Orphan insurance policy customer historical information abstract processing method

Abstract

The invention provides an orphan insurance policy customer historical information abstract processing method, and relates to the technical field of data processing, and the method comprises the steps: setting an insurance policy monitoring state flow control point if it is detected that an insurance policy state is changed into an orphan insurance policy; the process control point is located at an initial triggering node of policy state change, an instant node for confirming the orphan policy state and a starting node of the service handover process; based on matching analysis of the flow control points and the monitoring time sequence nodes, a geometric control area is obtained, and gridding segmentation is carried out on the geometric control area; obtaining a data collection adjustment value based on the distribution characteristics of the grids; performing optimization adjustment on the cross-source heterogeneous historical data according to the adjustment value; and based on the adjusted cross-source heterogeneous historical data, obtaining a customer interaction timeline through semantic alignment and time sequence reconstruction processing. According to the invention, customer active characteristics and service preferences can be quantified, and the service handover efficiency and customer service quality are improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for processing historical information summaries of orphan policy customers. Background Technology

[0002] In the insurance industry, especially in the life insurance sector, orphan policies have become a core pain point restricting service quality and the value extraction of existing customers. With the continuous reduction of the insurance agent force, a large number of policies are experiencing service gaps due to the departure of their original agents.

[0003] While some companies manually assign orphan policies to new agents or community grid workers, the depth of service and responsiveness are difficult to guarantee due to the limited manpower available to each agent. Other companies have introduced system integration tools such as ESB (Enterprise Service Bus) to connect heterogeneous systems such as policy management and CRM, but this only achieves preliminary data aggregation and cannot solve the semantic conflicts and temporal sequence problems of cross-source data. A few digital tools can generate basic customer profiles, but they mostly remain at the level of information listing and lack in-depth analysis of customer interaction patterns and potential needs.

[0004] The current handover of orphan policy services faces various bottlenecks: First, data collection lacks specificity. For example, existing systems mostly use a full-data scraping model without filtering data based on policy status change nodes (such as agent resignation triggers, orphan policy confirmation, and handover initiation), resulting in redundant handover materials and missing key information. Second, customer information is severely fragmented. Heterogeneous data scattered across multiple channels, such as APP interaction logs, customer service call recordings, and offline policy maintenance records, lacks semantic alignment and temporal reconstruction, making it difficult to form a coherent service trajectory. Third, analysis and summarization capabilities are weak. It is impossible to quantify core characteristics such as customer activity periods and channel preferences. The resulting handover documents are mostly unstructured information piles, requiring new service personnel to spend a lot of time sorting them out, and easily overlooking deeper customer needs, ultimately leading to a decline in service experience and insufficient mining of existing customers. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method for processing historical information summaries of orphan policy customers, which can solve the problems of semantic conflicts and temporal breaks in cross-source heterogeneous data, quantify customer activity characteristics and service preferences, and improve service handover efficiency and customer service quality.

[0006] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: In a first aspect, a method for processing historical information summaries of orphan policyholders, the method comprising: If a policy status change to an orphan policy is detected, a policy monitoring status process control point is set. This control point is located at the initial trigger node of the policy status change, the instant the orphan policy status is confirmed, and the initiation node of the service handover process. Based on the matching analysis between the process control point and the monitoring time sequence nodes, a geometric control region is obtained, and this region is then divided into grids. Data aggregation adjustment values ​​are obtained based on the grid's distribution characteristics. These adjustment values ​​are then used to optimize and adjust the cross-source heterogeneous historical data. Based on the adjusted cross-source heterogeneous historical data, a customer interaction timeline is obtained through semantic alignment and temporal reconstruction. Based on the customer interaction timeline, in-depth analysis and reasoning are performed using a large language model to obtain analysis results that include a set of key service events and customer profile data. Based on the customer profile data in the analysis results, the cluster boundaries of key customer service events are defined; different service events are mapped to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; by calculating the convex hull boundary of the discrete points, the geometric boundaries of dense areas and boundary ranges of customer service interactions in the spatiotemporal dimension are identified; based on the geometric boundaries, the customer's active time periods and core service modes are quantified to obtain quantitative analysis results. Based on the quantitative analysis results, a structured service handover summary for business roles is obtained. The structured service handover summary covers basic customer information, historical interaction timeline, customer insights, and personalized service suggestions.

[0007] Furthermore, if a policy status change to an orphan policy is detected, a policy monitoring status process control point is set. This process control point is located at the initial trigger node of the policy status change, the instantaneous node confirming the orphan policy status, and the initiation node of the service handover process. Based on the matching analysis between the process control point and the monitoring time sequence nodes, a geometric control region is obtained, and this region is then divided into grids. Data aggregation adjustment values ​​are obtained based on the grid's distribution characteristics. The cross-source heterogeneous historical data is then optimized and adjusted according to these adjustment values, including: Establish process control points synchronized with the time sequence during the monitoring of policy status; the process control points are located at the initial trigger node of policy status change, the instant node of confirming the status of orphan policies, and the start node of service handover process. Map process control points to coordinate points on the monitoring timeline plane; construct polygonal geometric control regions based on these coordinate points; The polygonal geometric control region is divided into uniform grids; based on the distribution density and location characteristics of the grid cells, the data aggregation adjustment value is calculated. Based on the data collection adjustment value, the collection scope and collection priority of cross-source heterogeneous historical data are dynamically optimized to obtain the optimized collection parameters. Based on the optimized aggregation parameters, a cross-system data aggregation operation is performed to obtain an adjusted cross-source heterogeneous historical dataset.

[0008] Furthermore, based on the adjusted cross-source heterogeneous historical data, a customer interaction timeline is obtained through semantic alignment and temporal reconstruction, including: Entity parsing is performed on records from each data source in the adjusted cross-source heterogeneous historical dataset to identify core entities related to customers and policies; Based on the identified core entities, a semantic association network is established across data sources; through the semantic association network, interaction records from different sources are semantically aligned to form a semantically unified set of customer interaction records; Time information is extracted from a semantically unified set of customer interaction records to obtain structured and unstructured data; the record timestamps are directly obtained from the structured data, and natural language processing is performed on the unstructured data to identify the implicit time information, which is then standardized. Using all standardized time information as a timeline benchmark, customer interaction records are reconstructed in time sequence; based on the time order, a coherent customer interaction timeline is obtained.

[0009] Furthermore, based on the customer interaction timeline, in-depth analysis and reasoning are performed using a large language model to obtain analytical results including a set of key service events and customer profile data, including: The customer interaction timeline is input into a pre-trained large language model; based on the sequence understanding capability of the large language model, the customer interaction timeline is semantically parsed segment by segment; key semantic segments in the timeline are captured through an attention mechanism to identify service events with business significance. Service events are classified and merged in multiple dimensions to obtain a structured set of key service events; Analyze the implicit semantic relationships in a structured set of key service events; infer the customer's deeper intentions and unexpressed potential needs based on the contextual logical relationships of the event sequence. By integrating the structured set of key service events with the inferred deep customer intentions and potential needs, customer profile data containing quantitative characteristics is obtained; the customer profile data and the set of key service events together constitute the analysis results.

