service weight of the service preference type is 0, the service preference type is input into a random function, the push weight is selected, and the weight is set to the weight corresponding to the service preference type.

[0020] In one embodiment, the step of generating and outputting a service message according to the recommended service weight includes:

[0021] Generate the corresponding service message according to the recommended service weight;

[0022] Use a distributed stream processing platform to set the sending time of the service message and select the sending group;

[0023] The corresponding service message recommended contents are sent to the sending group in batches according to the sending time.

[0024] In one embodiment, the step of iteratively training the initialization constant model to obtain an updated constant model includes:

[0025] Calculating the residual of each user feature information sample under the initialization constant model;

[0026] fitting the residuals;

[0027] The result of the residual fitting is input into the initialization constant model to obtain the update constant model.

[0028] In one embodiment, the step of acquiring user feature information and corresponding business rule information in the feature table according to the identification ID includes:

[0029] Identify the identification ID input by the user;

[0030] Searching the feature table for all data associated with the identification ID to obtain the user feature information;

[0031] According to the user characteristic information, business rule information corresponding to the user characteristic information is obtained in a rule engine table.

[0032] A second aspect of the present invention provides a message processing device based on user characteristic information, comprising:

[0033] An acquisition module is used to obtain user feature information and business rule information;

[0034] A first processing module, configured to calculate the user's preferred service type based on the user characteristic information;

[0035] A second processing module, configured to calculate a recommended service weight according to the preferred service type and the business rule information;

[0036] The third processing module is used to generate and output a service message according to the recommended service weight.

[0037] A third aspect of the present invention provides a message processing system based on user characteristics, the system comprising at least one processor; and

[0038] a memory communicatively connected to the at least one processor; wherein,

[0039] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned message processing method based on user characteristics.

[0040] A fourth aspect of the present invention provides a non-volatile computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the above-mentioned user feature-based message processing method.

[0041] Beneficial effects: The present invention discloses a message processing method, device, system and medium based on user characteristics, the method comprising obtaining user characteristic information and business rule information; calculating the user's preferred service type according to the user characteristic information; calculating the recommended service weight according to the preferred service type and the business rule information; generating and outputting a service message according to the recommended service weight. According to the technical solution of the present invention, it is possible to use user characteristics to predict and model the user's preferences, improve the click-through rate of message push and reduce the operating manpower cost, effectively use user characteristic information to accurately predict user preferences, and also consider the policy rules set by the business, effectively improve the user's click-through rate of push messages, and save human resources, which can accurately push service content to users and improve user satisfaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0043] Figure 1 A flow chart of a message processing method based on user characteristics provided by an embodiment of the present invention;

[0044] Figure 2 A flowchart of step S100 in the message processing method based on user characteristics provided in an embodiment of the present invention;

[0045] Figure 3 A flowchart of step S200 in the message processing method based on user characteristics provided in an embodiment of the present invention;

[0046] Figure 4 A flowchart of step S300 in the message processing method based on user characteristics provided in an embodiment of the present invention;

[0047] Figure 5 A flowchart of step S400 in the message processing method based on user characteristics provided in an embodiment of the present invention;

[0048] Figure 6 A flowchart of step S220 in the message processing method based on user characteristics provided in an embodiment of the present invention;

[0049] Figure 7 A flowchart of step S120 in the message processing method based on user characteristics provided in an embodiment of the present invention;

[0050] Figure 8 A schematic diagram of functional modules of a message processing device based on user characteristics provided by an embodiment of the present invention;

[0051] Fig. 9 A schematic diagram of the hardware structure of a user feature-based message processing system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. The embodiments of the present invention are described below in conjunction with the accompanying drawings.

[0053] Message processing based on user characteristics is an efficient personalized marketing strategy that builds a refined user portrait by deeply analyzing user behavior, preferences, background and other characteristics. Using machine learning technology, this strategy can predict user preferences for different types of content and achieve accurate message push.

[0054] In the "Good Car Owner" message recommendation project, the application of this strategy has significantly improved user experience and interaction rate. By analyzing the user's driving habits, vehicle type, maintenance records and other data, the system can automatically push personalized vehicle maintenance tips, insurance services, driving safety advice and other information. This customized message not only increases the user's click-through rate, but also enhances the user's loyalty and stickiness to the "Good Car Owner" app. At the same time, the automated push process greatly reduces the cost of manual screening and sending messages, and improves operational efficiency. Taking into account business rules and user feedback, this strategy can be continuously optimized to ensure the relevance and timeliness of message push, providing users with a more intimate and valuable service experience.

