Middle-aged and elderly private domain user arriving method and device based on dynamic preference prediction

Abstract

The invention relates to the technical field of private domain reaching, and discloses a middle-aged and elderly private domain user reaching method and device based on dynamic preference prediction.The method comprises the steps that original behavior data and external environment data of a target user are obtained and preprocessed, and a structured time sequence data table of the target user is obtained; inputting the structured time sequence data table into a dynamic preference prediction model for prediction to obtain a touch response probability density curve of the target user in a future preset time period; obtaining to-be-reached content metadata, and generating a reaching task list of the target user in a future preset time period in combination with the reaching response probability density curve; and extracting the reaching content and the reaching time in the reaching task list, and pushing the reaching content to the target user according to the reaching time. According to the method, the response probability curve is generated based on the behavior and environment data, the user disgusting and shielding rate is reduced, the high response probability window is matched in combination with the content metadata, and the time and the content are synthesized, so that the pushed content is accurately reached in the user active period.

Description

Technical Field

[0001] This invention relates to the field of private domain outreach technology, specifically to a method and apparatus for reaching middle-aged and elderly private domain users based on dynamic preference prediction. Background Technology

[0002] In private domain traffic operations, direct communication with users through WeChat messages, community announcements, and mini-program push notifications is a key means to improve conversion and retention. However, blindly reaching out to the middle-aged and elderly, a specific group, can easily have negative effects. First, the middle-aged and elderly have relatively fixed schedules and are generally more sensitive to mobile phone notifications. Sending messages during work hours, lunch breaks, or late at night not only results in low open rates but may also be perceived as "digital harassment," leading users to block or even leave the community. Second, the online activity of middle-aged and elderly users is often strongly correlated with family life, health habits, and the broadcast time of specific media content, which is quite different from the fragmented and nighttime activity patterns of younger users. Middle-aged and elderly users are worried about missing important offers or health information but also do not want to be frequently disturbed by irrelevant information. Therefore, pushing appropriate content at the right time is crucial.

[0003] In existing private domain outreach technologies, most operational tools use fixed-time mass messaging, ignoring the activity patterns of individual users, resulting in poor effectiveness and a bad user experience. A few systems make simple inferences based on users' recent active time, but this method is lagging, cannot predict future activity windows, and does not consider the impact of external factors (such as holidays and weather). General models cannot capture the activity patterns unique to middle-aged and elderly users, which are dominated by their lifestyles, and the prediction results are often inaccurate. Timing prediction and consideration of the attributes of the content to be reached are isolated from each other, failing to achieve the optimal combination of "reaching users at the time when they are most likely to be interested in a certain type of content". Summary of the Invention

[0004] This invention provides a method and apparatus for reaching middle-aged and elderly private domain users based on dynamic preference prediction, in order to solve the problem of ignoring the individual differences of middle-aged and elderly users and the independent nature of time prediction and content recommendation, which leads to a low degree of matching between the timing of the reach and the recommended content.

[0005] In a first aspect, the present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction, the method comprising: Obtain the target user's raw behavioral data and external environment data, and preprocess the raw behavioral data and external environment data to obtain a structured time-series data table of the target user; The structured time series data table is input into the dynamic preference prediction model for prediction, and the reach response probability density curve of the target user in the future preset time period is obtained. Obtain metadata of the content to be reached, and generate a list of reach tasks for the target user in the future within a preset time period by combining the reach response probability density curve; Extract the content and time of the outreach task list, and push the content to the target user based on the time of the outreach.

[0006] The present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. Based on multi-dimensional data of target users, it dynamically predicts the life patterns of middle-aged and elderly users, takes into account individual differences, and solves the problem of blindly reaching and disturbing users. It generates response probability curves based on behavioral and environmental data, avoids sensitive periods such as rest periods, reduces user aversion and blocking rate, and combines content metadata to match high response probability windows. By comprehensively considering time and content, it enables the target user's attention to content to be accurately reached during the user's active period, thereby improving content click-through rate and conversion efficiency.

[0007] In one optional implementation, after pushing the content to the target user based on the time of contact, the method further includes: Acquire the target user's response behavior and use the outreach task and the target user's response behavior as feature samples of the target user; The dynamic preference prediction model is iteratively optimized using feature samples.

[0008] The present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. It learns from each reach result and continuously adapts to changes in the behavior patterns of middle-aged and elderly users, achieving true intelligent evolution. The reach feedback loop continuously iterates and optimizes the model. Long-term use can continuously enhance the matching degree of reach timing and content, which not only fits the habits of middle-aged and elderly users, but also helps to reduce costs and increase efficiency in private domain operations.

[0009] In one optional implementation, the raw behavioral data and external environment data are preprocessed to obtain a structured time-series data table of the target user, including: Semantic enhancement and response time back-calculation are performed on the raw behavioral data to add scene labels to the raw behavioral data; All raw behavioral data and external environment data are aligned in chronological order to construct a structured time-series data table containing timestamps.