[0010] Furthermore, based on the customer profile data in the analysis results, clustering boundaries are defined for key customer service events; different service events are mapped to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; by calculating the convex hull boundary of the discrete points, the geometric boundaries of dense regions and boundary ranges of customer service interactions in the spatiotemporal dimension are identified; based on the geometric boundaries, customer active periods and core service patterns are quantified to obtain quantitative analysis results, including: Extract a set of key service events from the analysis results; Map each service event in the set of key service events to a two-dimensional coordinate system; the horizontal axis of the two-dimensional coordinate system represents the time of the event, and the vertical axis represents the interaction channel type code; Perform convex hull boundary calculation on a discrete point set in a two-dimensional coordinate system; construct the minimum convex polygon containing all discrete points by traversing the outer boundary of the point set. Based on the minimum convex polygon, we identify dense areas of customer service interactions in the spatiotemporal dimension; by calculating the relative positional relationship between the boundary of the minimum convex polygon and the coordinate axis, we determine the boundary range of customer active periods and core service modes. Based on the characteristics of the boundary range, the customer's active time period and core service mode are quantitatively labeled; the quantitative labeling results are then integrated with customer profile data to obtain quantitative analysis results.

[0011] Furthermore, based on the characteristics of the boundary range, customer activity periods and core service modes are quantitatively labeled; the quantitative labeling results are then integrated with customer profile data to obtain quantitative analysis results, including: The quantitative annotation results are analyzed to extract the distribution parameters of active time periods and service mode feature parameters; Based on the distribution parameters of active time periods and the characteristic parameters of service modes, a customer service behavior feature vector is constructed; the feature vector includes time distribution density, channel preference intensity and service frequency characteristics. The customer service behavior feature vector is fused with customer profile data at the feature level; the importance of various features is weighted through a weight allocation mechanism to obtain an enhanced customer profile. Based on the enhanced customer profile, customer service strategy matching suggestions are obtained; these suggestions include the best contact time, preferred communication channels, and personalized service themes. The enhanced customer profile was integrated with customer service strategy matching suggestions to obtain the final quantitative analysis results.

[0012] Furthermore, based on the quantitative analysis results, a structured service handover summary oriented towards business roles was obtained. This summary covers basic customer information, historical interaction timelines, customer insights, and personalized service recommendations, including: The final quantitative analysis results will be used to extract enhanced customer profiles and customer service strategy matching suggestions. Based on a preset service summary generation template, basic customer information, historical interaction timelines, and enhanced customer profiles are structurally integrated to form a core data block containing a customer overview and historical behavioral characteristics. Based on customer service strategy matching suggestions, personalized service recommendations are obtained; these personalized service recommendations include recommended communication time slots, preferred contact channels, and key service topics. Personalized service suggestions are combined and arranged with core data blocks; based on the reading habits of business roles, a service handover summary with a complete structure and highlighting key points is obtained; Standardize the format of the service handover summary to obtain a structured service handover summary that meets the requirements of the business system.

[0013] Secondly, a system for processing historical information summaries of orphan policyholders includes: The configuration module is used to set policy monitoring status process control points if a policy status change to an orphan policy is detected. These control points are located at the initial trigger node of the policy status change, the instant the orphan policy status is confirmed, and the start node of the service handover process. Based on the matching analysis between the process control points and monitoring time-series nodes, a geometric control region is obtained, and this region is then divided into grids. Data aggregation adjustment values ​​are obtained based on the grid's distribution characteristics. These adjustment values ​​are then used to optimize and adjust cross-source heterogeneous historical data. The processing module, based on the adjusted cross-source heterogeneous historical data, obtains the customer interaction timeline through semantic alignment and temporal reconstruction. The reasoning module is used to perform in-depth analysis and reasoning based on the customer interaction timeline using a large language model to obtain analysis results that include a set of key service events and customer profile data. The calculation module is used to define the cluster boundaries of key customer service events based on customer profile data in the analysis results; map different service events to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; identify the geometric boundaries of dense areas and boundary ranges of customer service interactions in the spatiotemporal dimension by calculating the convex hull boundary of the discrete points; and quantify customer active periods and core service modes based on the geometric boundaries to obtain quantitative analysis results. The summary module is used to generate a structured service handover summary for business roles based on the quantitative analysis results. The structured service handover summary covers basic customer information, historical interaction timeline, customer insights and personalized service suggestions.

[0014] Thirdly, a computing device, comprising: One or more processors; A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0015] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0016] The above-described solution of the present invention has at least the following beneficial effects: By employing process control points for key policy status changes (initial trigger, orphan policy confirmation, handover initiation), constructing geometric control areas, and calculating data aggregation adjustment values ​​in a grid-like manner, this approach overcomes the problems of redundant handover materials and missing key information caused by existing systems that capture all data without considering policy node screening. This achieves precise optimization of the scope and priority of cross-source data aggregation and improves data availability. Furthermore, by using techniques to establish semantic association networks through cross-source data entity parsing, standardizing (using timestamps for structured data, and processing unstructured data with NLP) time information, and reconstructing the time sequence, this approach overcomes the problems of lack of semantic alignment and time sequence integration in heterogeneous multi-channel data, making it difficult to form a coherent service trajectory. This achieves the effect of constructing a complete customer interaction timeline and helping service personnel quickly grasp the customer's historical service logic. Finally, by inputting the customer interaction timeline into a pre-trained large language model, key service points are identified through sequence understanding and attention mechanisms. This technology, by analyzing service events and inferring deeper customer intentions, overcomes the limitations of existing methods that merely list information and fail to analyze implicit customer needs. This allows for the creation of customer profiles with quantified characteristics and the accurate capture of potential customer demands. Furthermore, by mapping key service events to two-dimensional discrete points (time + channel), calculating convex hull boundaries to identify spatiotemporally dense areas, and quantifying active periods and core service modes, it overcomes the limitations of existing technologies in quantifying customer activity characteristics and channel preferences, and the lack of data support for service strategies. This allows for the identification of optimal contact times and channels, improving service accuracy. Finally, by integrating customer information and enhanced profiles based on preset templates, and combining content with standardized formats based on business role habits, it overcomes the problems of unstructured handover documents and high costs in acquiring information for service personnel. This results in the generation of structured handover summaries that highlight key points, significantly improving service handover efficiency. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating a method for processing historical information summaries of orphan policy customers, provided by an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of an orphan policy customer history information summary processing system provided by an embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0020] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a method for processing historical information summaries of orphan policy customers, the method comprising the following steps: Step 1: If a policy status change to an orphan policy is detected, a policy monitoring status process control point is set. The process control point is located at the initial trigger node of the policy status change, the instant node of confirming the orphan policy status, and the start node of the service handover process. Based on the matching analysis between the process control point and the monitoring time sequence node, a geometric control area is obtained, and the geometric control area is divided into grids. Based on the distribution characteristics of the grid, a data aggregation adjustment value is obtained. The cross-source heterogeneous historical data is optimized and adjusted according to the adjustment value. Step 2: Based on the adjusted cross-source heterogeneous historical data, a customer interaction timeline is obtained through semantic alignment and temporal reconstruction. Step 3: Based on the customer interaction timeline, conduct in-depth analysis and reasoning using a large language model to obtain analysis results including a set of key service events and customer profile data; Step 4: Based on the customer profile data in the analysis results, cluster the key customer service events and define their boundaries; map different service events to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; identify the dense areas and geometric boundaries of customer service interactions in the spatiotemporal dimension by calculating the convex hull boundary of the discrete points; quantify the customer's active time periods and core service modes based on the geometric boundaries to obtain quantitative analysis results. Step 5: Based on the quantitative analysis results, obtain a structured service handover summary for business roles. The structured service handover summary covers basic customer information, historical interaction timeline, customer insights, and personalized service suggestions.