[0055] In previous projects, only operations personnel were used to configure the message content pushed to users. Due to constraints such as time, region, and personnel costs, the push accuracy was low, and it was easy to push content that users were not interested in or even disgusted with, which not only wasted the operations personnel's time but also failed to achieve business results.

[0056] In order to solve the above problems, the present invention proposes a message processing method based on user characteristics, see Figure 1 , Figure 1A flowchart of an embodiment of the message processing method based on user features provided by the present invention. The message processing method based on user features provided in this embodiment is applied to a system consisting of a terminal device, a network and a server, wherein the network is a medium for providing a communication link between the terminal device and the server, which may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.; the operating system on the terminal device may include a handheld device operating system (iPhone operating system, iOS system), Android system or other operating system, and the terminal device is connected to the server through the network to achieve interaction, thereby performing operations such as receiving or sending data, and specifically may be various electronic devices with a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, portable computers and desktop servers, etc. Figure 1 As shown, the method specifically comprises the following steps:

[0057] S100, obtaining user feature information and business rule information;

[0058] S200, calculating the user's preferred service type according to the user characteristic information;

[0059] S300, calculating the recommended service weight according to the preferred service type and the business rule information;

[0060] S400: Generate and output a service message according to the recommended service weight.

[0061] In this embodiment, S100, obtaining user feature information and business rule information, that is, when a user operates on our platform and provides user feature information, our system will use this key information as an index to quickly retrieve and collect the user's detailed feature information in the huge user feature database. These feature information may include multi-dimensional data such as the user's personal profile, historical behavior data, preference settings, consumption patterns, etc. At the same time, the system will also refer to the business rule library to obtain specific business rule information related to the user, which includes but is not limited to the conditions for providing services, user stratification strategies, standards for participating in promotional activities, and customized pricing policies. By integrating these user characteristics and business rules, the system can accurately understand user needs and provide them with tailored services and content recommendations, thereby achieving personalized user experience and efficient business processes.

[0062] S200. Calculate the user's preferred service type based on the user characteristic information. Input the user information into the GBDT model, and obtain the score of the user's service type preference through the model calculation, so as to obtain the user's preferred service type. Among them, the service types include: ETC, refueling, annual inspection agency, charging, parking payment, maintenance, calling a designated driver, car repair, car rental, taxi and car washing, etc. The GBDT model, that is, the Gradient Boosting Decision Tree model, is a commonly used ensemble learning algorithm and an implementation method of the Gradient Boosting technology. GBDT makes predictions by constructing multiple decision trees. Each new decision tree is optimized on the basis of all previous trees to correct the residuals of the previous model, thereby improving the prediction accuracy of the overall model.

[0063] S300, calculating the recommended service weight according to the preferred service type and the business rule information. The preference score and the initialized business rule score are added to obtain the final recommended service weight. At the same time, in order to increase the scalability of users, a lower initial weight is given to the service type that does not hit the preference and business rule.

[0064] S400, generating and outputting a service message according to the recommendation service weight. Generating push content according to the recommendation service weight, writing the content to be recommended into a database for unified storage, and then sending the recommended content corresponding to the user group in batches at a fixed time.

[0065] The technical solution of the present application can accurately identify the user's preferred service type by analyzing a large amount of user data, and push relevant services accordingly, greatly improving the user experience and satisfaction. At the same time, the recommendation system must also take into account the set business rule information to ensure that the recommendation not only meets the user's needs, but also meets the company's business goals to achieve a win-win situation. The system collects multi-dimensional feature information of users, including basic information, historical behavior data, and real-time behavior data, and uses advanced data analysis technologies such as machine learning and deep learning algorithms to conduct in-depth analysis of user behavior patterns and preferences. For example, an e-commerce platform may analyze a user's browsing history, purchase history, and search habits to identify the categories of goods or services that the user may be interested in. Finally, combined with business rule information, the system can balance user needs with business goals. When generating recommendations, it is not only based on the user's personal preferences, but also takes into account factors such as the company's inventory management, sales strategy, and market positioning. For example, if a service is not the user's first choice, but the company currently has a promotion, the system will appropriately adjust the recommendation weight to achieve the promotion purpose, while also ensuring the relevance of the recommendation as much as possible to avoid user dissatisfaction.

[0066] Compared with the existing technology, it can use user characteristics to predict user preferences, improve the click-through rate of message push and reduce operating labor costs, effectively use user characteristic information to accurately predict user preferences, and also consider the policy rules set by the business, effectively improve the user's click-through rate of push messages, and save human resources. This can accurately push service content to users and improve user satisfaction.

