Insurance customer marketing task scheduling method and device, electronic equipment and storage medium

Abstract

This application relates to the fields of artificial intelligence applications and computer information processing technology, and can be applied to the fintech field. It discloses a method, device, electronic device, and storage medium for scheduling insurance customer marketing tasks. The method includes: acquiring policy records and interaction records of target customers within a preset time range, and generating a customer behavior sequence by sorting them chronologically; inputting the customer behavior sequence into a customer stage identification model, and outputting customer stage parameters; filtering candidate tasks based on the insurance marketing context; generating customer state features and task call features through a task scheduling controller, calling the task scheduling model to calculate selection parameters and generate a target task sequence; sequentially calling marketing task modules and transmitting processing results to generate marketing content and execution parameters; sending marketing content and obtaining customer responses, updating the marketing context and task status, and executing the next scheduling. This method improves the timeliness of customer stage identification, the synergy of marketing tasks, and the efficiency of marketing scheduling.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence applications and computer information processing technology, and can be applied to the financial technology field, particularly to a method, device, electronic device and storage medium for scheduling insurance customer marketing tasks. Background Technology

[0002] With the application of computer information processing technology in insurance business, insurance institutions typically process customers' policy records, product browsing records, consultation records, and interaction records through customer management systems and marketing management systems, and send marketing information such as insurance product introductions, renewal reminders, and protection plan descriptions to customers. Existing insurance customer marketing systems mostly follow preset customer tags, customer grouping rules, and fixed marketing processes, periodically analyzing customer data and then reaching customers through channels such as SMS, applications, instant messaging tools, or online customer service. The behavioral data generated by insurance customers during product browsing, consultation, policy application, and after-sales service stages has dynamic and time-related characteristics, and related data may also be distributed across different business systems. Existing technologies are prone to problems such as asynchronous data updates, lagging customer status information, and inconsistent data used in different processing stages when processing customer data from multiple sources. For example, if a customer has already entered the product comparison or policy consultation stage, the system may still send basic product introductions based on earlier browsing records, resulting in marketing information that does not match the customer's current needs. Furthermore, existing insurance customer marketing systems typically operate according to fixed workflows for customer analysis, marketing content generation, business function calls, and information transmission. Changes in customer behavior, insurance products, or marketing channels often necessitate reconfiguration of processing rules, interface parameters, or call order. When multiple marketing tasks are executed in parallel, issues such as duplicate sending, unreasonable sending order, or inconsistent content output from different channels may arise. Customer feedback regarding marketing information, such as clicks, inquiries, ignoring, or rejections, is also usually periodically aggregated before being used for subsequent processing, resulting in a lag in the utilization of feedback data. Summary of the Invention

[0003] The main technical problems addressed by the implementation method of this application are the untimely identification of customer status, insufficient coordination in marketing processes, and lag in customer response data processing in traditional insurance customer marketing technology.

[0004] To address the aforementioned technical problems, the first technical solution adopted in this application is as follows: Obtain policy records and interaction records of the target customer within a preset time range, sort them according to the record generation time to obtain a customer behavior sequence; input the customer behavior sequence into a pre-trained customer stage recognition model, process the behavioral changes and time correlations in the customer behavior sequence through the customer stage recognition model, and output customer stage parameters corresponding to the target customer; read the insurance marketing context of the target customer, and filter candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters; generate customer state features through a task scheduling controller based on the customer stage parameters and the insurance marketing context, and generate task call features corresponding to each candidate marketing task based on the calling conditions and interface descriptions of each candidate marketing task; input the customer state features and the task call features corresponding to each candidate marketing task into a pre-trained task... The scheduling model calculates task selection parameters for each candidate marketing task; selects target marketing tasks based on the task selection parameters, generates execution priorities based on the task selection parameters of each target marketing task, and generates a sequence of target marketing tasks based on the ranking of the execution priorities; according to the sequence of target marketing tasks, the task scheduling controller generates task call requests corresponding to each marketing task module, calls the marketing task modules sequentially, and inputs the processing result of the previous marketing task module into the next marketing task module to generate insurance marketing content and marketing execution parameters; sends the insurance marketing content to the marketing channel corresponding to the marketing execution parameters, and obtains customer response information generated by the target customer in response to the insurance marketing content; updates the insurance marketing context and the execution status of each marketing task based on the customer response information, and executes the next insurance customer marketing task scheduling according to the updated insurance marketing context.

[0005] To solve the above-mentioned technical problems, the second technical solution adopted in this application is: providing an insurance customer marketing task scheduling device, comprising: a customer behavior construction module, used to acquire policy records and interaction records of target customers within a preset time range, sort them according to the record generation time, and obtain a customer behavior sequence; a customer stage identification module, used to input the customer behavior sequence into a pre-trained customer stage identification model, process the behavior changes and time correlations in the customer behavior sequence through the customer stage identification model, and output customer stage parameters corresponding to the target customer; a candidate task filtering module, used to read the insurance marketing context of the target customer, and filter candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters; a scheduling feature generation module, used to generate customer status features according to the customer stage parameters and the insurance marketing context through a task scheduling controller, and generate task call features corresponding to each candidate marketing task according to the calling conditions and interface descriptions of each candidate marketing task; and a task selection calculation module, used to calculate the customer status features and each candidate marketing task. The system takes the following steps: A task call feature is input into a pre-trained task scheduling model, which calculates task selection parameters for each candidate marketing task; a target task orchestration module selects target marketing tasks based on the task selection parameters, generates execution priorities based on the task selection parameters of each target marketing task, and generates a target marketing task sequence based on the ranking of the execution priorities; a marketing task execution module generates task call requests corresponding to each marketing task module according to the target marketing task sequence, calls the marketing task modules sequentially, and inputs the processing result of the previous marketing task module into the next marketing task module to generate insurance marketing content and marketing execution parameters; a marketing content push module sends the insurance marketing content to the marketing channels corresponding to the marketing execution parameters and obtains customer response information generated by the target customer in response to the insurance marketing content; and a marketing status update module updates the insurance marketing context and the execution status of each marketing task based on the customer response information, and executes the next insurance customer marketing task scheduling according to the updated insurance marketing context.

[0006] To solve the above-mentioned technical problems, the third technical solution adopted in the embodiments of this application is: to provide an electronic device, including: 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 to enable the at least one processor to perform the insurance customer marketing task scheduling method as described above.

[0007] To solve the above-mentioned technical problems, the fourth technical solution adopted in the embodiments of this application is: to provide a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by an electronic device, the electronic device performs the insurance customer marketing task scheduling method as described above.

[0008] Unlike related technologies, this application performs time-series processing on policy records and interaction records to identify the customer's current stage by combining changes in customer behavior. The customer stage, insurance marketing context, and candidate marketing tasks are then input into a task scheduling controller to generate a sequence of target marketing tasks, which are then continuously invoked into the corresponding marketing task modules. Since the processing results of the previous task can be passed to the next task, and customer response information can be used to update subsequent scheduling statuses, this reduces duplicate or mismatched marketing tasks, improves the timeliness of customer status identification, enhances the synergy between marketing tasks, and increases task scheduling efficiency. Attached Figure Description

[0009] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0010] Figure 1 This is a schematic diagram of the operating environment of the insurance customer marketing task scheduling method provided in the embodiments of this application.

[0011] Figure 2 This is a schematic diagram of the execution flow of the insurance customer marketing task scheduling method provided in the embodiments of this application.

[0012] Figure 3 This is a schematic diagram of the execution flow for generating a target marketing task sequence in the insurance customer marketing task scheduling method provided in this application embodiment.

[0013] Figure 4 This is a schematic diagram of the execution process for generating insurance marketing content and marketing execution parameters in the insurance customer marketing task scheduling method provided in this application embodiment.

[0014] Figure 5 This is a schematic diagram of the system structure of the insurance customer marketing task scheduling device provided in the embodiments of this application.

[0015] Figure 6 This is a schematic diagram of the hardware structure of an electronic device for executing an insurance customer marketing task scheduling method provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Software tools, components, or servers not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0017] It should be noted that, unless otherwise specified, the various features in the embodiments of this application can be combined with each other, all of which are within the protection scope of this application. Furthermore, although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device schematic diagram or the order in the flowchart.

[0018] Unless otherwise defined, all technical and scientific terms used in this specification have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The term "and / or" as used in this specification includes any and all combinations of one or more of the associated listed items.

[0019] To facilitate understanding of this embodiment, a detailed description of an insurance customer marketing task scheduling method disclosed in this application embodiment will be provided first. Please refer to [link / reference]. Figure 1 , Figure 1 This is a schematic diagram of the operating environment of the insurance customer marketing task scheduling method provided in the embodiments of this application, such as... Figure 1 As shown, the execution subject of the insurance customer marketing task scheduling method provided in this application embodiment is generally an electronic device with a certain computing power, such as a computer device. In some possible implementations, the insurance customer marketing task scheduling method can be implemented by a processor calling computer-readable instructions stored in memory. Figure 1 The computer equipment mentioned can be a server. A server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms. This can be understood as... Figure 1 The number of computer devices shown is merely illustrative and can be expanded in any number according to actual needs.

[0020] Please continue reading. Figure 2 , Figure 2This is a schematic diagram of the execution flow of the insurance customer marketing task scheduling method provided in the embodiments of this application, such as... Figure 2 As shown, it includes the following steps: S1. Obtain the policy records and interaction records of the target customer within a preset time range, sort them according to the time of record generation, and obtain the customer behavior sequence.

[0021] Specifically, the preset time range is used to define the boundaries of records participating in customer behavior analysis, and can be set according to the insurance business cycle or customer interaction frequency. Policy records are used to represent business events such as insurance purchase, renewal, premium payment, policy maintenance, claims, and policy cancellation, while interaction records are used to represent interaction events such as browsing, querying, consulting, replying, and refusing. Each record is configured with its generation time and record type, and organized according to chronological order, so that the customer behavior sequence can represent the sequential order and continuous changes between business events. In fintech scenarios, records of policy expiration, renewal page browsing, premium query, and quotation feedback generated by auto insurance customers within the last ninety days can be obtained, and a customer behavior sequence can be constructed according to the record generation time. Through the above processing, input data with a clear time structure can be provided for customer stage identification.