[0010] The present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. By semantic enhancement to add scene tags and response time backtracking to complete the link, intuitive behaviors are transformed into scene information that fits the lives of middle-aged and elderly people, enabling subsequent models to accurately understand user behavior motivations and avoid misjudgment. By aligning behavior and environmental data by time, the messy raw behavior data and external environmental data are integrated into a structured sequence with timestamps, providing an accurate and complete input basis for subsequent prediction models and greatly improving the accuracy of reaching timing prediction.

[0011] In one optional implementation, the dynamic preference prediction model includes: a time-series feature extraction layer and a survival analysis prediction layer. A structured time-series data table is input into the dynamic preference prediction model for prediction, resulting in a probability density curve of the target user's reach response over a preset future time period, including: The structured time series data table is transformed into the input features of the dynamic preference prediction model, and the time series feature extraction layer is used to capture the long-term cross-period dependencies in the input features to obtain a context feature vector containing the user's cross-period patterns. The context feature vector is input into the survival analysis prediction layer to perform survival analysis and prediction, and the conditional risk function is obtained. The negative log-partial likelihood loss function is used to perform end-to-end joint training of the temporal feature extraction layer and the survival analysis prediction layer to minimize the deviation between the model prediction and the user's actual response behavior, thereby obtaining the model optimization parameters. Based on the model optimization parameters, the conditional risk function is transformed into a reach response probability density function. Based on the target user's current behavior data and external environment data, the reach response probability density function is used to generate the reach response probability density curve of the target user for a future preset time period.

[0012] The present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. By capturing cross-cycle patterns through a time-series feature extraction layer, it can accurately identify long-term behavioral patterns of middle-aged and elderly users, such as their focus on financial management after receiving pension payments, avoiding misjudgments based on a single time dimension. The survival analysis prediction layer adapts to the common situation of middle-aged and elderly users not responding, learning the rationality of timing from non-clicks, rather than simply judging them as not interested. End-to-end joint training further reduces prediction bias, and the final generated probability density curve can clearly show the user's future activity window, which not only fits their daily routines but also makes the selection of subsequent outreach timing more reasonable, significantly improving message response rate.

[0013] In one alternative implementation, the structured time-series data table is transformed into input features for a dynamic preference prediction model, including: Based on the timestamps in the structured time-series data table, the corresponding behavioral data or external environment data are transformed into periodic features; The semantic information in the periodic features is embedded and converted into dense vectors. The periodic features and dense vectors together form the input features of the dynamic preference prediction model.

[0014] The present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. It transforms timestamps into periodic features, which not only cover the general daily / weekly cycle, but also implicitly fit the fixed routines of middle-aged and elderly people. By embedding processing, semantic information is transformed into dense vectors, allowing scene tags to be understood and associated by the model. The input features formed by the combination of the two not only retain the temporal pattern of user behavior, but also incorporate the semantic information of the life scenes of middle-aged and elderly people, effectively improving the accuracy of subsequent predictions.

[0015] In one optional implementation, metadata of the content to be reached is obtained, and a list of reach tasks for the target user over a future preset time period is generated by combining the reach response probability density curve, including: Based on the preset operation rule base and the first preset probability threshold, the effective time window within the future preset time period is selected. The effective time window is used to characterize the time period within the future preset time period that meets the global constraints in the operation rule base and has a probability density not lower than the first preset probability threshold. Determine the content type of the metadata of the content to be reached, and based on the content type, filter the optimized time window for the reaching task from the effective time window.

[0016] In one alternative implementation, optimizing the time window for reaching tasks is selected from the effective time window based on content type, including: If the content type is regular content, then the time point with the highest probability density is selected as the optimization time window for regular content based on the reach response probability density curve. If the content type is time-sensitive, a time value decay function is introduced. Based on the time value decay function and the reach response probability density curve, a comprehensive evaluation is conducted to determine the optimal time window for the time-sensitive content. If the content type is care content, the optimization time window for care content is determined based on the preset operation rule base and the second preset probability threshold, where the second preset probability threshold is less than the first preset probability threshold.

[0017] The present invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. By pre-setting a rule base and probability thresholds to filter effective time windows, it avoids sensitive time periods for middle-aged and elderly users such as late nights and lunch breaks, and locks in high response probability intervals, reducing invalid outreach from the source. Furthermore, by combining content type matching to optimize time windows, regular content is selected based on probability peaks, time-sensitive content takes into account urgency, and caring content is adapted to comfortable time periods. This not only conforms to the daily routines of middle-aged and elderly people, but also achieves precise matching of content and timing, improves user experience, and significantly improves the conversion efficiency of private domain outreach.

[0018] Secondly, this invention provides a device for reaching middle-aged and elderly private domain users based on dynamic preference prediction. The device includes: The multidimensional data acquisition module is used to acquire the target user's raw behavioral data and external environment data, and to preprocess the raw behavioral data and external environment data to obtain the target user's structured time-series data table; The dynamic preference prediction module is used to input structured time series data tables into the dynamic preference prediction model for prediction, and obtain the reach response probability density curve of the target user in the future preset time period. The outreach timing arbitrator is used to obtain metadata of the content to be reached and combine it with the outreach response probability density curve to generate a list of outreach tasks for the target user in the future within a preset time period. The strategy execution and feedback module is used to extract the content and time of the outreach task list, and push the content to the target user based on the time of the outreach.