[0067] In one embodiment, Figure 2 As shown, in step S100 of acquiring user feature information and business rule information, the method further includes:

[0068] S110, obtaining an identification ID input by a user;

[0069] S120: Acquire user feature information and corresponding business rule information in the feature table according to the identification ID.

[0070] In this embodiment, the identification ID refers to the unique identification of the user, and the corresponding user feature information and service setting rule information can be obtained in the feature table and rule engine table according to the user's aospsid (the unique identification of the user).

[0071] In one embodiment, Figure 3 As shown, in step S200, in calculating the user's preferred service type according to the user characteristic information, the steps further include:

[0072] S210, obtaining an initialization constant model;

[0073] S220, iteratively training the initialized constant model to obtain an updated constant model;

[0074] S230 , performing weighted sum on the updated constant models to obtain a final constant model.

[0075] In this embodiment, in step S210, an initial constant model is obtained: First, an initial machine learning model needs to be obtained as a starting point. For example, the constant model f0(x) can take the mean of the target value of the training data: Among them, L is the loss function, y i is the target value, γ is the initialization constant, and n is a natural number representing the number of decision trees.

[0076] In step S220, the initialization constant model is iteratively trained to obtain an updated constant model, and the initialization constant model is iteratively trained using a training data set. In each iteration, the prediction result of the model is calculated by forward propagation, and then compared with the actual label to calculate the error. Next, the parameters of the constant model are updated by back propagation to reduce the error. This process is repeated many times until the performance of the model reaches a satisfactory level or reaches a preset number of training rounds. The prediction results of the new decision tree are added to the initialization constant model to obtain the updated constant model. This model can be used for new data prediction tasks, or for further optimization and adjustment. The updated model can be saved to a file for subsequent loading and use.

[0077] In step S230, if Figure 4 As shown, the update constant model is weighted and summed by the following formula: Get the final constant model f(x).

[0078] In one embodiment, in step S300, the steps further include:

[0079] S310, combining the preferred service type with the business rule information, and when the user is identified as a mandatory push group according to the combination result, the service weight of the corresponding service preference type is set to 1, and the service weights of other service preference types are set to 0;

[0080] S320: If the service weight of the service preference type is 0, input the service preference type into a random function, select a pushed weight, and set the weight to the weight corresponding to the service preference type.

[0081] A random function usually refers to a function used to generate random numbers in programming. Different programming languages ​​may have different implementations and libraries to support random number generation, but their core functions are similar, that is, to generate a random value within a specific range.

[0082] In this embodiment, the user's service preferences and business rule information are combined to determine how to push services to the user. The following is a detailed description of this method:

[0083] Collect user preferences: The system first needs to collect the user's service type preferences. These preferences may come from direct user input, historical behavior analysis, or other user profiling technologies.

[0084] Define business rules: Business rules are a set of conditions that determine when a service should be pushed. These rules may be based on time, location, user status, historical interactions, and other factors.

[0085] Combining preferences and rules: The system combines the user's service preferences with the business rule information. This combination process may involve matching the user's preferences with the conditions in the rules.

[0086] Identify the forced push group: If the combined result shows that the user meets the conditions of a certain business rule, then the user is identified as a forced push group. This means that for this user, a certain type of push service is inevitable.

[0087] Set service weight: For users who are identified as mandatory push users, the service weight of the corresponding service preference type is set to 1, indicating that the service will be pushed first. The service weight of other service preference types that are not mandatory push is set to 0, indicating that these services will not be pushed.

[0088] Handling the situation where the rules are not hit: If the user's service preference type does not hit any business rules, the system will use a random function (such as the random function, the random generation function in Python, which can affect the random generation result by setting the weights of each part according to the weight size in the list) to decide whether to push the service and the weight of the push.

[0089] Select push weight: A random function selects a weight value based on a certain probability distribution (which may be uniform or weighted based on other factors). This weight value is then set as the weight corresponding to the service preference type.

[0090] Execute push: Finally, the system executes the push of services according to the set weights. Services with higher weights are more likely to be pushed to users.

[0091] This approach allows the system to ensure compliance with business rules while also being able to flexibly adjust service push policies based on user preferences and randomness. Doing so can improve user satisfaction while ensuring that important services can reach users in a timely manner.

[0092] In one embodiment, if Figure 5 As shown, in step S400, the steps further include:

[0093] S410, generating the corresponding service message according to the recommended service weight;

[0094] S420, using a distributed stream processing platform to set the sending time of the service message and select a sending group;

[0095] S430: Send the corresponding service message recommended contents to the sending group in batches according to the sending time.