[0022] As an optional implementation, the process of obtaining the customer behavior sequence in step S1 above may also specifically include the following steps S11 to S15.

[0023] S11. Read the policy records and interaction records corresponding to the target customer.

[0024] Specifically, policy records can include business records such as policy application, renewal, premium payment, policy maintenance, claims, and policy surrender, while interaction records can include operation records such as page browsing, premium inquiry, online consultation, message reply, and price feedback. When reading policy records and interaction records, records can be filtered according to target customer identifiers and preset time ranges, limiting subsequent processing to business events and interaction events formed by the same target customer within the same analysis period.

[0025] S12. Standardize the time format of policy records and interaction records, and correct the record generation time according to the record source.

[0026] Specifically, records from different sources may use time fields with different date formats, time precision, or time zone bases, and may also have record time offsets due to data transmission, batch writing, or asynchronous processing. Unifying the time format can convert different time fields into the same time representation, and the record generation time correction can adjust the time value according to the time base corresponding to the record source, transmission delay, or writing rules, so that records from different sources can be compared and arranged on the same timeline.

[0027] S13. Delete records with duplicate content and merge multiple records belonging to the same consecutive operation into one behavior node.

[0028] Specifically, duplicate records can be multiple identical records generated when a target customer repeatedly performs actions on the same page, message, or business entry point. Deleting duplicate records reduces the interference of repeated events on the number and density of actions. Continuous operations can be identified according to the operation object, operation type, and time interval between adjacent operations, merging multiple operation records formed around the same business purpose into behavior nodes. Behavior nodes represent a relatively complete customer behavior, and can record the behavior type, start time, end time, and associated business object.

[0029] S14. Use a sliding time window to divide each behavior node, and arrange the behavior nodes in each sliding time window according to the corrected record generation time.

[0030] Specifically, a sliding time window is used to organize behavioral nodes according to continuous and partially overlapping time ranges. The sliding time window can be configured with a window length and a sliding step size. The window length defines the time range covered by a single local behavioral analysis, while the sliding step size controls the degree of overlap between adjacent sliding time windows. Arranging the behavioral nodes within the window according to the corrected record generation time sequence preserves the chronological relationship between behavioral nodes within the local time range and provides ordered input for identifying short-term behavioral changes and continuous interaction processes.

[0031] S15. Connect the first and last behavior nodes in adjacent sliding time windows to obtain a customer behavior sequence arranged continuously in time.

[0032] Specifically, adjacent sliding time windows may contain some of the same behavior nodes, or the behavior process may be segmented due to the window boundaries. Connecting the first and last behavior nodes in adjacent sliding time windows can establish a sequential relationship based on the record generation time of the behavior nodes and the business object, and merge recurring behavior nodes so that the local behavior fragments in different sliding time windows form a time-continuous sequence of customer behavior.

[0033] As an example, records of policy expiration, renewal quote inquiries, renewal page views, online inquiries, and quote feedback generated by auto insurance customers within the last 90 days can be read. The time field of each record is uniformly converted to year, month, day, hour, minute, and second format, and the record generation time is corrected according to the write latency of each record source. Multiple page views and price inquiries generated consecutively within five minutes related to the same renewal quote can be merged into a single renewal quote behavior node. A sliding time window with a length of seven days and a sliding step of one day is used to divide each behavior node. The behavior nodes within each sliding time window are arranged according to the corrected record generation time. Connecting the first and last behavior nodes in adjacent sliding time windows yields a customer behavior sequence encompassing the process of policy expiration, quote inquiries, online inquiries, and quote feedback.

[0034] Through the above steps S11 to S15, the impact of time format differences, repetitive operations, and window boundary segmentation between different record sources on customer behavior sequences can be reduced, forming customer behavior sequences with clear time order, continuous behavior nodes, and less repetitive content, providing structured input for subsequent customer stage identification.

[0035] S2. Input the customer behavior sequence into the pre-trained customer stage identification model. The customer stage identification model processes the behavioral changes and time correlations in the customer behavior sequence and outputs the customer stage parameters corresponding to the target customer.

[0036] Specifically, the customer stage identification model transforms multiple behavioral nodes arranged sequentially in time into customer stage parameters representing the current customer stage of a target customer. The customer behavior sequence serves as the input data for the model, with each behavioral node representing a relatively complete customer action. These nodes can include behavior type, business object, click frequency, dwell time, response status, and record generation time. The model comprises a time-series embedding layer, a time-series correlation layer, and a stage output layer, connected sequentially. The time-series embedding layer converts different types of behavioral information into time-series embedding features with uniform dimensions. The time-series correlation layer handles the temporal dependencies between the current behavioral node and its predecessors. The stage output layer maps the time-series correlation features to matching parameters corresponding to each preset customer stage. The time-series embedding features are intermediate data at the behavioral node level, the time-series correlation features represent the continuous change process between multiple behavioral nodes, and the customer stage parameters are model output data at the customer behavior sequence level. Preset customer stages can include the potential awareness period, demand emergence period, price comparison decision period, purchase execution period, and post-sales evaluation period. Customer stage parameters can be represented by a sequence of matching parameters corresponding to each preset customer stage, or they can also include the customer stage identifier with the largest matching parameter. Customer stage parameters do not directly generate marketing content, but rather serve as input for subsequent candidate marketing task selection and task scheduling.

[0037] As an optional implementation, the above-mentioned customer stage identification model includes a time series embedding layer, a time series correlation layer, and a stage output layer. The process of inputting the customer behavior sequence into the pre-trained customer stage identification model in step S2 may also specifically include the following steps S21 to S26.

[0038] S21. Input the customer behavior sequence into the time series embedding layer.

[0039] Specifically, before the customer behavior sequence is input into the time series embedding layer, the behavior nodes can be organized according to a preset sequence length. The preset sequence length can be set based on the customer behavior density and the model input dimension. When the number of behavior nodes is greater than the preset sequence length, behavior nodes closer to the current time can be retained, or behavior nodes can be selected according to a preset sampling interval. When the number of behavior nodes is less than the preset sequence length, padding positions can be added at the end of the customer behavior sequence, and a masking identifier corresponding to the padding position can be generated. The masking identifier is used to prevent the subsequent time series association layer from calculating the association weight of the padding position. The organized customer behavior sequence is input into the time series embedding layer according to the chronological order of the behavior nodes, so that customer behavior sequences of different lengths can be processed using a unified data structure.

[0040] S22. Extract the behavioral features of each behavioral node in the customer behavior sequence, and generate time and location features according to the order of the behavioral nodes.

[0041] Specifically, the time-series embedding layer extracts behavioral features and temporal location features from behavioral nodes. Behavioral features characterize the operation content corresponding to the behavioral node and can be generated based on behavior type, business object, click frequency, dwell time, number of operations, and response status. For example, behavior type can distinguish between page browsing, premium inquiry, plan comparison, online consultation, and insurance application; business object can distinguish between insurance products, coverage, premium plans, and service items. Temporal location features characterize the arrangement and temporal distribution of behavioral nodes in the customer behavior sequence and can be generated based on the sequence position of the behavioral node, record generation time, time interval between adjacent behaviors, and time difference from the current time. Behavioral features indicate what operation the target customer performed, while temporal location features indicate when the corresponding operation occurred and its position within the customer behavior change process.

[0042] S23. Integrate the behavioral features and time location features corresponding to each behavioral node to obtain the time series embedding features corresponding to each behavioral node.

[0043] Specifically, the time-series embedding layer maps the behavioral features and temporal location features corresponding to the same behavioral node to the same feature dimension, and then fuses the mapped behavioral features and temporal location features. The fusion method can be feature addition, linear mapping after feature concatenation, or weighted processing according to preset fusion parameters. The time-series embedding layer can also perform normalization processing on the fusion result to ensure that the numerical ranges corresponding to different types of behavioral features are consistent, resulting in time-series embedded features corresponding to each behavioral node. Each time-series embedded feature contains both behavioral content information and temporal location information. These time-series embedded features form an embedded feature sequence according to the order of the behavioral nodes, and the embedded feature sequence is input into the temporal correlation layer.

[0044] S24. Embed the time series into the feature input time series association layer and calculate the association weight between each behavior node and the preceding behavior node.

[0045] Specifically, the temporal association layer can include one or more sequentially connected temporal association units. Each temporal association unit includes a feature mapping unit, an association calculation unit, and a weight normalization unit. The feature mapping unit maps the time-series embedded features corresponding to the current behavior node and its predecessor behavior nodes, respectively, to obtain the query features corresponding to the current behavior node, the key features corresponding to each predecessor behavior node, and the value features corresponding to each predecessor behavior node. The association calculation unit calculates the association parameters between the query features and each key feature. These association parameters represent the degree of content association and temporal association between the current behavior node and each predecessor behavior node. The weight normalization unit normalizes each association parameter to obtain the association weights between the current behavior node and each predecessor behavior node. When calculating the association weights, a temporal order masking method can be used to restrict the current behavior node to only establish associations with behavior nodes preceding it, preventing later-occurring behavior nodes from participating in the feature calculation of previous behavior nodes. When multiple temporal association units are sequentially connected, the output of the previous-level temporal association unit serves as the input of the next-level temporal association unit, thereby processing short-term behavioral relationships and long-term behavioral relationships spanning multiple behavior nodes level by level.

[0046] S25. Aggregate the embedded features of each time series according to the association weight to generate time-series association features that characterize the change process of the target customer's behavior.