[0021] In this embodiment of the invention, when a change in the status of an orphan policy is detected, by setting three core process control points—initial trigger, status confirmation, and handover initiation—and combining geometric control area division and gridded calculation of adjustment values, the scope and priority of cross-source heterogeneous historical data collection can be accurately optimized, avoiding information redundancy or missing key data caused by traditional full-volume crawling. Based on the adjusted data, semantic alignment and temporal reconstruction can unify the semantics and organize the temporal logic of heterogeneous data scattered across different channels (such as APP logs and customer service records), forming a coherent customer interaction timeline, solving the previous problems of data fragmentation and service trajectory breaks. The large language model provides in-depth timeline analysis, effectively identifying key service events, refining customer profiles, and uncovering deeper needs, overcoming the limitations of traditional information listing that fails to reveal potential customer demands. By mapping key events to two-dimensional discrete points of time and channel and calculating convex hull boundaries, it can quantify customer activity periods and core service models, avoiding the blindness of relying on experience to judge customer preferences. The structured service handover summary integrates basic customer information, interaction timelines, in-depth insights, and personalized suggestions, rather than being a collection of scattered information. New service personnel can quickly grasp the core content, significantly reducing information processing costs and ensuring the efficiency and continuity of service handover.

[0022] In a preferred embodiment of the present invention, step 1 above may include: Step 1.1: Establish time-synchronized process control points during policy status monitoring. These control points are located at the initial trigger node for policy status changes, the instantaneous node confirming the status of orphan policies, and the initiation node of the service handover process. Specifically, this includes: real-time monitoring of the service attribution status of all policies; when it is detected that the original service agent for a policy has left and a new service personnel has not been immediately assigned, resulting in an empty service attribution field for that policy, immediately triggering the initial trigger node record for policy status changes and generating a precise time-series marker for this node containing the change time, original agent information, and policy number; subsequently, verifying the original agent's departure approval documents and whether the customer has recently been contacted by other service personnel. Historically, when all verification conditions are met and it is confirmed that there is no service contact person for the policy, the moment the verification is completed, the node confirming the orphan policy status is marked, and the time sequence record is updated synchronously to ensure that the timestamp of this node is later than the initial trigger node, and the time sequence relationship between the two is traceable. Finally, a handover preparation instruction is sent to the service allocation management area. When the service allocation management area matches a suitable new service personnel according to the customer's region and policy type, and obtains a handover task order containing the new service personnel information and the handover task deadline, the service handover process start node is marked, and this node also comes with complete time sequence information. Throughout the entire process, the establishment of process control points is always synchronized with the time sequence monitoring system.

[0023] Step 1.2: Map the process control points to coordinate points on the monitoring timeline. Based on these coordinate points, construct a polygonal geometric control region. This includes: first, retrieving the established process control points, extracting the precise timestamp and status change association parameters corresponding to each node, and mapping this information to coordinate points on the monitoring timeline. The horizontal axis of the monitoring timeline is set to a unified monitoring time axis, accurate to the second, and the vertical axis is set to the type code value of the policy status change. The vertical axis code for the initial trigger node is 01, the vertical axis code for the instant node confirming the orphan policy status is 02, and the vertical axis code for the service handover process initiation node is 03. After mapping all coordinate points, connect all coordinate points sequentially according to the time sequence of the initial trigger node, the instant node confirming the orphan policy status, and the service handover process initiation node to form a closed polygonal geometric control region. The boundary of this geometric control region precisely defines the critical time interval and status change range from the policy entering the orphan state to the formal initiation of service handover.

[0024] Step 1.3: Perform uniform grid segmentation on the polygonal geometric control region. Based on the distribution density and location characteristics of the grid cells, calculate the data aggregation adjustment value. Specifically, this includes: performing uniform grid segmentation on the polygonal geometric control region, setting the size of the grid cells by referring to the size of the temporal span and the variation amplitude of the state coding value within the region during segmentation; for example, in the short temporal span from the initial trigger node to the instant node of confirming the orphan policy status, the time dimension of the grid cells is set to 10 seconds; in the long temporal span from the instant node of confirming the orphan policy status to the start node of the service handover process, the time dimension of the grid cells is set to 30 seconds, while maintaining all grid cells in the state coding dimension. The dimensions are standardized to ensure that each grid cell corresponds to a specific time range and a clear state coding interval. After segmentation, each grid cell is scanned to count the amount of cross-source heterogeneous data contained in each grid cell. This data includes basic policy information, recent customer consultation records, and information on former agents leaving the company, thereby determining the data distribution density of each grid cell. At the same time, the position of each grid cell within the polygonal geometric control area is recorded, and the straight-line distance between each grid cell and the three process control points is calculated. Grid cells that are closer to the control points are assigned higher position weights. Finally, the data distribution density of each grid cell is multiplied by its position weight to obtain the data aggregation adjustment value corresponding to that grid cell.

[0025] Step 1.4: Based on the data aggregation adjustment values, dynamically optimize the aggregation scope and collection priority of cross-source heterogeneous historical data to obtain optimized aggregation parameters. Specifically, this includes: collecting the data aggregation adjustment values ​​of all grid cells, sorting these adjustment values ​​from smallest to largest, and dividing them into three priority levels: high, medium, and low. The top 30% of adjustment values ​​are classified as high priority, the middle 40% as medium priority, and the bottom 30% as low priority. For grid cells corresponding to different priorities, set the data aggregation scope and collection order respectively. For the data source corresponding to a high-priority grid cell, set the aggregation scope to all data within that grid cell and its associated upstream and downstream data. For example, not only data from that grid cell but also other data from the grid cell itself is collected. For customer consultation records within the original time frame, relevant interaction data within 10 minutes before and after the consultation record is also collected, and this type of data is set as the priority collection target. For data sources corresponding to medium-priority grid units, the collection scope is limited to the core data within that grid unit, such as policy status change records, and the collection order is set to be performed after the high-priority data collection is completed. For data sources corresponding to low-priority grid units, the collection scope only retains the key identification data within that grid unit, such as customer number and policy number, and the collection order is set to be the last, collected only as needed when the high- and medium-priority data collection is complete. Through such dynamic adjustments, optimized collection parameters are obtained, including data source priority, collection scope boundaries, and collection order.

[0026] Step 1.5: Based on the optimized aggregation parameters, perform cross-system data aggregation operations to obtain the adjusted cross-source heterogeneous historical dataset. Specifically, this includes: importing the optimized aggregation parameters into the cross-system data acquisition engine; first, sending data acquisition requests to the corresponding data source systems according to the data source priorities in the aggregation parameters; for example, first sending a high-priority request for policy status change records to the policy management system, then a medium-priority request for customer consultation records to the customer service system, and finally a low-priority request for former agent departure identification data to the agent management system; during the data acquisition process, filtering the data returned by each data source according to the aggregation range boundaries in the aggregation parameters, removing redundant data exceeding the boundaries, such as excluding other customer data unrelated to the orphan policy and historical data earlier than one hour before the initial trigger node; simultaneously, performing preliminary standardization processing on the collected data of different formats, uniformly converting the date and encoding formats from different systems into a preset standard format; after completing data filtering and standardization, associating and integrating all data by customer number and policy number, ultimately forming the adjusted cross-source heterogeneous historical dataset.