[0096] In this embodiment, user data is collected: First, the system needs to collect user service preferences, historical behavior data, personal attributes and other related information. These data are the basis for generating recommended service weights.

[0097] Calculate service weights: Based on the method mentioned above, the system calculates the weights of each user for different service types. This includes combining user preferences with business rules, identifying mandatory push groups, and using random functions to assign weights to service preference types that do not match the rules.

[0098] Generate recommended content: For each user, the system generates personalized recommended content based on the calculated service weights. This may involve content selection, sorting, and formatting to ensure that the recommended content is relevant and attractive.

[0099] Through scheduled task settings: In the distributed stream processing platform, set up scheduled tasks to batch process the sending of recommended content. These tasks can be based on time (such as a specific time of day), events (such as user behavior), or other trigger conditions. Second, batch processing: When the scheduled task is triggered, the system will batch process the recommended content to be sent. This includes retrieving the recommended content from the database or message queue, and performing any necessary data processing. Third, distribute the recommended content: The processed recommended content is sent to the target user through the distributed stream processing platform. This process may be achieved through email, SMS, in-app notifications, or other communication channels. Fourth, monitoring and optimization: The system monitors the sending effect of recommended content, including key performance indicators such as user open rate, click rate, conversion rate, etc. Based on these feedbacks, the system continuously adjusts the recommendation algorithm and business rules to optimize the recommendation effect. Fifth, use logging: All sending activities, user interactions, and system responses are recorded in the log for subsequent analysis and auditing. Finally, through the feedback loop: User feedback on recommended content (such as clicks, likes, and shares) is collected and used to further refine user portraits and service weights, forming a continuous optimization cycle.

[0100] When the service message is generated, the service message can be sent to the corresponding sending group through the distributed stream processing platform. In other words, the corresponding service message can be selected for sending according to different sending groups, ensuring that the service messages received by different groups of people are what they need, thereby achieving the effect of precise push.

[0101] A distributed stream processing platform is a system that can process data streams in real time. It is deployed in large-scale clusters, abstracts the data processing process in the form of a directed acyclic graph, and reasonably distributes tasks to servers in the cluster through a scheduling algorithm. It can process millions of messages per second and is suitable for scenarios that require large-scale data processing. Through a distributed architecture, it can be easily expanded to thousands of nodes to cope with growing data traffic and workloads. This platform is scalable, real-time, and open, and can maximize cluster utilization, speed up data processing, improve overall system throughput, balance loads, and reduce resource usage.

[0102] For example, Apache Kafka is a typical distributed stream processing platform developed by the Apache Software Foundation for building real-time data pipelines and streaming applications. Kafka's architectural design supports high throughput, low latency, and scalability. It uses Apache ZooKeeper to manage cluster metadata and distributed consistency. Kafka version 3.0 introduced the KRaft mode to completely replace ZooKeeper567. In addition, Kafka's applicable scenarios include log collection, real-time data stream processing, message queues, and event-driven architectures.

[0103] In this way, the system can efficiently process and send a large amount of personalized recommendation content while ensuring the accuracy and timeliness of the recommendations. The use of a distributed stream processing platform makes this process highly scalable and reliable.

[0104] In one embodiment, Figure 6 As shown, step S220 iteratively trains the initialized constant model to obtain an updated constant model, including:

[0105] S221, calculating the residual under the initialization constant model for each user characteristic information sample;

[0106] S222, fitting the residual;

[0107] S223. Input the result of the residual fitting into the initialization constant model to obtain the updated constant model.

[0108] In this embodiment, step S221, for each sample, use the initialization constant model to make a prediction. Compare the prediction result with the actual label, calculate the error (residual), and store the residual of each sample. For each round of iteration, m = 1, 2, ..., M, for each user feature information sample, use the formula: Calculate the residual under the initialization constant model, where m is the number of iterations and f m-1(x) represents the initialization constant model, r im Indicates that the i-th sample is in the current model f m-1 (x i ), and represents the effect of L on f(x i ) Find the partial derivative; model f m-1 (x i ), which reflects the error of the model's current prediction.

[0109] In step S222, a new decision tree h is constructed. m (x) Fit the residual, here using the mean square error In the case of: r im =y i -f m-1 (x u ), where y represents the target value, y i represents the target value of the i-th sample.