[0047] Specifically, the temporal association layer performs weighted aggregation of the value features corresponding to each preceding behavior node according to the association weight between the current behavior node and each preceding behavior node, thus obtaining the associated behavior feature corresponding to the current behavior node. Preceding behavior nodes with larger association weights occupy a larger proportion in the weighted aggregation result, while those with smaller association weights occupy a smaller proportion. The temporal association layer can perform residual fusion between the associated behavior feature and the time series embedded feature corresponding to the current behavior node, and perform normalization and nonlinear mapping on the residual fusion result to obtain the temporal processing feature corresponding to the current behavior node. For multi-level temporal association units, association weight calculation, feature aggregation, and feature mapping can be repeated. After processing each behavior node, the temporal processing features can be weighted and summarized according to the validity identifier and time position of each behavior node to generate temporal association features that characterize the overall behavioral change process of the target customer. The temporal association features can represent the change path of the target customer from page browsing to premium inquiry, plan comparison, online consultation, or insurance application.

[0048] S26. Input the time-series correlation features into the stage output layer, calculate the matching parameters between the target customer and each preset customer stage, and output the customer stage parameters according to each matching parameter.

[0049] Specifically, the stage output layer receives the temporal correlation features output by the temporal correlation layer and maps these features to the parameter space corresponding to the preset customer stages through one or more fully connected mapping units. When there are multiple preset customer stages, the stage output layer outputs the same number of initial stage parameters as the preset number of customer stages. The stage output layer normalizes each initial stage parameter to obtain matching parameters corresponding to the potential awareness period, demand emergence period, price comparison decision period, purchase execution period, and post-sales evaluation period, respectively. Customer stage parameters can include all matching parameters, or they can include the maximum value of the matching parameters, the customer stage identifier corresponding to the maximum value of the matching parameters, and the differences between the matching parameters. The differences between the matching parameters can be used to represent the degree of discrimination of the customer stage identification results. When the matching parameters corresponding to two customer stages are close, the matching parameters corresponding to the two customer stages can be retained for further processing in the subsequent task scheduling process in conjunction with the insurance marketing context.

[0050] As an example, a customer stage identification model can employ a feature engineering module, a time-series embedding layer, a Transformer encoder, and a fully connected classification head, connected in sequence. The feature engineering module receives raw behavior logs arranged chronologically, extracts behavioral features such as behavior type, business object, number of operations, dwell time, response status, and time interval between adjacent behaviors, and inputs these features into the time-series embedding layer in the order of the behavior nodes. The time-series embedding layer fuses the behavioral features with their corresponding temporal location features to obtain the time-series embedded features for each behavior node. The Transformer encoder comprises multiple sequentially connected encoding units. Each encoding unit processes the temporal correlation between the current behavior node and its preceding behavior nodes through attention calculations and performs weighted aggregation of the time-series embedded features to obtain temporal correlation features representing the process of customer behavior change. The fully connected classification head maps the temporal correlation features to the parameter space corresponding to each preset customer stage, calculates the matching parameters corresponding to the potential awareness period, demand emergence period, price comparison decision period, purchase execution period, and after-sales evaluation period, and outputs the customer stage identifier corresponding to each matching parameter and the maximum value of the matching parameter as the customer stage parameter output. The output customer stage parameters can be further input into the task scheduling controller as input for subsequent marketing task selection and task orchestration.

[0051] As another example, behavioral nodes formed by car insurance customers within the past 30 days, such as checking policy expiration, browsing renewal pages, inquiring about premiums, comparing protection plans, and consulting with agents, can be input into the time series embedding layer. The time series embedding layer extracts the behavior type, business object, and time interval between adjacent behaviors for each behavioral node, and generates time position features according to the order of the behavioral nodes. After fusion processing, the time series embedding features corresponding to each behavioral node are obtained. The time series correlation layer calculates the correlation weight between the agent consultation behavior node and the premium inquiry and protection plan comparison behavior nodes, and aggregates the time series embedding features according to the correlation weight. The stage output layer calculates the matching parameters for each preset customer stage based on the aggregated time series correlation features. When the matching parameter corresponding to the price comparison decision period is higher than the matching parameters corresponding to other preset customer stages, the customer stage parameter containing the price comparison decision period identifier is output.

[0052] Through the above steps S21 to S26, the business content, arrangement position, time interval and previous and subsequent relationships of the behavior nodes can be retained during the customer stage identification process. This allows the customer stage parameters to reflect the continuous behavioral changes of the target customer, providing a more accurate parameter basis for subsequent screening of marketing tasks according to the current customer stage.

[0053] As an optional implementation, historical customer behavior sequences corresponding to multiple historical customers can be obtained. Customer stage labels are generated based on actual business events such as product browsing, premium inquiries, plan comparisons, insurance purchases, and after-sales interactions that occurred within the corresponding time period for each historical customer, forming training samples for the customer stage identification model. Before training, the historical customer behavior sequences are time-aligned, truncated, or padded, and a training sample set and a validation sample set are created. During training, the historical customer behavior sequences are sequentially input into the time series embedding layer, the time series association layer, and the stage output layer to obtain the prediction matching parameters corresponding to each preset customer stage. The classification loss is calculated based on the prediction matching parameters and the customer stage labels, and the weight parameters and bias parameters of each layer are updated based on the classification loss. Training parameters may include preset sequence length, embedded feature dimension, number of time series association units, learning rate, training batch size, and maximum training epochs. Training stops when the classification loss corresponding to the validation sample does not decrease within a consecutive preset epochs, or when the maximum training epochs are reached, resulting in a pre-trained customer stage identification model.

[0054] S3. Read the insurance marketing context of the target customer and filter candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters.

[0055] Specifically, the insurance marketing context records the target customer's historical task execution status, current marketing status, and marketing interactions within the current marketing cycle. Customer stage parameters represent the degree of matching between the target customer and each preset customer stage. The marketing task library stores marketing tasks such as renewal reminders, protection plan introductions, premium calculations, manual consultation guidance, and deferred marketing, along with their corresponding triggering conditions. The candidate marketing task screening process first loads matching marketing task constraints according to the customer stage parameters, then combines the insurance marketing context to determine whether each marketing task meets the requirements for execution timing, execution frequency, and task connection, thereby excluding marketing tasks that do not match the current customer stage, have already been executed, or do not yet meet the conditions for invocation. Through this process, the scope of subsequent task scheduling can be narrowed, and candidate marketing tasks can be kept consistent with the target customer's current stage and historical marketing processes.

[0056] As an optional implementation, the process of screening candidate marketing tasks in step S3 above may also specifically include the following steps S31 to S36.

[0057] S31. Obtain the historical task execution status and current marketing status recorded in the insurance marketing context.

[0058] S32. Load marketing task constraints that match the customer stage parameters.

[0059] S33. Extract the task trigger conditions corresponding to each marketing task in the marketing task library.

[0060] Specifically, historical task execution status indicates whether each marketing task has been executed, the number of executions, the execution time, and the execution result. Current marketing status indicates the current marketing progress of the target customer. Marketing task constraints limit the scope of marketing tasks that can participate in the screening at different customer stages, and may include the applicable stage of the task, the allowed execution period, and the maximum number of executions. Task triggering conditions describe the specific conditions that a marketing task must meet to enter the candidate scope, and may include customer stage conditions, marketing status conditions, historical execution conditions, and task invocation conditions. Marketing task constraints and task triggering conditions respectively handle task scope limitation and specific condition judgment, avoiding the direct input of all marketing tasks in the marketing task library into the subsequent task scheduling process.

[0061] S34. Match customer stage parameters, current marketing status, and historical task execution status with the triggering conditions of each task.

[0062] S35. Retain marketing tasks that meet the marketing task constraints and task triggering conditions.

[0063] S36. Delete marketing tasks that have task conflicts, are executed repeatedly, or have incomplete calling conditions to obtain candidate marketing tasks.

[0064] Specifically, the system can compare customer stage parameters, current marketing status, and historical task execution status with the task triggering conditions of each marketing task, and record the matching results for each task triggering condition. Marketing tasks that meet all marketing task constraints and whose necessary triggering conditions are successfully matched are retained. Task conflicts indicate that two marketing tasks cannot be executed simultaneously in terms of execution content, timing, or marketing channel. Duplicate execution indicates that the same marketing task has been completed within a preset interval. Incomplete calling conditions indicate that a marketing task lacks necessary input parameters or available calling interfaces. After completing conflict detection, duplicate detection, and calling condition checks, the remaining marketing tasks constitute candidate marketing tasks.

[0065] As an example, when the customer stage parameters of a car insurance customer indicate that the customer is in the price comparison decision-making period, marketing task constraints applicable to this period can be loaded, such as comparing protection plans, explaining premium discounts, and providing manual consultation guidance. If the insurance marketing context records that the premium calculation task has been executed, the protection plan comparison task has not been executed, and the current marketing status is "waiting for the customer to choose a plan," then the customer stage parameters, the current marketing status, and the historical task execution status are matched with the task triggering conditions of each marketing task. After filtering out already executed premium calculation tasks, other tasks that conflict with the execution timing of manual consultation guidance tasks, and tasks lacking necessary plan parameters, the protection plan comparison task and the premium discount explanation task can be selected as candidate marketing tasks.

[0066] Through steps S31 to S36 above, the scope of marketing tasks can be narrowed down step by step by combining customer stage, current marketing progress and historical task execution status, and marketing tasks that conflict, duplicate or do not meet the conditions for invocation can be excluded, providing candidate marketing tasks that are in line with the current marketing status for subsequent task selection and execution priority calculation.

[0067] S4. Generate customer status characteristics based on customer stage parameters and insurance marketing context through the task scheduling controller, and generate task call characteristics corresponding to each candidate marketing task based on the calling conditions and interface descriptions of each candidate marketing task.