[0019] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method described in the first aspect or any corresponding embodiment thereof.

[0020] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0021] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of an application scenario according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the first process of the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the second process of the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to an embodiment of the present invention; Figure 4 This is a logical diagram of the arbiter for the timing of reaching middle-aged and elderly private domain users in the method based on dynamic preference prediction according to an embodiment of the present invention. Figure 5This is a complete flowchart of a specific embodiment of the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to an embodiment of the present invention; Figure 6 This is a structural block diagram of a device for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to an embodiment of the present invention; Figure 7 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0025] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0026] As an optional application scenario of this invention, such as Figure 1 As shown, this system for reaching private domain users of middle-aged and elderly people may include at least one terminal device and at least one server. Figure 1 The system is illustrated in the example, which includes a computer 101, a mobile terminal 102, and a server 103, and the terminal devices such as the computer 101 and the mobile terminal 102 are connected to the server 103 through a network 110.

[0027] Specifically, the terminal device can be a smartphone, tablet, laptop, PDA, desktop computer, game console, smart TV, smart wearable device, in-vehicle terminal, VR (Virtual Reality) device, AR (Augmented Reality) device, etc. Server 103 can be a standalone physical server, a server cluster, a distributed system, or a cloud server providing cloud services. Network 110 can be a wired or wireless network, examples of which include, but are not limited to, the Internet, corporate intranet, local area network, wide area network, mobile communication network, and combinations thereof.

[0028] This invention provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction. By dynamically predicting the probability distribution curve of each middle-aged and elderly private domain user's most likely message viewing and strongest willingness in the future, the method accurately locates the timing of reaching them, thereby achieving the effect of accurately locating the timing of reaching different content and improving message click-through rates.

[0029] According to an embodiment of the present invention, an embodiment of a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0030] This embodiment provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction, which can be used in the aforementioned computer system. Figure 2 This is a flowchart of a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to an embodiment of the present invention, as follows: Figure 2 As shown, the process includes the following steps: Step S201: Obtain the target user's original behavioral data and external environment data, and preprocess the original behavioral data and external environment data to obtain the target user's structured time series data table.

[0031] Specifically, the target users include middle-aged and elderly users. Raw data is collected from the private domain ecosystems of these users and constructed into a structured sequence aligned with the timeline for model consumption. Raw behavioral data is collected in real-time or near real-time from platforms such as WeChat Work, social groups, and mini-programs through Application Programming Interfaces (APIs). This raw behavioral data includes, but is not limited to: communication logs: the time, type (text / image / link), and parties involved in the conversation (customer service / group chat / friends) of messages sent / received by users; content interaction logs: actions and timestamps of clicking product links, watching live streams, participating in group activities, posting comments, and sharing articles; and mini-program behavior logs: events such as login, page browsing time, time spent on product details pages, and completion of check-in.

[0032] External environmental data can be obtained from publicly available data services, including but not limited to: weather data (temperature, precipitation, air quality, etc.), statutory holidays, weekday / weekend markings, and major local events (such as community activities and festival celebrations).

[0033] The raw behavioral data and external environment data are preprocessed in a structured manner to obtain a structured time series data table, which includes fields including but not limited to: timestamp, behavior type (semantic label), behavior metadata, weather, holiday markers, etc. This time series data table serves as the direct input for the subsequent dynamic preference prediction model.

[0034] Step S202: Input the structured time series data table into the dynamic preference prediction model for prediction to obtain the reach response probability density curve of the target user in the future preset time period.

[0035] Specifically, based on the structured behavioral sequence constructed from the preceding modules, a dynamic preference prediction model is used to accurately predict the time window during which middle-aged and elderly users are willing to respond to private domain messages within a preset time period (e.g., 24-48 hours). The dynamic preference prediction model adopts a hybrid architecture of "temporal feature extraction + survival analysis prediction," which is specifically adapted to the periodicity and delayed response characteristics of middle-aged and elderly users' behavior.

[0036] It needs to capture users' long-term behavioral patterns (such as their financial concerns after receiving monthly pension payments) and learn from unresponsive data. Ultimately, it should output an intuitive and usable reach response probability density curve. This curve can clearly show the active time window of the target users, such as the typical multi-peak curve for middle-aged and elderly users (one probability peak each in the morning, afternoon, and evening).

[0037] Step S203: Obtain the metadata of the content to be reached, and generate a list of reach tasks for the target user in the future preset time period by combining the reach response probability density curve.

[0038] Specifically, the system obtains metadata about the content to be reached through a timing arbitrator, and generates a list of outreach tasks for the target user within a preset time period by combining this metadata with the outreach response probability density curve. The metadata includes: content type (promotion / care / information), urgency level, and target user group. Combined with the outreach response probability density curve, a structured outreach task list for the target user within a preset time period is determined. This list includes: user identifier (ID), recommended outreach timestamp, estimated response probability, and decision-making basis.

[0039] Step S204: Extract the reach content and reach time from the reach task list, and push the reach content to the target user according to the reach time.