[0027] In this embodiment of the invention, by establishing three core process control points synchronized with the time sequence, the entire critical node of the orphan policy from state change triggering, orphan identity confirmation to handover initiation is precisely anchored. This avoids subsequent data processing deviating from the core scenario due to missing critical nodes, providing a clear time sequence and state benchmark for accurately defining the data processing scope. Mapping the process control points to time-series plane coordinates and constructing a polygonal geometric control region can accurately define the critical time and state range of the orphan policy from entering the orphan state to initiating the handover, effectively eliminating data interference from irrelevant times and states, and solving the information redundancy problem caused by the lack of clear boundaries in traditional full-data crawling. Geometric control... By segmenting regions into grids and calculating adjustment values ​​based on grid data density and location characteristics, data collection priorities are quantified, avoiding the blind reliance on experience to judge data importance. The data collection scope and priority are dynamically optimized based on these adjustment values, with high-priority data being collected comprehensively and medium- and low-priority data streamlined as needed. This ensures that critical information such as customer inquiries and policy changes is not missing while reducing redundant data, improving the accuracy and efficiency of data collection. Based on optimized parameters, cross-system data collection and filtering of redundant data are performed, and the data is then unified in format. Finally, data is integrated by customer and policy number, resulting in a high-quality cross-source heterogeneous dataset that solves the problems of chaotic data formats and low correlation in traditional cross-system data collection.

[0028] In a preferred embodiment of the present invention, step 2 above may include: Step 2.1 involves entity parsing of records from various data sources in the adjusted cross-source heterogeneous historical dataset to identify core entities related to customers and policies. Specifically, the adjusted cross-source heterogeneous historical dataset includes records from multiple data sources, such as APP interaction logs, customer service call recordings, offline policy maintenance records, and the policy management system. For each type of data source record, keywords are extracted from the text content of the APP interaction logs to identify customer identity information such as ID number and name, as well as basic policy information such as policy number and type of insurance. Then, entity annotation is performed on the transcribed text of customer service call recordings to locate service event entities such as policy maintenance applications and claims consultations mentioned in the call, along with corresponding customer and policy identifiers. Simultaneously, structured fields and unstructured notes in the offline policy maintenance records are parsed to extract customer and policy information related to policy maintenance business type and processing time. Through this multi-dimensional parsing, identity entities directly related to customers and basic and business entities directly related to policies are selected from all data source records to form a core entity list.

[0029] Step 2.2: Based on the identified core entities, establish a cross-data source semantic association network. This network is used to semantically align interaction records from different sources, forming a semantically unified set of customer interaction records. Specifically, this involves: using core entities as the connection basis, first assigning a unique identifier code to each core entity. For example, a customer's ID number corresponds to a unique customer ID code, a policy number corresponds to a unique policy ID code, and a service event entity corresponds to a unique event type code. Records from different data sources are then bound to these identifier codes. For instance, the record of customer A inquiring about policy B terms in the APP interaction log is associated with customer A's ID code, policy B's ID code, and the event type code for the term inquiry. Using these identifier codes as network nodes, semantic connections are established between records from different data sources that are associated with the same customer, the same policy, or the same service event, forming a cross-data source semantic association network. Based on this network, semantic alignment is performed on interaction records from different sources. For example, the term inquiries in the APP log and the insurance term inquiries in the customer service records are uniformly labeled as "policy term inquiry semantic tags," ultimately integrating them to form a semantically unified set of customer interaction records.

[0030] Step 2.3 involves extracting time information from the semantically unified set of customer interaction records to obtain structured and unstructured data. For structured data, the record timestamps are directly obtained. For unstructured data, natural language processing is performed to identify implicit time information, which is then standardized. Specifically, the records in the set are first divided into structured and unstructured data based on their data type. For structured data, such as the policy maintenance processing time and policy activation time recorded in the policy management system, as well as the operation timestamps in the APP interaction logs, the time information in these fields is directly read to obtain standard format record timestamps. For unstructured data, such as the statement in the customer service call recording that someone wanted to apply for a policy loan last Wednesday, or the content of the address change materials submitted at the beginning of this month in the offline policy maintenance record notes, vague time expressions like "last Wednesday" and "the beginning of this month" are first identified. Then, combined with the current time and context information, time conversion is performed, converting "last Wednesday" into a specific year, month, and day format, and converting "the beginning of this month" into the standard date of the 1st of the current month. Finally, all converted time information is uniformly adjusted to the set format to complete the time information standardization.

[0031] Step 2.4: Using all standardized time information as the timeline benchmark, reconstruct the customer interaction records in time sequence. Based on the time sequence, integrate and sort to obtain a coherent customer interaction timeline. Specifically, this includes: using standardized time information as the sole timeline benchmark, firstly, binding each record in the semantically unified customer interaction record set with the corresponding standardized time information to form a pair of record content + standardized time; then, performing a preliminary sorting of all pairs of pairs according to the order of standardized time. For pairs of pairs with identical times, a secondary sorting is performed based on the priority of the interaction channel (e.g., offline security records have higher priority than customer service records, and customer service records have higher priority than APP logs) to ensure that records from the same time have a clear arrangement logic; after sorting, these records are sequentially linked in time sequence to form a complete customer interaction timeline. The timeline clearly presents the business, questions, or feedback requests that customers handled through a specific channel at a specific time. This coherent timeline completely restores the customer's historical service trajectory.

[0032] In this embodiment of the invention, by performing entity parsing on cross-source heterogeneous datasets, the core entities related to customers and insurance policies are accurately identified, solving the problem of chaotic entity identification and difficulty in associating them in different data sources; By constructing a semantic association network based on core entities and completing semantic alignment, the semantic differences between different data sources are effectively eliminated, integrating scattered interaction records into a semantically unified set and avoiding information fragmentation caused by semantic conflicts. Time information is processed separately for structured and unstructured data. Structured data is directly extracted with timestamps, while unstructured data is identified and its implicit time is standardized by NLP. This solves the problems of inconsistent time information formats and the lack of explicit time for some data in cross-source data, ensuring that all interaction records have a unified timeline benchmark. The interaction records are reconstructed and sorted according to the standardized time as the benchmark, ultimately forming a coherent customer interaction timeline. This solves the problems of chaotic timeline and broken service trajectories in traditional cross-source data, allowing new service personnel to clearly trace the chronological logic of customer historical interactions and quickly grasp the customer service context.

[0033] In a preferred embodiment of the present invention, step 3 above may include: Step 3.1: Input the customer interaction timeline into a pre-trained large language model; based on the sequence understanding capability of the large language model, perform segmented semantic parsing of the customer interaction timeline; capture key semantic segments in the timeline through an attention mechanism to identify service events with business significance, specifically including: dividing the coherent customer interaction timeline into segments according to natural time (e.g., by day, or by an event interval exceeding 24 hours), obtaining several continuous interactive text segments; then input these text segments into the pre-trained large language model, and analyze the semantic logic in the text segments segment by segment using sequence understanding capability, such as identifying customer interactions through... When customer service calls inquire about policy loan application requirements, the subject (customer), action (inquiry), and object (policy loan application requirements) are clearly defined. Simultaneously, an attention mechanism is used to assign higher weight to fragments containing business keywords such as claims, policy maintenance, insurance application, and loans, focusing on capturing these key semantic fragments. Then, combined with a pre-defined business event dictionary for the insurance industry (including event type definitions such as consultation, processing, and feedback), the key semantic fragments are matched as service events with business significance. For example, inquiring about policy loan application requirements is marked as a policy loan consultation event, and submitting an address change application is marked as a customer information maintenance event.