[0068] Specifically, the task scheduling controller handles the data transformation and transmission relationships between customer stage parameters, insurance marketing context, and candidate marketing tasks. Customer stage parameters represent the degree of matching between the target customer and each preset customer stage, while the insurance marketing context records the target customer's historical task execution status, current marketing status, and existing customer responses. The task scheduling controller reads the customer stage parameters and insurance marketing context, performs field mapping, encoding, and normalization on data with different formats and value ranges, generating customer status features. These features centrally characterize the target customer's current customer stage, current marketing progress, and historical marketing task execution status. Each candidate marketing task is configured with invocation conditions and interface descriptions. Invocation conditions limit the customer stage, marketing status, and prerequisite task status that the corresponding candidate marketing task must meet to execute, while the interface description specifies the input parameters the candidate marketing task needs to receive and the output content it can return. Based on the invocation conditions and interface descriptions of each candidate marketing task, the task scheduling controller extracts the applicable task stage, prerequisite task requirements, input parameter types, parameter sources, output content types, and invocation restrictions, and converts the extracted content into task invocation features with unified structure and dimensions. Each candidate marketing task corresponds to a set of task invocation features, while customer status features correspond to the overall status of the target customer in this task scheduling. The customer status features are status data at the customer level, while the task invocation features are descriptive data at the candidate marketing task level; their data sources and representation objects differ. The candidate marketing tasks selected in the previous step limit the scope of tasks to be processed in this step. The customer status features and task invocation features generated in this step serve as inputs to the task scheduling model in the next step, enabling the model to calculate the task selection parameters for each candidate marketing task based on the target customer's current status and the invocation requirements of each candidate marketing task. As an optional implementation, the above task scheduling model includes a policy decision network, and the interface description includes input parameter descriptions and output templates. Please continue reading. Figure 3 , Figure 3 This is a schematic diagram of the execution flow for generating a target marketing task sequence in the insurance customer marketing task scheduling method provided in this application embodiment. For example... Figure 3 As shown, the process of generating the target marketing task sequence in step S4 above may also specifically include the following steps S41 to S47.

[0069] S41. According to the preset task scheduling control protocol, the customer stage parameters and insurance marketing context are encapsulated into context state.

[0070] Specifically, the pre-defined task scheduling control protocol specifies the organization of customer status information, candidate marketing task information, and task scheduling results. The protocol can set context state fields, candidate task fields, target task fields, and execution priority fields. Customer stage parameters represent the current customer stage of the target customer, while the insurance marketing context represents the historical marketing task execution status, current marketing status, and existing customer responses. The task scheduling controller, according to the protocol, converts the customer stage parameters and insurance marketing context into context states with unified field names, data types, and arrangement order. Context states can be represented using context state vectors or structured state objects to avoid the customer stage parameters and insurance marketing context using different data formats and thus being unable to participate in task selection calculations together.

[0071] S42. Read the task identifier, calling conditions, input parameter descriptions and output templates of each candidate marketing task, and generate a list of candidate task descriptions based on the correspondence between the task identifier and the calling conditions, input parameter descriptions and output templates.

[0072] Specifically, each candidate marketing task can be configured with a unique task identifier, which is then associated with invocation conditions, input parameter descriptions, and output templates. Invocation conditions indicate the customer stage, marketing status, and preceding task status that the candidate marketing task must meet to be allowed to execute. Input parameter descriptions indicate the parameter names, types, and sources required when invoking the marketing task module. Output templates indicate the result fields and format returned by the marketing task module after processing. After reading the description information of each candidate marketing task, the corresponding invocation conditions, input parameter descriptions, and output templates are combined into candidate task description items using the task identifier as an index. These candidate task description items are then organized according to the order of the candidate marketing tasks to generate a candidate task description list. The candidate task description list corresponds to the set of available tasks in the task scheduling control protocol.

[0073] S43. Parse the context state and candidate task description list through the task scheduling controller to extract customer state features and task call features.

[0074] Specifically, the task scheduling controller parses the context state and candidate task description list according to the task scheduling control protocol. For the context state, it extracts customer stage matching parameters, current marketing status, historical task execution status, and historical customer responses, and encodes or normalizes content with different value ranges to obtain customer status features. For the candidate task description list, it extracts task type, applicable customer stage, required input parameters, output content type, and invocation restrictions, and converts the extracted content into task invocation features with unified dimensions. Each candidate marketing task corresponds to a set of task invocation features, and the customer status features and each set of task invocation features together constitute the input data of the task scheduling model.

[0075] S44. Call the business rule base to match customer status characteristics with the calling conditions of each candidate marketing task, and generate the callable status of each candidate marketing task based on the matching results.

[0076] Specifically, the business rule base can store rules for applicable stages of marketing tasks, execution frequency rules, execution time rules, prerequisite task rules, and task mutual exclusion rules. The task scheduling controller compares customer status characteristics with the invocation conditions of each candidate marketing task. When the customer status characteristics meet all the necessary invocation conditions corresponding to a candidate marketing task, the invocation status of the candidate marketing task is set to invocationable. When any necessary invocation condition is not met, the invocation status of the candidate marketing task is set to non-invocationable. The invocation status of each candidate marketing task can be arranged into an invocation status vector according to the task order in the candidate task description list, thereby enabling the strategy decision network to distinguish between invocationable and non-invocationable marketing tasks during the task selection calculation process.

[0077] S45. Input customer status characteristics, task invocation characteristics, and invocation status into the strategy decision network to calculate the task selection parameters for each candidate marketing task.

[0078] Specifically, the strategy decision network is used to calculate the degree of selectivity between the current customer state and each candidate marketing task. In one optional implementation, the strategy decision network can employ a pre-trained reinforcement learning strategy network. The strategy decision network receives customer state features, task invocation features corresponding to each candidate marketing task, and invocation status, respectively. It then fuses the customer state features and task invocation features and processes the fused features through multi-level mapping. For candidate marketing tasks whose invocation status is "not invocationable," the corresponding network output can be masked. For candidate marketing tasks whose invocation status is "invocationable," the strategy decision network outputs the corresponding task selection parameters. The task selection parameters can be represented by selection probability or selection score, used to characterize the degree of matching between the corresponding candidate marketing task and the current customer state and the current marketing process.

[0079] S46. Filter out candidate marketing tasks that cannot be called based on their callable status, and select the target marketing task according to the task selection parameters.

[0080] Specifically, first, uncallable candidate marketing tasks are removed from the candidate marketing task pool based on their callable status. Then, the task selection parameters of the remaining candidate marketing tasks are compared. Target marketing tasks can be selected using a task selection parameter threshold; candidate marketing tasks whose task selection parameters reach the preset threshold are selected as target marketing tasks. Target marketing tasks can also be selected using a preset number of tasks, selecting a preset number of candidate marketing tasks in descending order of task selection parameters. The number of target marketing tasks can be one or more, allowing a single task scheduling to select either a single marketing task or a combination of marketing tasks that need to be executed consecutively.

[0081] S47. Generate execution priorities based on the task selection parameters of each target marketing task, and generate a sequence of target marketing tasks based on the ranking results of the execution priorities.

[0082] Specifically, the task scheduling controller reads the task selection parameters corresponding to each target marketing task and generates the execution priority of each target marketing task according to the sorting result of the task selection parameters from largest to smallest. Target marketing tasks with larger task selection parameters are assigned higher execution priorities, and target marketing tasks with smaller task selection parameters are assigned lower execution priorities. The target marketing tasks are arranged in order of execution priority to obtain a target marketing task sequence. The target marketing task sequence can include the task identifier, execution priority, and sorting position of the target marketing tasks, and serves as the input for subsequently generating task invocation requests and sequentially invoking marketing task modules. The target task field and execution priority field in the task scheduling control protocol are used to record the selected marketing tasks and their corresponding execution order, respectively.

[0083] As an optional implementation, the preset task scheduling control protocol can use structured protocol messages to record input and output data during the task scheduling process. The protocol messages include a context state vector field, an available task set field, a selected task field, and an execution priority field. Specifically, the context state vector field records the context state vector generated based on customer stage parameters and the insurance marketing context; the available task set field records the identifiers of candidate marketing tasks that meet business rules and invocation conditions; the selected task field records the identifier of the target marketing task selected by the strategy decision network based on the context state vector and the task invocation characteristics of each candidate marketing task; and the execution priority field records the execution order of the target marketing task in the target marketing task sequence. By parsing these fields, the task scheduling controller can establish a unified data transfer relationship between customer status, candidate marketing tasks, and task scheduling results. For example: { "Context state vector": [0.85, -0.32, 1.20], "Available Task Set": ["Discount Calculation Task", "Policy Comparison Task", "Coverage Plan Explanation Task"] Select the task: "Policy Comparison Task". "Execution Priority": 1 }

[0084] In the aforementioned protocol message, the parameters in the context state vector represent the target customer's customer stage, current marketing status, and historical task execution status, respectively. The available task set includes marketing tasks that are in a callable state after being matched according to business rules. The selected task represents the target marketing task output by the strategy decision network. An execution priority of 1 indicates that the corresponding target marketing task will be executed first in the current target marketing task sequence. The parameter values ​​and task names mentioned above are only used to illustrate the field organization method of the task scheduling control protocol and do not constitute a limitation on the dimensions of the context state vector, the types of candidate marketing tasks, or the number of execution priorities.

[0085] As an example, when the customer stage parameters of a car insurance customer indicate that the customer is in the price comparison decision period, and the insurance marketing context records that the premium calculation task has been completed but the policy comparison task has not been executed, the customer stage parameters, current marketing status, and historical task execution status can be encapsulated into a context state. Candidate marketing tasks include premium discount calculation tasks, policy comparison tasks, and manual consultation guidance tasks. The task scheduling controller reads the task identifier, invocation conditions, input parameter descriptions, and output templates of each candidate marketing task, generates a candidate task description list, extracts customer status features from the context state, and extracts task invocation features from the candidate task description list. The business rule base matches the customer status features with the invocation conditions of each candidate marketing task, marking tasks lacking necessary invocation parameters as uninvokeable. The strategy decision network calculates the task selection parameters for the remaining candidate marketing tasks. When the task selection parameters for the policy comparison task are the highest, and those for the premium discount calculation task are the second highest, the policy comparison task and the premium discount calculation task can be selected as target marketing tasks, and corresponding execution priorities and target marketing task sequences are generated according to the sorting results of the task selection parameters.