[0040] Specifically, based on the task list, the tasks are sorted by time and precisely scheduled using a message middleware (such as RabbitMQ) to send messages by calling the push platform API at the specified time. The target user's content to be pushed and the corresponding recommended delivery time are extracted. At the time corresponding to the recommended delivery time, the content is sent to the target user through private channels (such as WeChat Work or social groups), ensuring that the push content reaches the user during periods of high user response.

[0041] The method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment dynamically predicts the life patterns of middle-aged and elderly users based on multi-dimensional data of target users, takes into account individual differences, and solves the problem of blindly reaching out and disturbing users. It generates response probability curves based on behavioral and environmental data, avoids sensitive periods such as rest periods, reduces user aversion and blocking rate, and combines content metadata to match high response probability windows. By comprehensively considering time and content, it enables the target user's attention to content to be accurately reached during the user's active period, thereby improving content click-through rate and conversion efficiency.

[0042] In some optional implementations, after pushing the content to the target user based on the time of contact, the method further includes: Step S205: Obtain the response behavior of the target user and use the outreach task and the target user's response behavior as feature samples of the target user.

[0043] Specifically, it tracks user behavior in real time after being reached. If the channel supports read receipts, it records whether the message has been read. It also tracks whether the user has clicked on a link, participated in an activity, or completed a conversion within a preset time after being reached (adapted to the delayed response characteristics of middle-aged and elderly users, usually set to 1-24 hours).

[0044] The key information of this outreach (such as the time of outreach and the type of content) is used as features, and the user's feedback behavior (such as read, click, or no response) is used as tags to form a labeled feature sample. At the same time, time window matching is used to solve the correlation problem of delayed response among middle-aged and elderly people (for example, if a user clicks 3 hours after the outreach, the system will associate the click behavior with the original outreach action).

[0045] In addition to the tagged feedback data that serves as feature samples, a delivery execution status report is output synchronously, recording basic information such as whether the delivery was successful and the push time.

[0046] Step S206: Iteratively optimize the dynamic preference prediction model using feature samples.

[0047] Specifically, labeled feedback data is used as feature samples, and incremental training is selected. By continuously learning the latest behavioral patterns of target users (such as adjustments to daily routines and changes in interests), the dynamic preference prediction model is continuously adapted to the changes in the behavior of middle-aged and elderly users, thereby achieving continuous optimization and self-adaptation of the model and gradually improving the matching accuracy of the timing of outreach and content.

[0048] A complete closed-loop data system of prediction, arbitration, enforcement, and feedback was established. In particular, a time window matching mechanism was designed to address the delayed response characteristics of middle-aged and elderly users, ensuring the correct association between positive and negative samples.

[0049] The method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment learns from each reach result and continuously adapts to changes in the behavior patterns of middle-aged and elderly users, achieving true intelligent evolution. The reach feedback loop continuously iterates and optimizes the model. Long-term use can continuously enhance the matching degree between reach timing and content, which not only fits the habits of middle-aged and elderly users, but also helps to reduce costs and increase efficiency in private domain operations.

[0050] This embodiment provides a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction, which can be used in the aforementioned computer system. Figure 3 This is a flowchart of a method for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to an embodiment of the present invention, as follows: Figure 3 As shown, the process includes the following steps: Step S301: Obtain the target user's original behavioral data and external environment data, and preprocess the original behavioral data and external environment data to obtain the target user's structured time series data table.

[0051] Specifically, step S301 includes: Step S3011: Perform semantic enhancement and response time back-calculation on the original behavioral data to add scene labels to the original behavioral data.

[0052] Specifically, semantic enhancement is performed on the raw behavioral data to add scene labels to the raw behaviors. For example, the raw behavioral data "sending a good morning emoji in the 'Happy Family' group at 7:00 am" is labeled as "behavior: family social morning ritual"; the raw behavioral data "clicking on the pushed news after the news broadcast" is labeled as "behavior: content consumption during a fixed media period", providing high-level semantics for the model to understand the motivations behind the behaviors.

[0053] Many private messaging services (such as WeChat Work and group push notifications) currently lack "read receipts." When you send a message, you have no idea whether the recipient didn't see it or saw it but ignored it. This means the system only knows if a user clicked a link, but not when they saw the message and clicked it, making it impossible to accurately record the time difference between message sending and user viewing. For example, if a message with a product link is sent at 10:00 AM, and the user clicks the link at 10:30 AM, the system will automatically treat 10:30 AM (the time the user clicked the link) as the time the user saw the message, thus completing the message's reading time record. What were originally fragmented data points of "message sending time" and "user click time" become a complete response sequence: "Message sent at 10:00 AM → User viewed at 10:30 AM → User clicked the link at 10:30 AM."

[0054] Step S3012: Align all raw behavioral data and external environment data in chronological order to construct a structured time-series data table containing timestamps.

[0055] Specifically, taking users as units, all raw behavioral data and external environmental data are aligned in chronological order. All behavioral data of the target user (communication, content interaction, mini-program operation) and corresponding external environmental data (weather at the time, whether it is a holiday) are matched one by one along the time axis. At the same time, a uniform granularity (30 minutes or 1 hour) is set, such as the time period "9:00-9:30". The following are also recorded: the user clicked the health link, it was cloudy at the time, and it was a Wednesday, a weekday, to form structured time segment data. Multiple time segment data are combined to form a structured time series data table.