[0110] In step S223, the prediction result of the new decision tree is added to the initialization constant model and multiplied by a learning rate η to control the step size to obtain the update constant model: m(x) =f m-1(x) +ηh m(x) . This model can be used for new data prediction tasks, or for further optimization and adjustment. The updated model can be saved to a file for subsequent loading and use. The new decision tree is fitted on the prediction error (i.e., residual) of all previous trees, thereby gradually reducing the prediction error of the overall model, effectively using user feature information to accurately predict user preferences, and also considering the policy rules set by the business, effectively improving the user's click rate on push messages and saving human resources, which can accurately push service content to users and improve user satisfaction.

[0111] In the field of machine learning and deep learning, h m (x) usually represents a function or model in a decision tree, where h can represent a specific decision tree, m represents the index or version number of the model, and x is the input feature. m (x) can be viewed as the prediction or decision function of the model m for the input x. This function maps the input feature x to an output, which can be a categorical label or a continuous value, depending on whether it is a classification or regression problem.

[0112] Decision trees organize data in a tree structure. Each internal node represents a decision rule on a feature attribute, and each leaf node represents the final prediction result. At each node of the decision tree, the algorithm selects a feature for segmentation, usually the feature that makes the data have the maximum information gain on this feature. Information gain is an indicator used by decision trees to evaluate the effect of feature segmentation. It is based on the concept of entropy and is used to measure the uncertainty of data. Decision trees are constructed recursively, and each segmentation generates new nodes until the stopping condition is met, such as reaching the maximum depth, all data points belong to the same category, or the information gain is less than a certain threshold. The decision tree model is easy to understand and explain because it shows the decision process from the root node to the leaf node in an intuitive tree diagram. Decision trees are prone to overfitting, especially when the tree becomes very deep and complex. To solve this problem, pruning techniques can be used to reduce the complexity of the tree.

[0113] Decision trees divide data into different branches through a series of questions, and finally achieve the purpose of predicting the output. Each internal node represents a decision rule on a feature, and the leaf node represents the final prediction result. In the context of decision trees, h m (x) can be understood as the decision or prediction made by model m for a given input x. In practical applications, h m (x) It can be used in various scenarios, including but not limited to classification, regression, anomaly detection, etc. Decision trees are widely used in the fields of data science and machine learning because of their easy understanding and interpretation.

[0114] In one embodiment, if Figure 7 As shown, in step S120, the step of acquiring user feature information and corresponding business rule information in the feature table according to the identification ID includes:

[0115] S121, identifying the identification ID input by the user;

[0116] S122, querying all data associated with the identification ID in the feature table to obtain the user feature information;

[0117] S123. Acquire business rule information corresponding to the user feature information in a rule engine table according to the user feature information.

[0118] In this embodiment, the identification ID input by the user is first received, which is usually a unique identification code, such as aopsid, and the identification ID input by the user is used as a query condition to perform a search in the feature table, which may contain multiple fields, such as user ID, name, age, gender, consumption history, preference settings, etc. The system retrieves all records associated with the identification ID, which constitute the user's feature information set, parses the query results, and extracts key feature information, such as the user's purchasing power, activity, loyalty, etc. The feature information may need to be further cleaned and converted to adapt to subsequent rule matching.

[0119] Design a set of logic to match user feature information with rules in the rule engine table. Rules may include conditional statements, decision trees, scoring systems, etc., to determine whether a user meets a specific rule. Based on the user feature information, query the rule engine table to find all applicable business rules. Through these steps, the system can efficiently process the identification ID entered by the user, query and apply relevant business rules, while ensuring user experience and data security.

[0120] In one embodiment, the business rule information includes situations where push notifications are forced to fail: gasoline vehicles and electric vehicles can only push refueling and charging, respectively, and anti-harassment situations: annual inspections, ETC, maintenance and other periodic services need to be pushed according to the rules given by the business.

[0121] In a complete embodiment, user information acquisition: the user enters his unique identification ID, such as a user name or user ID, through the interface. After the system receives the identification ID, it triggers the information retrieval process to retrieve the user feature information and business rule information associated with the identification ID from the feature table of the backend database.

[0122] Matching of feature information and business rules: User feature information includes static features (such as age, gender, occupation) and dynamic features (such as browsing history, purchase history, and activity participation) of users. Business rule information covers the conditions for providing services, priority, and promotional activity qualifications.

[0123] Calculation of preferred service types: The system initializes a recommendation model based on the user's historical data and behavior patterns.