[0086] Through steps S41 to S47 above, the customer stage, insurance marketing context, business call rules, and candidate marketing task interfaces can be uniformly converted into inputs that the task scheduling model can process. The selection of target marketing tasks and the arrangement of execution order are completed through business rule matching and strategy decision network calculation, thereby providing a well-structured sequence of target marketing tasks for the continuous invocation of subsequent marketing task modules.

[0087] As an optional implementation, the strategy decision network may include a customer state encoding layer, a task feature encoding layer, a feature fusion layer, and a strategy output layer. The customer state encoding layer processes customer state features, the task feature encoding layer processes task invocation features corresponding to each candidate marketing task, the feature fusion layer fuses customer state encoding features and task encoding features, and the strategy output layer combines the callable states to output task selection parameters corresponding to each candidate marketing task. Historical insurance marketing task scheduling records can be obtained, and historical customer states, historical candidate marketing tasks, actually executed marketing tasks, and corresponding customer response results can be correlated to generate training samples for the task scheduling model. During training, historical customer state features, task invocation features, and callable states are input into the strategy decision network to obtain predicted task selection parameters; the strategy loss is calculated based on the predicted task selection parameters, the actually executed marketing tasks, and the corresponding customer response results, and the weight and bias parameters of the strategy decision network are updated based on the strategy loss. Training parameters may include the encoding feature dimension, learning rate, training batch size, and maximum training epochs. Training stops when the strategy loss corresponding to the validation sample meets the preset convergence condition or reaches the maximum training epochs, resulting in a pre-trained task scheduling model.

[0088] S5. Input the customer status characteristics and the task invocation characteristics corresponding to each candidate marketing task into the pre-trained task scheduling model, and calculate the task selection parameters of each candidate marketing task through the task scheduling model.

[0089] Specifically, the task scheduling controller organizes the data transfer between customer stage parameters, insurance marketing context, and candidate marketing tasks, and provides scheduling input to the task scheduling model according to a unified task scheduling control method. Customer stage parameters represent the current marketing stage of the target customer; the insurance marketing context represents the historical task execution status, current marketing status, and existing interaction results; and candidate marketing tasks limit the range of marketing tasks that can be selected in this scheduling. Each candidate marketing task is configured with invocation conditions and an interface description. The invocation conditions limit the customer state, marketing state, and preceding task state that the marketing task can execute, while the interface description specifies the input parameters the marketing task needs to receive and the output content it can return. After parsing the invocation conditions and interface descriptions, the task scheduling controller can convert candidate marketing tasks of different formats into task invocation features with a unified structure. These task invocation features, along with the customer stage parameters and the customer state features corresponding to the insurance marketing context, are then input into the pre-trained task scheduling model. The task scheduling model processes the adaptation relationship between the target customer's current state and each candidate marketing task, and outputs the task selection parameters corresponding to each candidate marketing task. Task selection parameters represent the degree of selection of candidate marketing tasks within the current insurance marketing context. These parameters are distinct from subsequent execution priorities. Task selection parameters are used to select target marketing tasks, while execution priorities are used to arrange the execution order of the selected target marketing tasks. Through this process, customer stages, historical marketing processes, and task invocation requirements can be incorporated into the task selection calculation, providing a unified parameter basis for subsequent target marketing task selection and execution order arrangement.

[0090] S6. Select target marketing tasks based on task selection parameters, generate execution priorities based on task selection parameters for each target marketing task, and generate a sequence of target marketing tasks based on the ranking of execution priorities.

[0091] Specifically, step S6 is used to convert the task selection parameters output by the task scheduling model into a sequence of target marketing tasks. Target marketing tasks can be selected from the available candidate marketing tasks according to the task selection parameters, and the execution priority can be set by combining the task selection parameters and the relationship with the preceding tasks to generate a sequence of target marketing tasks that records task identifiers and their arrangement positions.

[0092] S7. According to the target marketing task sequence, generate task call requests corresponding to each marketing task module through the task scheduling controller, call the marketing task modules in sequence, and input the processing results of the previous marketing task module into the next marketing task module to generate insurance marketing content and marketing execution parameters.

[0093] Specifically, step S7 converts the target marketing task sequence into a continuously transferable calling process between multiple marketing task modules. Each marketing task module can be configured as an independent service unit operating according to a unified calling specification. Different marketing task modules are used to perform processes such as customer information extraction, protection plan generation, premium discount calculation, marketing copy generation, marketing channel selection, or delivery timing calculation. The task scheduling controller reads the task identifier and arrangement position of the target marketing tasks sequentially according to the target marketing task sequence, and loads the input parameter description, execution logic, and output template of the corresponding marketing task module according to the task identifier. The input parameter description is used to limit the parameter name, parameter type, and parameter source that the marketing task module needs to receive. The execution logic is used to limit the specific processing procedure after the marketing task module receives the input content. The output template is used to limit the result fields and data format included in the task processing result. According to the unified calling specification, the task scheduling controller encapsulates the customer stage parameters, current marketing status, and existing business information in the insurance marketing context, or the task processing result generated by the previous marketing task module, into a task calling request corresponding to the current marketing task module. The current marketing task module processes the input content in the task calling request according to the execution logic and generates the task processing result according to the output template. The task scheduling controller verifies the field integrity, data format, and parameter value range of the task processing results. It then converts the verified task processing results into the input content required by the next marketing task module, thus establishing the result transfer relationship between adjacent marketing task modules. The output data of the previous marketing task module and the input data of the next marketing task module can be converted using the field mapping relationship in the unified calling specification; different marketing task modules are not required to use the same internal processing method. After completing all module calls in the target marketing task sequence, insurance marketing content and marketing execution parameters can be generated based on the task processing results of each marketing task module. The insurance marketing content represents the text, page information, or protection plan information sent to the target customer, while the marketing execution parameters represent execution requirements such as marketing channels, sending time, sending frequency, and content display method.

[0094] Through the above processing, multiple functionally independent marketing task modules can be continuously called according to the target marketing task sequence, and the processing results of the previous marketing task module can be directly used for the next marketing task module. This reduces the repeated reading and conversion of insurance marketing context between different marketing tasks, and provides a unified calling basis for the replacement, addition and combination of marketing task modules.

[0095] As an optional implementation, the aforementioned marketing task module is an independent service unit configured according to a unified invocation specification. Please continue reading. Figure 4 , Figure 4This is a schematic diagram illustrating the execution flow of generating insurance marketing content and marketing execution parameters in the insurance customer marketing task scheduling method provided in this application embodiment. For example... Figure 4 As shown, the process of generating task call requests corresponding to each marketing task module through the task scheduling controller according to the target marketing task sequence and calling the marketing task modules in sequence may also include the following steps S71 to S78.

[0096] S71. Read the current marketing tasks sequentially according to the target marketing task sequence, and load the marketing task module corresponding to the current marketing task.

[0097] S72. Read the input parameter description, execution logic, output template, and evaluation metrics of the marketing task module.

[0098] Specifically, marketing task modules can be registered as functionally independent service units in the task module library. Each marketing task module is configured with a corresponding task identifier, and a mapping relationship is established between the task identifier and the target marketing task. The task scheduling controller reads the current marketing task sequentially according to its position in the target marketing task sequence, queries the task module library using the task identifier of the current marketing task, and loads the marketing task module corresponding to the current marketing task. When loading a marketing task module, the input parameter description, execution logic, output template, and evaluation metrics of the marketing task module are read simultaneously. The input parameter description defines the parameter names, parameter types, and parameter sources required for the marketing task module to perform processing; the execution logic defines the business processing or model calculations performed by the marketing task module after receiving input content; the output template defines the output fields, field types, and organization format in the task processing results; and the evaluation metrics define the completeness, validity, or result quality requirements that the task processing results must meet. Different marketing task modules can use different internal algorithms and business processing logics, but they are all described according to the same calling elements, enabling the task scheduling controller to load and call marketing task modules with different functions in a unified manner.

[0099] S73. Based on the input parameter description, extract the input content from the insurance marketing context or the task processing result of the previous marketing task module.

[0100] S74. Generate a task invocation request based on the unified invocation standard and input content.

[0101] S75. Send a task invocation request to the marketing task module, and control the marketing task module to process the input content according to the execution logic and generate the task processing result.

[0102] Specifically, the task scheduler controller identifies the required input parameters and their corresponding sources for the current marketing task module based on the input parameter descriptions. When the current marketing task module is the first in the target marketing task sequence, it can extract input content such as customer stage parameters, current marketing status, existing policy information, and historical task execution results from the insurance marketing context. When the current marketing task module follows other marketing task modules, it can extract corresponding input content from the task processing results of the previous marketing task module, or it can extract input content from both the insurance marketing context and the task processing results of the previous marketing task module simultaneously. Following a unified calling standard, the task scheduler controller encapsulates the task identifier, input parameters, and corresponding values ​​of the marketing task module into a task call request and sends the request to the current marketing task module. The current marketing task module parses the task call request, processes the input content according to its execution logic, and generates a task processing result that conforms to the output template. Therefore, the task scheduler controller does not need to pre-define the internal implementation of each marketing task module; it only needs to complete the input content transmission and task processing result reception according to the unified calling standard.

[0103] As a specific implementation, the marketing task module may include a policy comparison and recommendation module. This module receives existing policy information and target insurance product parameters from the target customer. It extracts the coverage scope, coverage amount, deductible conditions, premium, and service content for each coverage item from both the existing policy information and the target insurance product parameters. Then, it calculates the coverage item score for each coverage item in the target insurance product according to preset scoring rules. The module weights the coverage item scores according to the weights corresponding to each coverage item, generating a recommendation index for the target insurance product. The recommendation index can be calculated using the following formula: in, Indicates the first i Recommendation index for each target insurance product n This indicates the number of insurance projects included in the comparison. Indicates the first j The weight corresponding to each protection project Indicates the first i The target insurance product in the first j The scoring of each coverage item; the weighting of each coverage item satisfies: The coverage score can be generated based on the differences between the target insurance product and existing policies in terms of coverage scope, coverage amount, deductible, premium, and service content. The policy comparison suggestion module generates coverage difference information, product recommendation results, and recommendation reasons based on the recommendation index and the differences between each coverage item, and outputs the coverage difference information, product recommendation results, and recommendation reasons as the task processing results.