[0034] Step 3.2 involves multi-dimensional classification and merging of service events to obtain a structured set of key service events. This includes: collecting all service events; firstly, classifying them by business event type into three main categories: consultation (e.g., policy terms consultation), processing (e.g., policy maintenance application), and feedback (e.g., complaints about claims delays); further subdividing each main category into subcategories (e.g., consultation into product consultation and process consultation); next, supplementing the classification by interaction channel, labeling each event with corresponding channel information (e.g., APP, customer service hotline); then, classifying by associated policy, binding events to corresponding policy numbers (e.g., an event associated with a customer's critical illness insurance policy or annuity policy); after classification, merging duplicate events for the same customer, of the same type, and within the same time period. For example, if a customer inquires about the claims progress of the same policy three times within a week via customer service hotline, merging these into one claims progress consultation event, labeling the number of consultations and the consultation time period; finally, organizing the classified and merged events according to the field format of event type, occurrence time, interaction channel, associated policy, and event details to form a structured set of key service events.

[0035] Step 3.3 involves analyzing the implicit semantic relationships within the structured set of key service events. Based on the contextual logic of the event sequence, the client's deeper intentions and unexpressed potential needs are inferred. This includes: retrieving the structured set of key service events, arranging the events chronologically to construct an event sequence (e.g., January: policy terms consultation event, February: education fund product consultation event); then analyzing the implicit semantic relationships within the event sequence, such as recognizing that both the education fund product consultation and the consultation on children's insurance age revolve around the theme of children's protection, indicating a clear semantic connection; next, reasoning is based on the contextual logic of the event sequence. For example, the client's continuous consultation on children-related insurance products for three months without mentioning their own protection needs suggests that their deeper intention is to purchase insurance protection for their children; the client's subsequent inquiry about policy cash value after consulting about policy loans suggests that their unexpressed potential need is for cash flow and a focus on the liquidity of policy assets; and combining this with the client's basic information (such as age and family structure) to aid in the reasoning. For instance, if the client is middle-aged and has children, the inference of their intention to purchase insurance for their children is strengthened.

[0036] Step 3.4 integrates the structured set of key service events with the inferred customer deep intentions and potential needs to obtain customer profile data containing quantitative features. The customer profile data and the set of key service events together constitute the analysis results, specifically including: extracting quantitative information from the structured set of key service events, such as 5 consultation events in the past six months (3 of which were related to children's insurance), 2 processing events (1 policy maintenance application and 1 premium payment), and 0 feedback events; then extracting the inferred customer deep intentions (such as purchasing insurance protection for children) and potential needs (such as focusing on education savings insurance products and cash flow needs); subsequently, integrating this information with the customer's basic information to construct customer profile data containing quantitative features; the behavioral feature field in the profile data records the quantitative information of key events (such as the number of consultations and the proportion of event types), the demand feature field records deep intentions and potential needs (such as strong demand for children's protection and potential demand for education savings insurance), and the basic feature field records the customer's basic information and policy information; finally, the customer profile data and the set of structured key service events are packaged together to form the analysis results, which retains specific event records and includes in-depth needs insights.

[0037] The construction and training process of the large language model is as follows: We collected multi-source data related to orphan policies within the insurance industry, including historical policy service records, customer-agent interaction texts (such as customer service call transcripts and APP consultation records), policy status change documents (such as orphan policy confirmation documents triggered by agent resignation), and service handover cases. This data, consistent with the adjusted cross-source heterogeneous historical dataset, constitutes the original corpus for model training. Simultaneously, this data was cleaned to remove duplicate and invalid information, ensuring the corpus contains complete customer interaction trajectories, business event descriptions, and service handover logic, laying a data foundation for the model's understanding of industry scenarios.

[0038] Based on the need for sequence analysis of customer interaction timelines, the Transformer architecture, with its strong sequence understanding capabilities, was chosen as the basic model framework. The architecture focuses on strengthening the attention mechanism module, enabling it to accurately capture key semantic segments in the text related to business events (such as claims inquiries and policy maintenance applications). It also adapts to the semantic association analysis needs of cross-source data, ensuring the model can handle semantic differences between texts from different sources.

[0039] The prepared insurance industry corpus was input into the base model for large-scale pre-training. During training, the focus was on optimizing the model's ability to recognize insurance terminology (such as orphan policies, policy maintenance, and claims) and its ability to classify business events in customer interaction texts (such as distinguishing between consultation and processing events). Simultaneously, by setting a sequence prediction task, the model learned the temporal logic of customer interaction records (such as the chronological relationship of consulting terms before applying for policy maintenance), enabling it to initially understand the timeline of customer interactions and laying the foundation for subsequent fine-tuning.

[0040] After pre-training, the model's event recognition capabilities are first trained using labeled key service event data as fine-tuning samples. This enables the model to automatically extract business-meaning service events from customer interaction texts, matching the accuracy requirements for event recognition in the invention. Secondly, sample data for inferring deeper customer intentions (such as data containing the correspondence between multiple inquiries about children's insurance and the intention to configure insurance for children) is introduced. The model is trained through causal inference tasks to improve its ability to uncover potential customer needs from event sequences. Finally, considering semantic alignment requirements, samples with synonymous expressions from different data sources (such as policy consultations and insurance policy inquiries) are used for training to enhance the model's semantic consistency capabilities.

[0041] After fine-tuning, model validation and optimization were performed. Validation data was taken from customer cases not used in the invention's training, including cross-source heterogeneous datasets, customer interaction timelines, and manually annotated key events and demand inference results. Consistency was verified by comparing the model's output of event recognition results, semantic alignment results, and demand inference results with the manually annotated data. If the model's recognition accuracy for a certain type of business event was low, samples of that type of event were added for further fine-tuning; if there were deviations in semantic alignment, the training weights of synonymous expression samples were increased until the model performance met the accuracy requirements for data analysis.

[0042] When deploying and iterating the model, the optimized model is integrated into the invention's system workflow for stages such as customer interaction timeline analysis and customer demand inference. In practical applications, new data processed by the model (such as newly added customer interaction records and manually corrected event annotations) is automatically collected. This data is periodically added to the training sample library, and the fine-tuning and verification process is repeated to ensure that the model continuously adapts to the ever-changing business scenarios in the handover of orphan policy services (such as the emergence of new interaction channels and the generation of new business events), ensuring that it can always provide reliable support for data analysis in the invention.

[0043] In this embodiment of the invention, by leveraging the sequence understanding and attention mechanism of a pre-trained large language model, it is possible to analyze and capture key semantic fragments segment by segment of a coherent customer interaction timeline, accurately identifying service events with business significance such as claims consultation and policy maintenance applications. This solves the problem of simply listing data and being unable to filter core business events from massive interaction records. The identified service events are classified and merged in multiple dimensions, organizing scattered events into a structured set of key service events, avoiding the chaotic and disordered state of events, and solving the problem of lack of organization and difficulty in quick sorting of events in traditional processing. By analyzing the implicit semantic relationships and contextual logic of the key service event set, it is possible to infer the customer's deep intentions and unexpressed potential needs from surface events, breaking through the limitations of the background technology in being unable to perceive the customer's implicit needs, and enabling a deeper understanding of the customer. Integrating the structured event set with the inferred deep intentions and potential needs forms customer profile data containing quantitative features, solving the problem that traditional customer files only list basic information and lack depth and quantitative support. Moreover, the analysis results composed of the profile and the event set provide comprehensive data support for quantifying customer characteristics and generating personalized service suggestions.

[0044] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1 involves extracting a set of key service events from the analysis results. This includes: first, retrieving the analysis results, which contain a structured set of key service events and customer profile data; second, separating the set of key service events from the analysis results based on preset event field identifiers (such as event type, occurrence time, and interaction channel), filtering out non-event data such as basic information and demand characteristics from the customer profile data; third, performing a completeness check on the extracted set of key service events to ensure that each event contains the core information required for subsequent steps, namely, a clear event occurrence time and corresponding interaction channel. If any event has missing information, it is automatically traced back to the original record for supplementation and improvement; through this process, a clean and complete set of key service events is obtained.