[0124] Through iterative training, the system continuously optimizes the model parameters. Each iteration adjusts the model by calculating the residual. The residual formula is: Calculate the residual under the initialization constant model, where m is the number of iterations and f m-1 (x) represents the initialization constant model, r im Indicates that the i-th sample is in the current model f m-1 (x i ), and represents the effect of L on f(x i )Find the partial derivative.

[0125] By mean square error In the case of: r im =y i -f m-1 (x i ) Fit the residuals and input the residual fitting results into the model, and update the constant model formula as follows: m(x) =f m-1(x) +ηh m(x) Where η is the learning rate, h m(x) is the newly fitted decision tree.

[0126] Recommended service weight calculation: The system combines the user's preferred service type with business rules, identifies the mandatory push group, and sets the service weight of these users to 1.

[0127] For non-mandatory push groups, if the service preference type does not match the business rules, a weight is assigned to the service type through a random function. The formula is: w = random (0, 1). The weight w reflects the push priority of the service.

[0128] Service message generation and output: Based on the calculated recommendation service weights, the system generates personalized service messages. Using a distributed stream processing platform, the system can send recommended content to target user groups in batches at regular intervals.

[0129] Message push execution: The system continuously optimizes the recommendation model and business rules based on user feedback and push results to improve the relevance of message push and user satisfaction.

[0130] The technical solution of the present application can use user characteristics to predict and model user preferences, improve the click-through rate of message push and reduce operating labor costs, effectively use user characteristic information to accurately predict user preferences, and also take into account the policy rules set by the business, effectively improve the user's click-through rate on push messages, and save human resources, which can accurately push service content to users and improve user satisfaction.

[0131] Another embodiment of the present invention provides a message processing device based on user characteristics, such as Figure 7 As shown, the device 1 comprises:

[0132] The acquisition module 11 is used to acquire user characteristic information and business rule information. This module is responsible for collecting and organizing user characteristic information and business rule information. User characteristic information may include basic information of the user, historical behavior data, real-time behavior data, etc. Business rule information involves the internal regulations of the enterprise regarding service type, priority, availability, etc.;

[0133] The first processing module 12 is used to calculate the user's preferred service type based on the user characteristic information. This module analyzes the user characteristic information provided by the first module and identifies and predicts the user's preferred service type by applying statistical methods, machine learning algorithms or deep learning models. For example, based on the user's past purchase history and browsing habits, the system can infer that the user may be interested in a certain type of product or service;

[0134] The second processing module 13 is used to calculate the recommended service weight according to the preferred service type and the business rule information. Based on the user preferred service type output by the second module and the business rule information obtained by the first module, this module is responsible for calculating the recommendation weights of different services. This involves matching the user's preferences with the business rules, determining which services should be recommended first, and how to adjust the recommendation order to maximize user satisfaction and corporate benefits;

[0135] The third processing module 14 is used to generate and output a service message according to the recommended service weight. Using the recommended service weight calculated by the third module, this module is responsible for generating a specific service recommendation message. This includes determining the content, format, time and other elements of the message, and then pushing the service message to the user through appropriate channels (such as email, SMS, APP notification, etc.).

[0136] The module referred to in the present invention refers to a series of computer program instruction segments that can perform specific functions, which is more suitable for describing the message processing execution process based on user characteristics than a program. For the specific implementation of each module, please refer to the corresponding method embodiment above, which will not be repeated here.

[0137] In one embodiment, the device 1 further comprises:

[0138] A receiving module, used to obtain an identification ID input by a user;

[0139] A search module, used to obtain user feature information and corresponding business rule information in the feature table according to the identification ID;

[0140] In one embodiment, the first processing module 12 includes:

[0141] The acquisition unit is used to obtain a simple initialization constant model. The main task of this unit is to initialize the model of the recommendation system. This includes setting the initial parameters of the model, defining the model architecture, loading necessary preprocessed data, etc. This stage is a key step in establishing the foundation of the model to ensure that the model can be correctly trained and optimized in the future;

[0142] A training unit is used to iteratively train the initialization constant model to obtain an updated constant model. After obtaining the initialization constant model, the training unit trains the model by inputting actual data. Using various machine learning algorithms, such as gradient descent, random forest or deep learning, the model will be iteratively trained multiple times to optimize its parameters to better fit the training data. This process may involve cross-validation, hyperparameter adjustment and other techniques to avoid overfitting and improve the generalization ability of the model;

[0143] The synthesis unit is used to perform a weighted sum of the update constant models to obtain the final constant model. After the model training is completed, the synthesis unit is responsible for integrating these models, which usually involves weighted sum of the outputs of different models. The weighted sum can be determined based on the performance of the model on the validation data set, and the model with better performance will have a higher weight. The final result is a composite model that comprehensively considers the prediction results of multiple models, which usually has better prediction performance and stability.