[0056] Each user has a unique structured time-series data table with core fields including: timestamp (accurate to 30 minutes / hour), behavior type (semantic tags, such as family social morning rituals), behavior metadata (such as the type of link clicked, browsing duration), weather conditions, holiday / workday markers, local event markers, etc., which fully restores the user's behavioral logic at different times and in different scenarios.

[0057] This involves semantically fusing external factors such as weather and holidays with native user behavior. By enhancing the semantics of behavior among middle-aged and elderly users, low-level behavioral data is elevated to features with life-scenario significance, enabling the model to understand the motivations behind behavior and predict pattern changes in specific situations.

[0058] The method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment uses semantic enhancement to add scene tags and response time to backtrack and complete the link, transforming intuitive behaviors into scene information that fits the lives of middle-aged and elderly people. This allows subsequent models to accurately understand user behavior motivations and avoid misjudgments. By aligning behavior and environmental data by time, the messy raw behavior data and external environmental data are integrated into a structured sequence with timestamps, providing an accurate and complete input basis for subsequent prediction models and greatly improving the accuracy of reaching timing prediction.

[0059] Step S302: Input the structured time series data table into the dynamic preference prediction model for prediction to obtain the reach response probability density curve of the target user in the future preset time period.

[0060] Specifically, the dynamic preference prediction model includes a time series feature extraction layer and a survival analysis prediction layer. This hybrid architecture, which deeply couples time series modeling with survival analysis, is key to simultaneously handling periodic patterns and right-censored data.

[0061] Step S302 above includes: Step S3021: The structured time series data table is transformed into the input features of the dynamic preference prediction model, and the time series feature extraction layer is used to capture the long-term cross-period dependencies in the input features to obtain a context feature vector containing the user's cross-period patterns.

[0062] Specifically, structured time-series data tables (containing information such as user behavior, time, and environment) are processed into input features that can be recognized by the dynamic preference prediction model. The dynamic preference prediction model includes a time-series feature extraction layer and a survival analysis prediction layer. After the input features are fed into the dynamic preference prediction model, the self-attention mechanism of the time-series feature extraction layer (usually a Transformer encoder) is used to mine long-term patterns of user behavior. For example, it can identify periodic behavioral associations across days / weeks / months, such as "users pay significantly more attention to financial management content 3 days after the monthly pension payment date" and "users are more likely to browse health information on Sunday afternoons." The captured cross-period patterns are integrated into a contextual feature vector, which serves as the core input for subsequent predictions.

[0063] In an optional implementation, step S3021 above, which transforms the structured time-series data table into input features for a dynamic preference prediction model, includes: Step a1: Based on the timestamps in the structured time series data table, convert the corresponding behavioral data or external environment data into periodic features.

[0064] Specifically, the timestamps in the structured time-series data table are broken down into two types of features: general periodic features and features specific to middle-aged and elderly people. The general periodic features refer to extracting "hours within a day" (e.g., 7 am, 7 pm) and "days of the week" (e.g., Tuesday, Sunday) to capture the user's daily and weekly cycle patterns. The features specific to middle-aged and elderly people refer to additionally labeling "morning exercise time (6-7 am)," "lunch break time (12-2 pm)," and "family time (6-8 pm)" and other life marker times to directly align with the daily routines of middle-aged and elderly people and reduce the model learning cost.

[0065] Step a2 involves embedding the semantic information in the periodic features and converting it into a dense vector. The periodic features and the dense vector together form the input features of the dynamic preference prediction model.

[0066] Specifically, for textual features that cannot be directly calculated, such as "behavioral semantic tags (e.g., 'using mini-programs at home on a rainy day')," "weather (e.g., 'heavy rain')," and "holidays (e.g., 'Double Ninth Festival')," embedding technology is used to transform them into a continuous string of numerical vectors (dense vectors).

[0067] The method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment transforms timestamps into periodic features, which not only cover the general daily / weekly cycle but also implicitly align with the fixed routines of middle-aged and elderly people. By embedding processing, semantic information is transformed into dense vectors, allowing scene labels to be understood and associated by the model. The input features formed by the combination of the two not only retain the temporal patterns of user behavior but also incorporate the semantic information of the life scenarios of middle-aged and elderly people, effectively improving the accuracy of subsequent predictions.

[0068] Step S3022: Input the context feature vector into the survival analysis prediction layer to perform survival analysis prediction and obtain the conditional risk function.

[0069] Specifically, the contextual feature vector is input into the survival analysis prediction layer (using a parameterized DeepSurv network) to learn the conditional risk function λ(t|X), where t represents the time since the last contact, and X represents the user's contextual features. The conditional risk function simulates the recovery of a user's willingness to respond to a message over time t since the last contact, within a specific user context. This architecture is naturally well-suited for handling the large amount of "no-response (right censoring)" data present among middle-aged and elderly users; that is, the model can learn from the fact that "the user received a message but did not click on it" to determine that the message may be inappropriate or that the user is temporarily uninterested.