[0045] Step 4.2: Map each service event in the set of key service events to a two-dimensional coordinate system. The horizontal axis of the two-dimensional coordinate system represents the event occurrence time, and the vertical axis represents the interaction channel type code. Specifically, this includes: taking the extracted set of key service events as input, first defining the parameter rules of the two-dimensional coordinate system; the horizontal axis is set as the event occurrence time axis, with the unit being hours, and the value range covering 24 hours of a day. If an event occurs across multiple days, the same hour of each day is regarded as the same horizontal axis value (e.g., Monday 9:00 and Tuesday 9:00 both correspond to the horizontal axis 9); the vertical axis is set as the interaction channel type code axis, assigning a unique integer code to different interaction channels, with APP corresponding to code 1, customer service phone corresponding to code 2, offline outlet corresponding to code 3, and SMS notification corresponding to code 4; then, traversing each event in the set of key service events, mapping the occurrence time of each event to a horizontal axis value, and converting the interaction channel into the corresponding vertical axis code, forming independent discrete coordinate points. For example, if an event occurs at 9:00 and is interacted through a customer service phone, it is mapped to the coordinate point (9, 2); through this mapping, the abstract service events are transformed into intuitive spatiotemporal coordinate data.

[0046] Step 4.3: Perform convex hull boundary calculation on the discrete point set in the two-dimensional coordinate system. By traversing the outer boundary of the point set, construct the minimum convex polygon containing all discrete points. Specifically, this includes: collecting all discrete coordinate points to form a discrete point set, and performing convex hull boundary calculation on the discrete point set. During the calculation, first find the point with the smallest x-coordinate and y-coordinate in the discrete point set as the starting point, and then sort the remaining points clockwise according to the polar angle of the line connecting them to the starting point. Next, select points sequentially from the sorted point set and form line segments with the points already on the convex hull boundary. Filter boundary points by judging the direction (convex or concave) of the line segments. If the newly formed line segment results in a concave overall shape, discard the intermediate points and retain only the points that maintain the convex direction as boundary points. By continuously traversing the outer boundary of the point set and filtering boundary points, finally connect all boundary points that meet the conditions to construct the minimum convex polygon containing all discrete points.

[0047] Step 4.4: Based on the minimum convex polygon, identify dense regions of customer service interaction in the spatiotemporal dimension; by calculating the relative positional relationship between the boundary of the minimum convex polygon and the coordinate axes, determine the boundary range of customer active periods and core service modes. Specifically, this includes: using the minimum convex polygon as the analysis object, by statistically analyzing the distribution of discrete points inside and on the boundary of the minimum convex polygon, locating areas with high point density, i.e., areas where the number of discrete points per unit area exceeds 1.5 times the average density, marking these areas as dense regions of customer service interaction in the spatiotemporal dimension (e.g., the area within the convex polygon with horizontal coordinates 9-11 and vertical coordinate 2, containing 5 discrete points, far exceeding the average density, is a dense region); Next, the relative positional relationship between the boundary of the smallest convex polygon and the coordinate axes is calculated: For the x-axis, the minimum and maximum x-coordinate values ​​of all points on the boundary of the convex polygon are extracted, and the interval formed by the two is the boundary range of the customer's active time period (e.g., minimum x-coordinate 8, maximum x-coordinate 12, corresponding to the active time period 8:00-12:00); For the y-coordinate, the y-coordinate codes corresponding to all points on the boundary of the convex polygon and inside are counted, and the code with the highest frequency is found. The interaction channel corresponding to this code is the boundary range of the core service mode (e.g., code 2 corresponds to customer service telephone, and the core service mode is to prioritize interaction through customer service telephone); Through this process, the geometric features are transformed into specific customer behavior features.

[0048] Step 4.5: Quantitatively label customer active time periods and core service modes based on the characteristics of the boundary range; integrate the quantitative labeling results with customer profile data to obtain quantitative analysis results, specifically including: first, quantitatively labeling the characteristics of the determined boundary range: for active time periods, label the specific time intervals and calculate the event percentage within that time period (e.g., events from 8:00 to 12:00 account for 60% of the total events of the day); for core service modes, label the core channels (e.g., customer service hotline) and the event percentage of that channel (e.g., customer service hotline interactions account for 80% of the total events), and label secondary channels and their corresponding percentages; then, retrieve customer profile data and add the quantitative labeling results to the behavioral quantitative feature fields of the customer profile, such as adding active time periods to the profile: weekday morning to afternoon (event percentage 60%), afternoon to afternoon (event percentage 30%); core service mode: customer service hotline (percentage 80%), secondary channel: APP (percentage 15%); after integration, use the customer profile data containing quantitative labeling information as the quantitative analysis result, which retains the deep needs insight of the customer profile and adds quantifiable behavioral features.

[0049] In this embodiment of the invention, extracting a set of key service events from the analysis results allows for focusing on the core objects of subsequent quantitative analysis, avoiding analytical bias caused by the inclusion of non-event data such as customer profiles, and providing clean analytical material for transforming events into calculable spatiotemporal data. Mapping each service event to a two-dimensional time + channel coordinate system transforms abstract event occurrence information into intuitive discrete coordinate points, clearly presenting the distribution patterns of events in the spatiotemporal dimension. This lays the data form foundation for locating core interaction areas using geometric methods, solving the problem of unvisualized spatiotemporal distribution and difficulty in capturing patterns in traditional analysis. Furthermore, calculating the convex hull boundary of the discrete points and constructing the minimum... Small convex polygons can automatically define the core distribution range of all event points, eliminating the interference of individual abnormal event points on the analysis results, and ensuring that the identified active areas and core patterns reflect the mainstream behavior of customers. Based on the identification of spatiotemporally dense areas and the determination of boundary ranges using the smallest convex polygons, the relative position of the polygons and coordinate axes can accurately quantify the customer's active time periods and core service modes. By integrating the quantitative annotation results with customer profile data, the customer profile is given new quantitative features such as active time periods and core channels, upgrading the profile from qualitative description to quantitative support. At the same time, the resulting quantitative analysis results provide accurate data basis for the subsequent generation of personalized service suggestions.

[0050] In a preferred embodiment of the present invention, step 4.5 above may include: Step 4.51 involves parsing the quantitative annotation results to extract active time period distribution parameters and service mode characteristic parameters. Specifically, this includes: the quantitative annotation results containing the boundary range of customer active time periods and relevant information about core service modes; decomposing the active time period information in the quantitative annotation results to extract active time period distribution parameters, specifically including the type of each active time period (e.g., weekday time periods, weekend time periods) and the relative proportion of different time periods (e.g., the proportion of events in a certain time period is higher than that in other time periods); simultaneously, parsing the core service mode information to extract service mode characteristic parameters, including the type of core interaction channels and the comparison of the usage intensity of core channels and secondary channels; through this parsing, the originally generalized quantitative annotation results are transformed into parameter data that can be used for analysis.