[0104] In one optional implementation, the weights corresponding to each coverage item can be obtained through a pre-trained weight generation model, which can employ an extreme gradient boosting model. When constructing training samples for the weight generation model, historical insurance product recommendation records, historical policy information, historical coverage item parameters, and corresponding historical transaction results can be obtained. Training samples are generated according to the correlation between insurance product recommendation records and historical transaction results. Each training sample includes coverage item features and transaction tags corresponding to each coverage item. The transaction tags indicate whether the customer has completed the purchase based on the corresponding insurance product.

[0105] During training, the coverage scope, coverage amount, deductible conditions, premium differences, and service content differences for each insurance item are input into the extreme gradient boosting model to obtain the predicted transaction parameters for historical insurance products. The training loss is then calculated based on these predicted transaction parameters and transaction labels. The extreme gradient boosting model constructs decision trees sequentially according to the training loss, updating the parameters of newly added split nodes and leaf nodes. Training parameters may include the maximum depth of the decision tree, learning rate, subsample ratio, and maximum number of iterations. Training stops when the training loss for the validation samples meets the preset convergence condition or reaches the maximum number of iterations.

[0106] After training, based on the split gain or cumulative contribution parameters of each coverage feature in each decision tree, importance parameters are generated for each coverage item. These importance parameters are then normalized to obtain the weights of each coverage item. These weights can be incorporated into the execution logic of the policy comparison and recommendation module to calculate the recommendation index for the target insurance product.

[0107] S76. Verify the task processing results based on the output template and evaluation indicators.

[0108] S77. Generate the input content for the next marketing task module based on the verified task processing results, and continue to call the next marketing task module.

[0109] Specifically, after receiving the task processing results generated by the current marketing task module, the task scheduling controller first checks whether the task processing results contain the necessary output fields according to the output template, and checks whether the data type, data format, and parameter values ​​of each output field meet the corresponding requirements. Then, it checks whether the task processing results meet the preset result requirements according to the evaluation indicators. Different marketing task modules can be configured with different evaluation indicators. For example, the evaluation indicators for the policy comparison suggestion module may include the completeness of the insurance projects being compared and the effective range of the recommendation index, while the evaluation indicators for the marketing content generation module may include the completeness of necessary business information and the degree of conformity with the content template. The verified task processing results are then mapped to the input parameters of the next marketing task module, converting the output fields in the current task processing results into the input fields required by the next marketing task module, and generating the input content for the next marketing task module. The task scheduling controller then reads the next marketing task according to the target marketing task sequence, repeating the process of loading the marketing task module, generating the task call request, verifying the task processing results, and transmitting the results, until all marketing tasks in the target marketing task sequence are completed.

[0110] S78. Generate insurance marketing content and marketing execution parameters based on the task processing results of each marketing task module.

[0111] Specifically, the task processing results of each marketing task module can respectively serve as intermediate calculation results, components of marketing content, and marketing execution control information. After the task scheduling controller completes the invocation of all marketing task modules, it can combine protection plan information, premium difference information, preferential information, and marketing prompt information according to preset content templates to generate insurance marketing content for target customers. At the same time, it extracts information such as marketing channel identifier, sending time, sending frequency, content display format, and subsequent interaction entry points from the task processing results to generate marketing execution parameters. Insurance marketing content is business content displayed to target customers, while marketing execution parameters are execution information that controls how and when insurance marketing content is sent; both serve as inputs for subsequent marketing content push steps.

[0112] As an example, a target marketing task sequence can sequentially include a policy comparison suggestion task, a discount information generation task, and a marketing content generation task. The task scheduler controller first loads the policy comparison suggestion module corresponding to the policy comparison suggestion task and reads the input parameter description, weighted scoring execution logic, output template, and recommendation index evaluation indicators for that marketing task module. Based on the input parameter description, it extracts the target customer's existing policy information and target insurance product parameters from the insurance marketing context and generates the task call request corresponding to the policy comparison suggestion task. The policy comparison suggestion module uses a weighted scoring algorithm to calculate the comparison parameters of each coverage item in the existing policy and the target insurance product, and generates coverage difference information and a recommendation index.

[0113] The task scheduling controller verifies the coverage difference information and recommendation index according to the output template and recommendation index evaluation indicators. It then converts the verified coverage difference information and recommendation index into the input content required by the discount information generation module. The discount information generation module combines the coverage difference information, recommendation index, and available discount rules to generate a discount description. Subsequently, the task scheduling controller inputs the coverage difference information, recommendation index, and discount description into the marketing content generation module to generate insurance marketing content that includes coverage differences, reasons for recommendation, and discount information. It also generates marketing execution parameters based on the channel selection results and delivery timing results generated during the execution of the target marketing task sequence.

[0114] Through steps S71 to S78 above, marketing task modules with different internal processing methods can be connected using a unified calling standard. Data transfer relationships between marketing task modules can be established through input parameter descriptions, output templates, and evaluation indicators. This enables the task scheduling controller to dynamically combine and continuously call multiple independent service units according to the target marketing task sequence, while retaining the ability to replace, add, or adjust each marketing task module individually.

[0115] S8. Send insurance marketing content to the marketing channels corresponding to the marketing execution parameters, and obtain customer response information from target customers in response to the insurance marketing content.

[0116] Specifically, marketing execution parameters specify the channels, timing, frequency, content format, and interaction points for sending insurance marketing content. Marketing channels can include application messaging, SMS, email, instant messaging, or human customer service. Before sending the marketing content, it can be converted to a format compatible with the marketing channel specified in the marketing execution parameters. Target customer identifiers, marketing task identifiers, and task scheduling identifiers are then written into the sending request, ensuring that customer response information is linked to the corresponding insurance marketing content and target marketing task. After the marketing content is sent, the delivery status returned by the marketing channel can be obtained, and customer responses such as viewing, clicking, querying, inquiring, purchasing, rejecting, or unsubscribing from the marketing content can be collected. Customer response information is generated based on these responses. Customer response information can include the response result, response time, response channel, corresponding marketing task identifier, and corresponding task scheduling identifier.

[0117] As an example, when marketing execution parameters instruct the sending of car insurance coverage plan comparison content via application messaging channels, the differences in coverage responsibilities, premium differences, preferential information, and consultation access can be organized according to the application message template, and the insurance marketing content can be sent to the corresponding application account of the target customer. After the target customer views the insurance marketing content, clicks on the coverage plan details, or enters the human consultation page, customer response information corresponding to the content viewing, plan inquiry, or consultation request is generated respectively, and the customer response information is linked to the insurance marketing content and target marketing task corresponding to this task scheduling.

[0118] Through the above processing, the actual responses of target customers to insurance marketing content can be collected, and feedback data can be provided for subsequent updates to the insurance marketing context, marketing task execution status, and task scheduling model.

[0119] S9. Update the insurance marketing context and the execution status of each marketing task based on the customer response information, and execute the next insurance customer marketing task scheduling according to the updated insurance marketing context.

[0120] Specifically, customer response information represents the actual interaction results of the target customer with the insurance marketing content, the insurance marketing context records the current state of the target customer during the continuous marketing process, and the execution status of the marketing task represents the execution progress of the corresponding marketing task in this scheduling. After obtaining the customer response information, it can be written into the execution record of the corresponding marketing task, updating the current marketing status, and converting the customer response information into a scheduling feedback value that the task scheduling model can process. When the scheduling feedback value reaches the preset update condition, the accumulated scheduling feedback value is used to correct the strategy parameters of the task scheduling model, so that subsequent task selection calculations can incorporate the actual response results generated by historical marketing tasks. The updated insurance marketing context continues to serve as the input for the next candidate marketing task screening and task scheduling processing, thus forming a continuous processing flow between insurance marketing content generation, marketing content delivery, customer response collection, and marketing task rescheduling.

[0121] As an optional implementation, the process of updating the insurance marketing context and the execution status of each marketing task based on the customer response information in step S8 above may also specifically include the following steps S81 to S87.

[0122] S91. Link customer response information with corresponding insurance marketing content, marketing channels, and target marketing task sequences.

[0123] S92. Identify the response result corresponding to the customer response information and write the response result into the execution record of the corresponding marketing task.

[0124] S93. Update the execution status of the corresponding marketing task according to the response result, and simultaneously update the current marketing status in the insurance marketing context.

[0125] Specifically, based on the task scheduling identifier, marketing task identifier, marketing content identifier, and marketing channel identifier recorded in the customer response information, the target marketing task sequence that generates the corresponding insurance marketing content can be found, and the customer response information can be associated with the insurance marketing content, marketing channel, and target marketing task sequence. Response results can be categorized according to customer operation type, such as delivery, viewing, clicking, plan inquiry, consultation, insurance application, rejection, or cancellation. After identifying the response result, the response type, response time, and response channel are written into the execution record of the corresponding marketing task, and the execution status of the marketing task is updated according to the response result. The execution status of the marketing task can include pending execution, in execution, sent, responded to, completed, and terminated. The current marketing status in the insurance marketing context is updated synchronously with the execution status of the marketing task. For example, when a target customer clicks the consultation entry, the manual consultation guidance task can be updated to "responded," and the current marketing status can be updated to "awaiting manual consultation"; when a target customer completes the insurance application operation, the corresponding marketing task can be updated to "completed," and the current marketing status can be updated to "insurance application completed."

[0126] S94. Based on the response results and the task selection parameters of the corresponding marketing task, generate the scheduling feedback value corresponding to the marketing task.

[0127] S95. Write the scheduling feedback value into the historical task execution result of the corresponding marketing task.

[0128] S96. When the accumulated scheduling feedback value meets the preset update conditions, the strategy parameters of the task scheduling model are corrected using the scheduling feedback value corresponding to each marketing task.