[0070] By deeply coupling the time-series modeling capabilities of Transformer with the censored data processing capabilities of survival analysis at the model level, it can not only capture the regular behavioral patterns of middle-aged and elderly users across cycles, but also effectively handle the large amount of non-response data and accurately predict the recovery curve of response intention, thus solving the problems of strong lag and inability to handle uncertainty in traditional methods.

[0071] Step S3023: Use the negative log-partial likelihood loss function to perform end-to-end joint training of the temporal feature extraction layer and the survival analysis prediction layer to minimize the deviation between the model prediction value and the user's actual response behavior, and obtain the model optimization parameters.

[0072] Specifically, by using the negative log-partial likelihood loss function, the temporal feature extraction layer and the survival analysis prediction layer are jointly trained end-to-end. This ensures that the goals of the two networks remain consistent (accurately predicting users' willingness to respond). By minimizing the deviation between the model's predicted values ​​and the actual user response behavior, the parameters of the two networks are continuously adjusted, ultimately yielding optimized model parameters that are suitable for the behavior of middle-aged and elderly users.

[0073] Step S3024: Based on the model optimization parameters, the conditional risk function is converted into a reach response probability density function, and based on the target user's current behavior data and external environment data, the reach response probability density function is used to generate the target user's reach response probability density curve for a future preset time period.

[0074] Specifically, based on the model optimization parameters, the abstract conditional risk function is transformed into a reach response probability density function. Combining the target user's current behavioral data and external environmental data (e.g., current weather, holiday status), the probability density function is used to calculate the user's response probability at each time point within a preset future time period (typically 24-48 hours), ultimately generating a reach response probability density curve PDF(t). This curve clearly shows the user's activity window (e.g., three probability peaks commonly seen among middle-aged and elderly users: "after morning exercise," "after lunch break," and "after evening family time"), providing a basis for selecting subsequent reach timing.

[0075] The method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment captures cross-cycle patterns through a time-series feature extraction layer, which can accurately identify long-term behavioral patterns of middle-aged and elderly users, such as their focus on financial management after receiving pension payments, avoiding misjudgments based on a single time dimension. The survival analysis prediction layer adapts to the common situation of middle-aged and elderly users not responding, learning the rationality of timing from non-clicks, rather than simply judging them as not interested. End-to-end joint training further reduces prediction bias, and the final generated probability density curve can clearly show the user's future activity window, which not only fits their work and rest habits, but also makes the selection of subsequent outreach timing more reasonable, significantly improving the message response rate.

[0076] Step S303: Obtain the metadata of the content to be reached, and generate a list of reach tasks for the target user in the future preset time period by combining the reach response probability density curve.

[0077] Specifically, step S303 includes: Step S3031: Based on the preset operation rule base and the first preset probability threshold, filter the effective time window within the future preset time period. The effective time window is used to characterize the time period within the future preset time period that meets the global constraints in the operation rule base and has a probability density not lower than the first preset probability threshold.

[0078] Specifically, the operation rule base contains predefined global constraints, such as prohibited contact periods (e.g., 22:00-7:00 the next day) and the maximum contact frequency per user. Based on the preset operation rule base and the first preset probability threshold, combined with the contact response probability density curve, invalid time periods with probability densities lower than the first preset probability threshold (e.g., 0.5) and that do not meet the global constraints are filtered out, thus obtaining the valid time periods.

[0079] Step S3032: Determine the content type of the metadata of the content to be reached, and based on the content type, filter the optimized time window for the reaching task from the effective time window.

[0080] Specifically, the content type of the metadata to be reached is determined by combining the behavioral characteristics, content value attributes, and operational goals of middle-aged and elderly private domain users.

[0081] The optimal time window for reaching a task is the best time to reach it, which can be achieved through a reach timing arbitrator, such as... Figure 4 The diagram shown is a logical schematic of the outreach timing arbitrator, which ultimately yields a structured outreach task list.

[0082] In some optional implementations, step S3032 above includes: Step b1: If the content type is regular content, then select the time point with the highest probability density as the optimization time window for regular content based on the reach response probability density curve.

[0083] Specifically, for regular content, the time point with the highest probability density is selected within the remaining valid time window to maximize the open rate.

[0084] Step b2: If the content type is time-sensitive content, a time value decay function is introduced. Based on the time value decay function and the reach response probability density curve, a comprehensive evaluation is conducted to determine the optimal time window for the time-sensitive content.

[0085] Specifically, for time-sensitive content (such as limited-time flash sales), a time value decay function is introduced to balance probability and timeliness, and a window with a slightly higher but not the highest probability, but an earlier time, may be selected.

[0086] Step b3: If the content type is care content, then the optimization time window for care content is determined based on the preset operation rule base and the second preset probability threshold. The second preset probability threshold is less than the first preset probability threshold.

[0087] Specifically, for specific care content for middle-aged and elderly people (such as health reminders), the probability threshold can be appropriately relaxed, and the first preset probability threshold can be adjusted to the second preset probability threshold, but it must be strictly ensured that it is done during a time when the user is comfortable (such as during the day) to avoid disturbing the user.