[0144] In one embodiment, the second processing module 13 includes:

[0145] The judgment unit is used to combine the preferred service type with the business rule information. When the user is identified as a forced push group according to the combination result, the service weight of the corresponding service preference type is set to 1, and the service weights of other service preference types are 0. The core task of this unit is to combine the user's preferred service type with the business rule information for analysis. If the user is identified as a "forced push group" (for example, based on a specific marketing strategy or the user's special needs) according to the combination result, then for this part of the user, the system will set the service weight of the corresponding service preference type to 1, and set the service weights of all other service preference types to 0. This ensures that specific service messages can be pushed to the target user group first;

[0146] The regeneration unit is used to input the service preference type into a random function if the service preference type of the rule information is not matched, select the weight to be pushed, and set the weight to the weight corresponding to the service preference type. In the case where the business rule information is not matched, that is, the user's service preference type does not have a clear matching rule, the regeneration unit intervenes at this time. It inputs the service preference type into a random function to select the weight to be pushed. Then, it is set to the weight corresponding to the service preference type according to the output of the random function. This unit ensures that even in the absence of clear business rule guidance, the system can push service messages to users with a certain probability distribution.

[0147] By setting the judgment unit, the system can implement a forced push strategy for specific user groups to enhance marketing effects or meet specific user needs. The regeneration unit ensures that the system can still make effective recommendations when it cannot clearly match the business rules, increasing the diversity and exploratory nature of the recommendations. Overall, this design improves the adaptability and user satisfaction of the recommendation system while maintaining the flexibility of the business strategy.

[0148] In one embodiment, the third processing module 14 includes:

[0149] A residual calculation unit, used for calculating the residual of each user feature information sample under the initialization constant model;

[0150] A fitting unit, used for fitting the residual;

[0151] An output unit is used to input the result of the residual fitting into the initialization constant model to obtain the updated constant model.

[0152] In one embodiment, the search module includes:

[0153] An identification unit, used to identify the identification ID input by the user;

[0154] A query unit, used for querying all data associated with the identification ID in the feature table to obtain the user feature information;

[0155] The matching unit is used to obtain business rule information corresponding to the user feature information in a rule engine table according to the user feature information.

[0156] Another embodiment of the present invention provides a message processing system based on user characteristics, such as Figure 8 As shown, the system 10 includes:

[0157] One or more processors 110 and memory 120, Figure 6A processor 110 is used as an example for description. The processor 110 and the memory 120 may be connected via a bus or other means. Figure 6 The example of connecting through bus is taken in the following.

[0158] The processor 110 is used to complete various control logics of the system 10, and it can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), a single-chip microcomputer, an ARM (Acorn RISC Machine) or other programmable logic device, a discrete gate or transistor logic, a discrete hardware component, or any combination of these components. In addition, the processor 110 can also be any traditional processor, microprocessor or state machine. The processor 110 can also be implemented as a combination of computing devices, for example, a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors combined with a DSP and / or any other such configuration.

[0159] The memory 120, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions corresponding to the message processing method based on user characteristics in the embodiment of the present invention. The processor 110 executes various functional applications and data processing of the system 10 by running the non-volatile software programs, instructions and units stored in the memory 120, that is, implements the message processing method based on user characteristics in the above method embodiment.

[0160] One or more units are stored in the memory 120. When executed by one or more processors 110, the message processing method based on user characteristics in any of the above method embodiments is executed, for example, the following steps are executed:

[0161] Obtain user feature information and business rule information;

[0162] Calculating the user's preferred service type based on the user characteristic information;

[0163] Calculating a recommended service weight according to the preferred service type and the business rule information;

[0164] A service message is generated and outputted according to the recommended service weight.

[0165] The memory 120 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the system 10, etc. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the system 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0166] An embodiment of the present invention provides a non-volatile computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the message processing method based on user characteristics in any of the above method embodiments is executed, for example, the following steps are performed:

[0167] Obtain user feature information and business rule information;

[0168] Calculating the user's preferred service type based on the user characteristic information;

[0169] Calculating a recommended service weight according to the preferred service type and the business rule information;

[0170] A service message is generated and outputted according to the recommended service weight.

[0171] As an example, non-volatile storage media can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) as external cache memory. By way of illustration and not limitation, RAM can be obtained in many forms such as synchronous RAM (SRAM), dynamic RAM, (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink DRAM (SLDRAM), and direct Rambus RAM (DRRAM). The disclosed memory components or memories of the operating environment described herein are intended to include one or more of these and / or any other suitable types of memory.