[0129] Specifically, a correspondence between response results and basic feedback parameters can be pre-established, and scheduling feedback values ​​can be generated according to the basic feedback parameters corresponding to the response results and the task selection parameters of the corresponding marketing tasks. Different response results can be configured with different basic feedback parameters. For example, applying for insurance, submitting inquiries, and querying insurance plans can be configured with positive basic feedback parameters, while refusing to receive and unsubscribing can be configured with negative basic feedback parameters, and content delivered but not viewed can be configured with lower basic feedback parameters. Task selection parameters are used to represent the degree of selection of marketing tasks by the task scheduling model before execution, and response results are used to represent the actual results after the marketing tasks are executed. By weighting the two, a scheduling feedback value reflecting the degree of correspondence between the model's selection results and the actual response results can be generated.

[0130] Scheduling feedback values ​​can be written into historical task execution results according to marketing task identifier, customer stage, and task scheduling time, and accumulated according to the same marketing task or the same customer stage. Preset update conditions can include the accumulated number of scheduling feedback values ​​reaching a preset sample size, or the time since the last model update reaching a preset update cycle. For example, a week can be set as the preset update cycle, and the accumulated scheduling feedback values ​​within that cycle can be processed centrally. When the preset update conditions are met, the customer state characteristics, task invocation characteristics, task selection parameters, and scheduling feedback values ​​corresponding to the execution of the marketing task can be combined into a strategy update sample and input into the strategy decision network of the task scheduling model. The strategy decision network calculates the adjustment direction of the strategy parameters based on the scheduling feedback values ​​and uses a gradient update method to correct the network weights and bias parameters. Positive scheduling feedback values ​​are used to improve the subsequent selection parameters of the corresponding marketing task under the corresponding customer state, while negative scheduling feedback values ​​are used to decrease the subsequent selection parameters of the corresponding marketing task under the corresponding customer state. The task invocation conditions and callable states configured in the business rule base do not change with the strategy parameter updates, enabling the strategy decision network to complete online adjustments within the range of candidate marketing tasks that satisfy the business rules.

[0131] S97. Read the updated insurance marketing context, re-filter candidate marketing tasks, and execute the next insurance customer marketing task scheduling.

[0132] Specifically, after updating the marketing task execution status and insurance marketing context, the system can reread the customer stage parameters, current marketing status, historical task execution status, and most recent customer response result for the target customer, and then match the marketing task triggering conditions again according to the updated insurance marketing context. Marketing tasks that have been completed, rejected, or repeatedly executed within a preset interval will no longer be included in the candidate marketing task range, while marketing tasks that are aligned with the latest customer response result can enter the next scheduling process. The task scheduling controller calculates the task selection parameters according to the re-selected candidate marketing tasks, generates a new sequence of target marketing tasks, and continues to execute the next insurance customer marketing task scheduling.

[0133] As an example, when a car insurance customer is in the price comparison decision-making period, the task scheduling controller sequentially calls the protection plan comparison task and the human consultation guidance task, and sends insurance marketing content through the application's messaging channel. After the target customer views the protection plan comparison content and clicks the human consultation entry, the view response and consultation response can be associated with the protection plan comparison task and the human consultation guidance task, respectively. The protection plan comparison task is updated to "completed," the human consultation guidance task is updated to "responded," and the current marketing status in the insurance marketing context is updated to "awaiting human consultation." Based on the positive basic feedback parameters corresponding to the consultation response and the task selection parameters of the human consultation guidance task, a positive scheduling feedback value is generated and written to the historical task execution results. After the accumulated scheduling feedback value meets the preset update conditions, the corresponding strategy is used to update the sample and correct the strategy parameters of the strategy decision network. In the next task scheduling, completed protection plan comparison tasks can be filtered out, and consultation appointment or agent access tasks can be selected based on the current marketing status of "awaiting human consultation."

[0134] Through steps S91 to S97 above, customer response information can be written into the marketing task execution record and insurance marketing context, and the scheduling feedback value can be used to correct the strategy parameters of the task scheduling model, so that the next insurance customer marketing task scheduling can be continuously adjusted in combination with historical response results.

[0135] It should be noted that the policy records, interaction records, customer response information, historical insurance product recommendation records, and historical transaction results involved in this application's implementation are all obtained with legal authorization and in compliance with data security and personal information protection requirements. Furthermore, they can undergo de-identification, anonymization, abnormal data cleanup, and access control before participating in model training or task scheduling. Training sample labels are used to represent customer behavior stages, marketing task execution results, or insurance product transaction results. The label generation process does not use indicators that measure the value of a natural person's life, personal dignity, or social value, nor does it differentiate customers based on sensitive personal information unrelated to insurance marketing tasks. The marketing task constraints, task triggering conditions, and invocation conditions in the business rule base are used to control the applicable stage, execution timing, number of executions, prerequisite task relationships, and task conflict relationships of marketing tasks, and do not include rules that violate laws, social morality, or harm public interests. Customer stage identification results, task selection parameters, recommendation index, and scheduling feedback values ​​are used solely for insurance marketing task screening, task execution order arrangement, and insurance marketing content generation. They are not used to measure the value of human life, nor are they the sole basis for insurance underwriting, insurance pricing, claims eligibility, or other decisions that significantly impact the rights and interests of natural persons. The delivery of insurance marketing content can also be controlled according to customer authorization scope, marketing channel usage rules, and unsubscription status. When a target customer refuses to receive or withdraws the corresponding authorization, the delivery of the corresponding insurance marketing content and subsequent marketing task scheduling will be stopped.

[0136] The insurance customer marketing task scheduling method provided in this application organizes policy records and interaction records in chronological order, and uses a customer stage identification model to process changes in customer behavior and temporal relationships to obtain customer stage parameters. Since customer stage parameters reflect the continuous behavioral state of customers, combining them with the insurance marketing context to filter candidate marketing tasks can reduce mismatched, duplicate, or uncallable marketing tasks. Furthermore, a task scheduling controller generates task selection parameters, execution priorities, and a sequence of target marketing tasks, and continuously calls multiple marketing task modules according to a unified calling standard to achieve inter-module processing result transfer; the marketing status and model parameters are updated based on customer response information, allowing subsequent scheduling to be continuously adjusted based on historical execution results. Thus, a closed-loop processing of customer stage identification, task selection, module calling, and response feedback can be formed, improving the automation level, execution continuity, and resource utilization efficiency of insurance marketing task scheduling.

[0137] Please continue reading. Figure 5 , Figure 5 This is a schematic diagram of the system structure of the insurance customer marketing task scheduling device provided in the embodiments of this application, such as... Figure 5As shown, the insurance customer marketing task scheduling device 50 includes: a customer behavior construction module 51, used to acquire policy records and interaction records of target customers within a preset time range, sort them according to the record generation time, and obtain a customer behavior sequence; a customer stage identification module 52, used to input the customer behavior sequence into a pre-trained customer stage identification model, process the behavior changes and time correlations in the customer behavior sequence through the customer stage identification model, and output the customer stage parameters corresponding to the target customer; a candidate task filtering module 53, used to read the insurance marketing context of the target customer, and filter candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters; a scheduling feature generation module 54, used to generate customer status features according to the customer stage parameters and insurance marketing context through the task scheduling controller, and generate task call features corresponding to each candidate marketing task according to the calling conditions and interface descriptions of each candidate marketing task; and a task selection calculation module 55, used to input the customer status features and the task call features corresponding to each candidate marketing task into the task selection calculation module 55. A pre-trained task scheduling model calculates task selection parameters for each candidate marketing task. A target task orchestration module 56 selects target marketing tasks based on the task selection parameters, generates execution priorities based on these parameters, and generates a target marketing task sequence based on the priority ranking. A marketing task execution module 57 generates task call requests for each marketing task module according to the target marketing task sequence, calls the modules sequentially, and inputs the processing results of the previous module into the next, generating insurance marketing content and marketing execution parameters. A marketing content push module 58 sends the insurance marketing content to the marketing channels corresponding to the marketing execution parameters and obtains customer response information from target customers. A marketing status update module 59 updates the insurance marketing context and the execution status of each marketing task based on customer response information and executes the next insurance customer marketing task scheduling according to the updated context.

[0138] It should be noted that the aforementioned insurance customer marketing task scheduling device can execute the insurance customer marketing task scheduling method provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects of the method. Technical details not described in detail in the embodiments of the insurance customer marketing task scheduling device can be found in the insurance customer marketing task scheduling method provided in the embodiments of this application.

[0139] Figure 6 This is a schematic diagram of the hardware structure of the electronic device for executing the insurance customer marketing task scheduling method provided in the embodiments of this application, such as... Figure 6 As shown, the electronic device 600 includes: One or more processors 610 and memory 620, Figure 6Take the 610 processor as an example.

[0140] The processor 610 and the memory 620 can be connected via a bus or other means. Figure 6 Taking the example of a connection between China and Israel via a bus.

[0141] The memory 620, 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 the program instructions / modules corresponding to the insurance customer marketing task scheduling method in the embodiments of this application. The processor 610 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions, and modules stored in the memory 620, thereby implementing the insurance customer marketing task scheduling method in the above-described method embodiments.

[0142] The memory 620 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the insurance customer marketing task scheduling device. Furthermore, the memory 620 may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 620 may optionally include memory remotely located relative to the processor 610, and these remote memories can be connected to the insurance customer marketing task scheduling device via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0143] The one or more modules are stored in the memory 620. When executed by the one or more processors 610, they execute the insurance customer marketing task scheduling method in any of the above method embodiments, for example, executing the above-described... Figure 2 Method steps S1 to S9, Figure 3 Method steps S41 to S47, Figure 4 Steps S71 to S78 in the method are implemented. Figure 5 The functions of modules 51-59 in the document.

[0144] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0145] This application provides a non-volatile computer-readable storage medium storing computer-executable instructions that are executed by one or more processors, for example... Figure 6One of the processors 610 can enable the one or more processors to execute the insurance customer marketing task scheduling method in any of the above method embodiments, for example, to perform the above-described... Figure 2 Method steps S1 to S9, Figure 3 Method steps S41 to S47, Figure 4 Steps S71 to S78 in the method are implemented. Figure 5 The functions of modules 51-59 in the document.