[0088] A business-aware intelligent arbitrator was built to transform pure probabilistic predictions into actionable operational decisions. By introducing mechanisms such as a time value decay function and a comfortable time slot guarantee, a dynamic balance is achieved among multiple objectives, including maximizing open rates, ensuring timeliness, and maintaining user experience, thus meeting the complex needs of private domain operations for middle-aged and elderly users.

[0089] The method for reaching middle-aged and elderly users in private domains based on dynamic preference prediction provided in this embodiment filters effective time windows through a preset rule base and probability thresholds. This avoids sensitive time periods for middle-aged and elderly users, such as late nights and lunch breaks, while locking in high response probability intervals, thus reducing invalid outreach from the source. Furthermore, by combining content type matching with optimized time windows, regular content is selected based on probability peaks, time-sensitive content takes into account urgency, and caring content is adapted to comfortable time periods. This not only aligns with the daily routines of middle-aged and elderly users but also achieves precise content-timing matching, improving user experience and significantly increasing the conversion efficiency of private domain outreach.

[0090] Step S304: Extract the reach content and reach time from the reach task list, and push the reach content to the target user based on the reach time. For details, please refer to [link to relevant documentation]. Figure 2 Step S203 of the illustrated embodiment will not be described again here.

[0091] In one specific embodiment, taking "private domain operation of a certain health product for middle-aged and elderly people" as an example, such as... Figure 5 The diagram shown is a complete flowchart of the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment. The specific reaching process includes: (1) Collect multi-source data: Obtain the original behavioral data of 55-year-old user Aunt Zhang from WeChat and health mini-program (e.g., sending good morning emojis in the family group at 7:00 every day and browsing the blood glucose monitor details page at 19:30), and at the same time obtain external environmental data (e.g., light rain today, Wednesday is a workday).

[0092] (2) Data cleaning and semantic enhancement: “Send a good morning emoji at 7:00” is labeled as “family social morning ritual”, and the reading time of the previous product message is deduced to be 19:30 by “clicking the product link at 19:30”.

[0093] (3) Constructing a time-series behavior sequence: Align the data such as “19:00-19:30: browsing blood glucose meter + rain + Wednesday” in 30-minute granularity to generate Aunt Zhang’s structured time-series table.

[0094] (4) Temporal feature extraction (Transformer): The model captures Aunt Zhang's cross-week pattern of "frequently browsing health products every Wednesday from 19:00 to 20:00 (after dinner)".

[0095] (5) Survival analysis prediction (DeepSurv): Input Aunt Zhang's time series characteristics and learn the conditional risk function that "3 days after the last push of health products, the willingness to respond recovers to 80%".

[0096] (6) Output probability density curve: Generate Aunt Zhang's response curve for the next 48 hours, showing that the response probability is 75% from 19:00 to 19:30 (after dinner).

[0097] (7) Candidate time window screening: Combining the operation rule base (prohibiting access after 22:00) and the probability threshold (≥60%), two effective time windows, "today 19:00-19:30" and "tomorrow 8:00-8:30", are selected.

[0098] (8) Multi-objective optimization decision: The content to be reached is "blood glucose meter promotion" (promotional category), and the time window with the highest probability "today 19:00-19:30" is selected as the optimization time window.

[0099] (9) Message delivery: At 19:00, push the message "Blood glucose meter 20% off for a limited time" to Aunt Zhang.

[0100] (10) Feedback monitoring: Record Aunt Zhang’s actions of clicking the link at 19:10 and placing the order at 19:15.

[0101] (11) Closed-loop learning and model optimization: "Push promotional content at 19:00 → Place order at 19:15" is used as a positive sample backflow to update the model parameters. Subsequently, for similar users, health product promotions will be pushed first after dinner.

[0102] This embodiment also provides a device for reaching middle-aged and elderly private domain users based on dynamic preference prediction. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0103] This embodiment provides a device for reaching middle-aged and elderly private domain users based on dynamic preference prediction, such as... Figure 6 As shown, it includes: The multidimensional data acquisition module 601 is used to acquire the target user's original behavioral data and external environment data, and to preprocess the original behavioral data and external environment data to obtain the target user's structured time-series data table.

[0104] The dynamic preference prediction module 602 is used to input the structured time series data table into the dynamic preference prediction model for prediction, and obtain the reach response probability density curve of the target user in the future preset time period.

[0105] The timing arbitrator 603 is used to obtain metadata of the content to be reached and generate a list of reach tasks for the target user in the future within a preset time period by combining the reach response probability density curve.

[0106] The strategy execution and feedback module 604 is used to extract the reach content and reach time from the reach task list, and push the reach content to the target user according to the reach time.

[0107] The device for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in this embodiment of the invention can execute the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method. Further functional descriptions of the above modules and units are the same as in the corresponding embodiments described above, and will not be repeated here.