[0172] In summary, the present invention discloses a message processing method, device, system and medium based on user characteristics, the method includes obtaining user characteristic information and business rule information; calculating the user's preferred service type according to the user characteristic information; calculating the recommended service weight according to the preferred service type and the business rule information; generating and outputting a service message according to the recommended service weight, the use of a distributed stream processing platform can efficiently process and send a large amount of personalized recommendation content, while ensuring the accuracy and timeliness of the recommendation, the new decision tree is fitted on the prediction error (i.e., residual) of all previous trees, thereby gradually reducing the prediction error of the overall model, and by accurately calculating and fitting the residual, the system can more accurately identify and correct errors in the prediction. The process of updating the constant model enables the system to gradually optimize its prediction performance, thereby providing higher quality recommendations, and this iterative learning method increases the self-improvement ability of the model, making it more powerful and adaptable.

[0173] According to the technical solution of the present invention, user characteristics can be used to predict user preferences, improve the click-through rate of message push and reduce operating labor costs, effectively use user characteristic information to accurately predict user preferences, and also consider the policy rules set by the business, effectively improve the user's click-through rate of push messages, and save human resources, which can accurately push service content to users and improve user satisfaction.

[0174] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium, and the computer program can include the processes of the above-mentioned method embodiments when executed. The storage medium can be a memory, a disk, a floppy disk, a flash memory, an optical storage device, etc.

[0175] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A message processing method based on user characteristics, characterized in that: include: Obtain user feature information and business rule information; Calculating the user's preferred service type based on the user characteristic information; Calculating a recommended service weight according to the preferred service type and the business rule information; A service message is generated and outputted according to the recommended service weight.

2. The message processing method based on user characteristics according to claim 1, characterized in that: The step of obtaining user feature information and business rule information includes: Get the identification ID entered by the user; The user feature information and corresponding business rule information are obtained in the feature table according to the identification ID.

3. The message processing method based on user characteristics according to claim 1, characterized in that: The step of calculating the user's preferred service type according to the user characteristic information comprises: Get the initialization constant model; Iteratively training the initialized constant model to obtain an updated constant model; The updated constant models are weighted summed to obtain a final constant model.

4. The message processing method based on user characteristics according to claim 1, characterized in that: The step of calculating the recommended service weight according to the preferred service type and the business rule information includes: The preferred service type is combined with the business rule information, and when the user is identified as a mandatory push group according to the combined result, the service weight of the corresponding service preference type is set to 1, and the service weights of other service preference types are set to 0; If the service weight of the service preference type is 0, the service preference type is input into a random function, a pushed weight is selected, and the weight is set to the weight corresponding to the service preference type.

5. The message processing method based on user characteristics according to claim 1, characterized in that: The step of generating and outputting a service message according to the recommended service weight comprises: Generate the corresponding service message according to the recommended service weight; Use a distributed stream processing platform to set the sending time of the service message and select the sending group; The corresponding service message recommended contents are sent to the sending group in batches according to the sending time.

6. The message processing method based on user characteristics according to claim 3 is characterized in that: The step of iteratively training the initialized constant model to obtain an updated constant model includes: Calculating the residual of each user feature information sample under the initialization constant model; fitting the residuals; The result of the residual fitting is input into the initialization constant model to obtain the update constant model.

7. The message processing method based on user characteristics according to claim 2, characterized in that: The step of acquiring user feature information and corresponding business rule information in the feature table according to the identification ID comprises: Identify the identification ID input by the user; Searching the feature table for all data associated with the identification ID to obtain the user feature information; According to the user characteristic information, business rule information corresponding to the user characteristic information is obtained in a rule engine table.

8. A message processing device based on user characteristics, characterized in that: include: An acquisition module is used to obtain user feature information and business rule information; A first processing module, configured to calculate the user's preferred service type based on the user characteristic information; A second processing module, configured to calculate a recommended service weight according to the preferred service type and the business rule information; The third processing module is used to generate and output a service message according to the recommended service weight.

9. A system for message processing based on user characteristics, characterized in that: The system includes at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the message processing method based on user characteristics according to any one of claims 1 to 7.

10. A non-volatile computer-readable storage medium, characterized in that: The non-volatile computer-readable storage medium stores computer-executable instructions. When the computer-executable instructions are executed by one or more processors, the one or more processors can execute the message processing method based on user characteristics as described in any one of claims 1-7.

Images