[0146] This application provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium. The computer program includes program instructions, which, when executed by an electronic device, enable the electronic device to perform the insurance customer marketing task scheduling method in any of the above method embodiments, for example, to perform the above-described... Figure 2 Method steps S1 to S9, Figure 3 Method steps S41 to S47, Figure 4 Steps S71 to S78 in the method are implemented. Figure 5 The functions of modules 51-59 in the document.

[0147] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0148] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software and a general-purpose hardware platform, or of course, using hardware. Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0149] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them; under the concept of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of this application as described above, which are not provided in detail for the sake of brevity; although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for scheduling insurance customer marketing tasks, characterized in that, include: Obtain policy records and interaction records of target customers within a preset time range, sort them according to the time the records were generated, and obtain a sequence of customer behavior. The customer behavior sequence is input into a pre-trained customer stage identification model. The customer stage identification model processes the behavioral changes and time correlations in the customer behavior sequence and outputs the customer stage parameters corresponding to the target customer. Read the insurance marketing context of the target customer, and filter candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters; The task scheduling controller generates customer status characteristics based on the customer stage parameters and the insurance marketing context, and generates task call characteristics corresponding to each candidate marketing task based on the calling conditions and interface descriptions of each candidate marketing task. The customer status features and the task invocation features corresponding to each candidate marketing task are input into the pre-trained task scheduling model, and the task selection parameters of each candidate marketing task are calculated through the task scheduling model. Select target marketing tasks according to the task selection parameters, generate execution priorities according to the task selection parameters of each target marketing task, and generate a target marketing task sequence according to the sorting result of the execution priorities; According to the target marketing task sequence, the task scheduling controller generates task call requests corresponding to each marketing task module, calls the marketing task modules in sequence, and inputs the processing result of the previous marketing task module into the next marketing task module to generate insurance marketing content and marketing execution parameters. The insurance marketing content is sent to the marketing channel corresponding to the marketing execution parameters, and the customer response information generated by the target customer in response to the insurance marketing content is obtained. The insurance marketing context and the execution status of each marketing task are updated based on the customer response information, and the next insurance customer marketing task scheduling is executed according to the updated insurance marketing context.

2. The insurance customer marketing task scheduling method according to claim 1, characterized in that, The process involves acquiring the target customer's policy records and interaction records within a preset time range, sorting them according to the record generation time, and obtaining a customer behavior sequence, including: Read the policy records and interaction records corresponding to the target customer; The time format of the policy records and the interaction records is standardized, and the record generation time is corrected according to the record source; Delete records with duplicate content and merge multiple records belonging to the same consecutive operation into one behavior node; A sliding time window is used to divide each behavior node, and the behavior nodes in each sliding time window are arranged according to the corrected record generation time. Connect the first and last behavior nodes in adjacent sliding time windows to obtain the customer behavior sequence arranged continuously in time.

3. The insurance customer marketing task scheduling method according to claim 2, characterized in that, The customer stage identification model includes a time series embedding layer, a time series correlation layer, and a stage output layer; The step of inputting the customer behavior sequence into a pre-trained customer stage identification model includes: The customer behavior sequence is input into the time series embedding layer; Extract the behavioral features of each behavioral node in the customer behavior sequence, and generate time location features according to the order of the behavioral nodes. By fusing the behavioral features and temporal location features corresponding to each behavioral node, the time series embedding features corresponding to each behavioral node are obtained; Each time series embedding feature is input into the time series association layer to calculate the association weight between each behavior node and the preceding behavior node. Aggregate the embedded features of each time series according to the association weight to generate time-series association features that characterize the change process of the target customer's behavior; The time-series correlation features are input into the stage output layer to calculate the matching parameters between the target customer and each preset customer stage, and the customer stage parameters are output according to each matching parameter.

4. The insurance customer marketing task scheduling method according to claim 1, characterized in that, The step of reading the insurance marketing context of the target customer and filtering candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters includes: Obtain the historical task execution status and current marketing status recorded in the insurance marketing context; Load marketing task constraints that match the customer stage parameters; Extract the task triggering conditions corresponding to each marketing task in the marketing task library; The customer stage parameters, the current marketing status, and the historical task execution status are matched with the triggering conditions of each task. Retain marketing tasks that satisfy the marketing task constraints and the task triggering conditions; The candidate marketing tasks are obtained by deleting marketing tasks that have task conflicts, are executed repeatedly, or have incomplete invocation conditions.

5. The insurance customer marketing task scheduling method according to claim 1, characterized in that, The task scheduling model includes a strategy decision network, and the interface description includes input parameter descriptions and output templates; the step of generating customer state features through the task scheduling controller based on the customer stage parameters and the insurance marketing context, and generating task invocation features corresponding to each candidate marketing task based on the invocation conditions and interface descriptions of each candidate marketing task, includes: According to the preset task scheduling control protocol, the customer stage parameters and the insurance marketing context are encapsulated into a context state; Read the task identifier, invocation conditions, input parameter descriptions, and output templates of each candidate marketing task, and generate a list of candidate task descriptions based on the correspondence between the task identifier and the invocation conditions, input parameter descriptions, and output templates; The task scheduling controller parses the context state and the candidate task description list to extract customer state features and task call features. The business rule base is invoked to match the customer status characteristics with the invocation conditions of each candidate marketing task, and the invocation status of each candidate marketing task is generated based on the matching results. The customer status characteristics, the task invocation characteristics, and the invocation status are input into the strategy decision network to calculate the task selection parameters for each candidate marketing task. Candidate marketing tasks that cannot be called are filtered out based on the callable status, and target marketing tasks are selected according to the task selection parameters. An execution priority is generated based on the task selection parameters of each target marketing task, and the target marketing task sequence is generated based on the sorting result of the execution priority.

6. The insurance customer marketing task scheduling method according to claim 1, characterized in that, The marketing task module is an independent service unit configured according to a unified invocation specification; the step of generating task invocation requests corresponding to each marketing task module through the task scheduling controller according to the target marketing task sequence, and sequentially invoking the marketing task modules, includes: The current marketing task is read sequentially according to the target marketing task sequence, and the marketing task module corresponding to the current marketing task is loaded. Read the input parameter description, execution logic, output template, and evaluation metrics of the marketing task module; Based on the input parameter description, extract the input content from the insurance marketing context or the task processing result of the previous marketing task module; The task invocation request is generated based on the unified invocation specification and the input content; Send the task invocation request to the marketing task module, and control the marketing task module to process the input content according to the execution logic and generate the task processing result; Verify the task processing results according to the output template and the evaluation indicators; The input content for the next marketing task module is generated based on the verified task processing results, and the next marketing task module is called again. The insurance marketing content and marketing execution parameters are generated based on the task processing results of each marketing task module.

7. The insurance customer marketing task scheduling method according to claim 1, characterized in that, The step of updating the insurance marketing context and the execution status of each marketing task based on the customer response information includes: The customer response information is associated with the corresponding insurance marketing content, marketing channels, and target marketing task sequences. Identify the response result corresponding to the customer response information and write the response result into the execution record of the corresponding marketing task; Update the execution status of the corresponding marketing task according to the response result, and simultaneously update the current marketing status in the insurance marketing context; Based on the response result and the task selection parameters of the corresponding marketing task, a scheduling feedback value corresponding to the marketing task is generated; Write the scheduling feedback value into the historical task execution result of the corresponding marketing task; When the accumulated scheduling feedback value meets the preset update conditions, the strategy parameters of the task scheduling model are corrected using the scheduling feedback value corresponding to each marketing task. Read the updated insurance marketing context, re-filter candidate marketing tasks, and execute the next insurance customer marketing task scheduling.

8. An insurance customer marketing task scheduling device, characterized in that, include: The customer behavior construction module is used to obtain the policy records and interaction records of target customers within a preset time range, sort them according to the record generation time, and obtain the customer behavior sequence. The customer stage identification module is used to input the customer behavior sequence into a pre-trained customer stage identification model, process the behavioral changes and time correlations in the customer behavior sequence through the customer stage identification model, and output the customer stage parameters corresponding to the target customer. The candidate task filtering module is used to read the insurance marketing context of the target customer and filter candidate marketing tasks that meet the task triggering conditions from the marketing task library according to the customer stage parameters. The scheduling feature generation module is used to generate customer status features based on the customer stage parameters and the insurance marketing context through the task scheduling controller, and to generate task call features corresponding to each candidate marketing task based on the calling conditions and interface descriptions of each candidate marketing task. The task selection calculation module is used to input the customer status features and the task invocation features corresponding to each candidate marketing task into the pre-trained task scheduling model, and calculate the task selection parameters of each candidate marketing task through the task scheduling model. The target task orchestration module is used to select target marketing tasks according to the task selection parameters, generate execution priorities according to the task selection parameters of each target marketing task, and generate a target marketing task sequence according to the sorting result of the execution priorities. The marketing task execution module is used to generate task call requests corresponding to each marketing task module through the task scheduling controller according to the target marketing task sequence, call the marketing task modules in sequence, and input the processing result of the previous marketing task module into the next marketing task module to generate insurance marketing content and marketing execution parameters. The marketing content push module is used to send the insurance marketing content to the marketing channels corresponding to the marketing execution parameters, and to obtain customer response information generated by the target customers in response to the insurance marketing content; The marketing status update module is used to update the insurance marketing context and the execution status of each marketing task according to the customer response information, and to execute the next insurance customer marketing task scheduling according to the updated insurance marketing context.

9. An electronic device, characterized in that, include: At least one processor; as well as, 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 to enable the at least one processor to perform the insurance customer marketing task scheduling method according to any one of claims 1-7.

10. A non-volatile computer-readable storage medium, characterized in that, The non-volatile computer-readable storage medium stores computer-executable instructions, which, when executed by an electronic device, cause the electronic device to perform the insurance customer marketing task scheduling method according to any one of claims 1-7.

Images