[0108] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0109] The following is a detailed reference. Figure 7This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 701, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 702 or a program loaded from memory 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the electronic device. The processor 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0110] Typically, the following devices can be connected to I / O interface 705: input devices 706 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 707 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 708 including, for example, magnetic tapes, hard disks, etc.; and communication devices 709. Communication device 709 allows electronic devices to exchange data via wireless or wired communication with other devices. Although Figure 7 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0111] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 709, or installed from a memory 708, or installed from a ROM 702. When the computer program is executed by the processor 701, it performs the functions defined in the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction according to embodiments of the present invention.

[0112] Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0113] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as recordable on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the method for reaching middle-aged and elderly private domain users based on dynamic preference prediction shown in the above embodiments is implemented.

[0114] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for reaching middle-aged and elderly private domain users based on dynamic preference prediction, characterized in that, The method includes: Obtain the target user's raw behavioral data and external environment data, and preprocess the raw behavioral data and external environment data to obtain a structured time-series data table of the target user; The structured time-series data table is input into the dynamic preference prediction model for prediction, and the reach response probability density curve of the target user in the future preset time period is obtained. Obtain the metadata of the content to be reached, and generate a list of reach tasks for the target user in the future within a preset time period by combining the reach response probability density curve; Extract the content and time of the outreach from the outreach task list, and push the content to the target user based on the outreach time.

2. The method according to claim 1, characterized in that, After pushing the content to the target user based on the reach time, the method further includes: Acquire the target user's response behavior, and use the outreach task and the target user's response behavior as feature samples of the target user; The dynamic preference prediction model is iteratively optimized using the feature samples.

3. The method according to claim 1, characterized in that, The raw behavioral data and external environment data are preprocessed to obtain a structured time-series data table of the target user, including: The original behavioral data is subjected to semantic enhancement and response time back-calculation processing to add scene labels to the original behavioral data; All raw behavioral data and external environment data are aligned in chronological order to construct a structured time-series data table containing timestamps.

4. The method according to claim 1, characterized in that, The dynamic preference prediction model includes: a time-series feature extraction layer and a survival analysis prediction layer. The structured time-series data table is input into the dynamic preference prediction model for prediction, resulting in the reach response probability density curve of the target user over a preset future time period, including: The structured time-series data table is transformed into input features for a dynamic preference prediction model, and the time-series feature extraction layer is used to capture long-term cross-period dependencies in the input features to obtain a context feature vector containing user cross-period patterns. The context feature vector is input into the survival analysis prediction layer to perform survival analysis prediction and obtain the conditional risk function. The negative log-partial likelihood loss function is used to perform end-to-end joint training of the temporal feature extraction layer and the survival analysis prediction layer to minimize the deviation between the model prediction and the user's actual response behavior, thereby obtaining the model optimization parameters. Based on the model optimization parameters, the conditional risk function is converted into a reach response probability density function, and based on the target user's current behavior data and external environment data, the reach response probability density function is used to generate a reach response probability density curve for the target user in the future for a preset time period.

5. The method according to claim 4, characterized in that, The structured time-series data table is transformed into input features for a dynamic preference prediction model, including: Based on the timestamps in the structured time-series data table, the corresponding behavioral data or external environment data are transformed into periodic features; The semantic information in the periodic features is embedded and converted into dense vectors. The periodic features and dense vectors together form the input features of the dynamic preference prediction model.

6. The method according to claim 1, characterized in that, Obtain the metadata of the content to be reached, and generate a list of reach tasks for the target user within a preset time period in conjunction with the reach response probability density curve, including: Based on a preset operation rule base and a first preset probability threshold, effective time windows within a future preset time period are filtered. The effective time window is used to characterize the time period within the future preset time period that satisfies the global constraints in the operation rule base and has a probability density not lower than the first preset probability threshold. Determine the content type of the metadata of the content to be reached, and based on the content type, filter the optimized time window for the reaching task from the effective time window.

7. The method according to claim 6, characterized in that, Based on content type, optimize time windows for reaching tasks are selected from the effective time window, including: If the content type is regular content, then the time point with the highest probability density is selected as the optimization time window for regular content based on the reach response probability density curve. If the content type is time-sensitive content, a time value decay function is introduced, and a comprehensive evaluation is conducted based on the time value decay function and the reach response probability density curve to determine the optimal time window for the time-sensitive content. If the content type is care content, then the optimization time window for care content is determined based on the preset operation rule base and the second preset probability threshold, wherein the second preset probability threshold is less than the first preset probability threshold.

8. A device for reaching middle-aged and elderly private domain users based on dynamic preference prediction, characterized in that, The device includes: The multidimensional data acquisition module is used to acquire the target user's raw behavioral data and external environment data, and to preprocess the raw behavioral data and external environment data to obtain the target user's structured time-series data table. The dynamic preference prediction module is used to input the structured time series data table into the dynamic preference prediction model for prediction, and obtain the reach response probability density curve of the target user in the future preset time period. The timing arbitrator is used to obtain metadata of the content to be reached and generate a list of reach tasks for the target user in the future within a preset time period by combining the reach response probability density curve. The strategy execution and feedback module is used to extract the reach content and reach time from the reach task list, and push the reach content to the target user according to the reach time.

9. An electronic device, characterized in that, include: A memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, the processor executing the computer instructions to perform the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the method of any one of claims 1 to 7.

Images