[0051] Step 4.52: Based on the active time period distribution parameters and service mode characteristic parameters, construct a customer service behavior feature vector. The feature vector includes time distribution density, channel preference intensity, and service frequency characteristics. Specifically, based on the active time period distribution parameters and service mode characteristic parameters, firstly, determine the time distribution density parameter. Based on the relationship between the relative number of events and the length of each active time period, classify the time distribution density into levels (e.g., high density, medium density, low density), which serves as the first part of the feature vector. Next, determine the channel preference intensity parameter. Based on the comparison of the usage intensity of core channels and secondary channels, classify the channel preference intensity into levels (e.g., strong preference, medium preference), and record the types of core channels and secondary channels, forming the second part of the feature vector. Finally, determine the service frequency characteristic parameter. Based on the relative number of service events per unit time, classify the service frequency into levels (e.g., high frequency, medium frequency, low frequency), and record the uniformity of event distribution per unit time, which serves as the third part of the feature vector. Combine these three parts of parameters in the order of time distribution density, channel preference intensity, and service frequency characteristics to form a complete customer service behavior feature vector.

[0052] Step 4.53 involves feature-level fusion of the customer service behavior feature vector and customer profile data. An enhanced customer profile is obtained by weighting various features according to their importance using a weighting mechanism. This includes: first, retrieving the customer service behavior feature vector, and then retrieving the customer profile data (containing basic customer information, deeper intentions, and potential needs). First, based on the needs of the service handover scenario, the weight priority of various features is determined, with the weight of the customer service behavior feature vector higher than that of the basic information in the customer profile, and the weight of deeper intentions and potential needs higher than that of non-critical content in the basic information. Then, the features in both types of data are standardized, converting all features into a unified description level (e.g., high, medium, low). After standardization, each feature is comprehensively integrated according to the set weight priority to form an enhanced customer profile containing key feature annotations.

[0053] Step 4.54: Based on the enhanced customer profile, obtain customer service strategy matching suggestions. These suggestions include optimal contact time periods, preferred communication channels, and personalized service themes. Specifically, based on the enhanced customer profile, firstly, select the most active time periods from the time distribution density characteristics of the enhanced customer profile, designate these periods as the optimal contact time periods, and mark the relative success rate of customer interactions within these periods. Next, extract the interaction channels with the highest channel preference intensity from the channel preference intensity characteristics, designate them as preferred communication channels, and supplement the relative response efficiency of these channels. Finally, combine the deep intentions and potential needs in the enhanced customer profile, match them with the insurance industry's pre-set service theme library, select personalized service themes highly relevant to the needs, and mark the core question types that customers may care about under these themes. Integrate the optimal contact time periods, preferred communication channels, and personalized service themes to form complete customer service strategy matching suggestions.

[0054] Step 4.55 integrates and analyzes the enhanced customer profile with the customer service strategy matching suggestions to obtain the final quantitative analysis results. Specifically, this includes: first, verifying the consistency between the key features in the enhanced customer profile (such as high time distribution density periods, high channel preference intensity channels, and core needs) and the corresponding content in the service strategy suggestions (such as optimal contact time, preferred communication channels, and personalized service themes) to ensure that the profile features and strategy suggestions are logically matched; if there are inconsistencies, the previous steps are re-checked to verify the parameter processing process and correct the deviations; after the verification is passed, the two types of information are structured and arranged in the logical order of customer basic information, core behavioral characteristics, in-depth needs insights, and service strategy suggestions; after the arrangement is completed, the final quantitative analysis results containing these structured contents are obtained, which have both the analytical depth of customer characteristics and clear service execution suggestions.

[0055] In this embodiment of the invention, the quantitative annotation results are analyzed to extract active time period distribution parameters and service mode feature parameters, transforming the fuzzy boundary range into clear quantitative data, thus solving the problem that active time periods and service modes can only be qualitatively described. Based on the extracted parameters, a customer service behavior feature vector containing time distribution density, channel preference intensity, and service frequency characteristics is constructed, integrating the scattered quantitative parameters into a structured feature system to avoid fragmented and disordered behavioral features. The customer service behavior feature vector is fused with the customer profile at the feature level, highlighting key information through a weight allocation mechanism to form an enhanced customer profile. This makes the profile no longer an information pile, but a comprehensive description with focus and hierarchy, solving the problems of traditional customer profiles lacking hierarchy and practicality. Based on the enhanced customer profile, service strategy suggestions including optimal contact time, preferred communication channels, and personalized service themes are obtained, directly providing new service personnel with a clear service direction, avoiding service personnel spending a lot of time sorting through information to determine strategies. The enhanced customer profile and service strategy suggestions are integrated into the final quantitative analysis result, which retains the deep characteristics of customers and includes clear service execution suggestions, combining analytical depth and practical value, solving the problems of unstructured handover documents and difficulty in obtaining key service suggestions.

[0056] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1: Extract enhanced customer profiles and customer service strategy matching suggestions from the final quantitative analysis results; Step 5.2: Based on the preset service summary template, the customer's basic information, historical interaction timeline and enhanced customer profile are structurally integrated to form a core data block containing customer overview and historical behavioral characteristics; Step 5.3: Based on the customer service strategy matching suggestions, obtain personalized service suggestions; the personalized service suggestions specifically include recommended communication time slots, preferred contact channels, and key service topics; Step 5.4: Combine and arrange personalized service suggestions with core data blocks; based on the reading habits of business roles, obtain a service handover summary with a complete structure and highlighting key points; Step 5.5: Standardize the format of the service handover summary to obtain a structured service handover summary that meets the requirements of the business system, specifically including: In this embodiment of the invention, enhanced customer profiles and service strategy suggestions are extracted, focusing on core information for service handover and avoiding interference from irrelevant data. This addresses the problems of cluttered handover information and lack of emphasis on core content in the background technology. Customer information, historical timelines, and enhanced profiles are integrated based on templates to form structured core data blocks, systematically organizing scattered information and making customer overviews and historical behavioral characteristics readily apparent. This solves the problems of fragmented information and difficulty in quickly grasping the full picture of customers in traditional handovers. Personalized service suggestions (recommended time periods, channels, and themes) are extracted from strategy suggestions, transforming abstract strategies into concrete and actionable service points. This provides new service personnel with clear action guidelines, avoiding problems of ambiguous service direction and high trial-and-error costs. The core data blocks and personalized suggestions are combined and arranged according to the reading habits of business roles to form a complete and focused handover summary, enabling new service personnel to quickly locate key information. The summary is standardized in format to meet the requirements of the business system, facilitating storage, retrieval, and circulation, improving the standardization and efficiency of the handover process, and solving the problems of difficult system management and poor reusability of traditional unstructured documents.

[0057] like Figure 2 As shown, embodiments of the present invention also provide a historical information summary processing system for orphan policy customers, comprising: The configuration module is used to set policy monitoring status process control points if a policy status change to an orphan policy is detected. These control points are located at the initial trigger node of the policy status change, the instant the orphan policy status is confirmed, and the start node of the service handover process. Based on the matching analysis between the process control points and monitoring time-series nodes, a geometric control region is obtained, and this region is then divided into grids. Data aggregation adjustment values ​​are obtained based on the grid's distribution characteristics. These adjustment values ​​are then used to optimize and adjust cross-source heterogeneous historical data. The processing module, based on the adjusted cross-source heterogeneous historical data, obtains the customer interaction timeline through semantic alignment and temporal reconstruction. The reasoning module is used to perform in-depth analysis and reasoning based on the customer interaction timeline using a large language model to obtain analysis results that include a set of key service events and customer profile data. The calculation module is used to define the cluster boundaries of key customer service events based on customer profile data in the analysis results; map different service events to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; identify the geometric boundaries of dense areas and boundary ranges of customer service interactions in the spatiotemporal dimension by calculating the convex hull boundary of the discrete points; and quantify customer active periods and core service modes based on the geometric boundaries to obtain quantitative analysis results. The summary module is used to generate a structured service handover summary for business roles based on the quantitative analysis results. The structured service handover summary covers basic customer information, historical interaction timeline, customer insights and personalized service suggestions.

[0058] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for processing historical information summaries of orphan insurance policy customers, characterized in that, The method includes: If a policy status change to an orphan policy is detected, a policy monitoring status process control point is set. This control point is located at the initial trigger node of the policy status change, the instant the orphan policy status is confirmed, and the initiation node of the service handover process. Based on the matching analysis between the process control point and the monitoring time sequence nodes, a geometric control region is obtained, and this region is then divided into grids. Data aggregation adjustment values ​​are obtained based on the grid's distribution characteristics. These adjustment values ​​are then used to optimize and adjust the cross-source heterogeneous historical data. Based on the adjusted cross-source heterogeneous historical data, a customer interaction timeline is obtained through semantic alignment and temporal reconstruction. Based on the customer interaction timeline, in-depth analysis and reasoning are performed using a large language model to obtain analysis results that include a set of key service events and customer profile data. Based on the customer profile data in the analysis results, the cluster boundaries of key customer service events are defined; different service events are mapped to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; by calculating the convex hull boundary of the discrete points, the geometric boundaries of dense areas and boundary ranges of customer service interactions in the spatiotemporal dimension are identified; based on the geometric boundaries, the customer's active time periods and core service modes are quantified to obtain quantitative analysis results. Based on the quantitative analysis results, a structured service handover summary for business roles is obtained. The structured service handover summary covers basic customer information, historical interaction timeline, customer insights, and personalized service suggestions.

2. The method for processing historical information summaries of orphan policy customers according to claim 1, characterized in that, If a policy status change to an orphan policy is detected, a policy monitoring status process control point is set; the process control point is at the initial trigger node of the policy status change, the instant node of confirming the orphan policy status, and the start node of the service handover process. Based on the matching analysis between process control points and monitoring time sequence nodes, the geometric control region is obtained, and the geometric control region is divided into grids. Data aggregation adjustment values ​​are obtained based on the distribution characteristics of the grid; The cross-source heterogeneous historical data was optimized and adjusted based on the adjustment values, including: Establish process control points synchronized with the time sequence during the monitoring of policy status; the process control points are located at the initial trigger node of policy status change, the instant node of confirming the status of orphan policies, and the start node of service handover process. Map process control points to coordinate points on the monitoring timeline plane; construct polygonal geometric control regions based on these coordinate points; The polygonal geometric control region is divided into uniform grids; based on the distribution density and location characteristics of the grid cells, the data aggregation adjustment value is calculated. Based on the data collection adjustment value, the collection scope and collection priority of cross-source heterogeneous historical data are dynamically optimized to obtain the optimized collection parameters. Based on the optimized aggregation parameters, a cross-system data aggregation operation is performed to obtain an adjusted cross-source heterogeneous historical dataset.

3. The method for processing historical information summaries of orphan policy customers according to claim 2, characterized in that, Based on the adjusted cross-source heterogeneous historical data, a customer interaction timeline is obtained through semantic alignment and temporal reconstruction, including: Entity parsing is performed on records from each data source in the adjusted cross-source heterogeneous historical dataset to identify core entities related to customers and policies; Based on the identified core entities, a semantic association network is established across data sources; through the semantic association network, interaction records from different sources are semantically aligned to form a semantically unified set of customer interaction records; Time information is extracted from a semantically unified set of customer interaction records to obtain structured and unstructured data; the record timestamps are directly obtained from the structured data, and natural language processing is performed on the unstructured data to identify the implicit time information, which is then standardized. Using all standardized time information as a timeline benchmark, customer interaction records are reconstructed in time sequence; based on the time order, a coherent customer interaction timeline is obtained.

4. The method for processing historical information summaries of orphan policy customers according to claim 3, characterized in that, Based on the customer interaction timeline, in-depth analysis and reasoning are performed using a large language model to obtain analytical results including a set of key service events and customer profile data, including: The customer interaction timeline is input into a pre-trained large language model; based on the sequence understanding capability of the large language model, the customer interaction timeline is semantically parsed segment by segment; key semantic segments in the timeline are captured through an attention mechanism to identify service events with business significance. Service events are classified and merged in multiple dimensions to obtain a structured set of key service events; Analyze the implicit semantic relationships in a structured set of key service events; infer the customer's deeper intentions and unexpressed potential needs based on the contextual logical relationships of the event sequence. By integrating the structured set of key service events with the inferred deep customer intentions and potential needs, customer profile data containing quantitative characteristics is obtained; the customer profile data and the set of key service events together constitute the analysis results.

5. The method for processing historical information summaries of orphan policy customers according to claim 4, characterized in that, Based on the customer profile data in the analysis results, the cluster boundaries of key customer service events are defined; different service events are mapped to discrete points on a two-dimensional plane according to their occurrence time and interaction channels; by calculating the convex hull boundary of the discrete points, the geometric boundaries of dense areas and boundary ranges of customer service interactions in the spatiotemporal dimension are identified. Based on geometric boundaries, customer activity periods and core service patterns are quantified, yielding quantitative analysis results, including: Extract a set of key service events from the analysis results; Map each service event in the set of key service events to a two-dimensional coordinate system; the horizontal axis of the two-dimensional coordinate system represents the time of the event, and the vertical axis represents the interaction channel type code; Perform convex hull boundary calculation on a discrete point set in a two-dimensional coordinate system; construct the minimum convex polygon containing all discrete points by traversing the outer boundary of the point set. Based on the minimum convex polygon, we identify dense areas of customer service interactions in the spatiotemporal dimension; by calculating the relative positional relationship between the boundary of the minimum convex polygon and the coordinate axis, we determine the boundary range of customer active periods and core service modes. Based on the characteristics of the boundary range, the customer's active time period and core service mode are quantitatively labeled; the quantitative labeling results are then integrated with customer profile data to obtain quantitative analysis results.

6. The method for processing historical information summaries of orphan policy customers according to claim 5, characterized in that, Based on the characteristics of the boundary range, the customer's active time period and core service mode are quantitatively labeled; By integrating the quantitative annotation results with customer profile data, quantitative analysis results are obtained, including: The quantitative annotation results are analyzed to extract the distribution parameters of active time periods and service mode feature parameters; Based on the distribution parameters of active time periods and the characteristic parameters of service modes, a customer service behavior feature vector is constructed; the feature vector includes time distribution density, channel preference intensity and service frequency characteristics. The customer service behavior feature vector is fused with customer profile data at the feature level; the importance of various features is weighted through a weight allocation mechanism to obtain an enhanced customer profile. Based on the enhanced customer profile, customer service strategy matching suggestions are obtained; these suggestions include the best contact time, preferred communication channels, and personalized service themes. The enhanced customer profile was integrated with customer service strategy matching suggestions to obtain the final quantitative analysis results.

7. The method for processing historical information summaries of orphan policy customers according to claim 6, characterized in that, Based on the quantitative analysis results, a structured service handover summary oriented towards business roles was obtained. This summary covers basic customer information, a timeline of historical interactions, customer insights, and personalized service recommendations, including: The final quantitative analysis results will be used to extract enhanced customer profiles and customer service strategy matching suggestions. Based on a preset service summary generation template, basic customer information, historical interaction timelines, and enhanced customer profiles are structurally integrated to form a core data block containing a customer overview and historical behavioral characteristics. Based on customer service strategy matching suggestions, personalized service recommendations are obtained; these personalized service recommendations include recommended communication time slots, preferred contact channels, and key service topics. Personalized service suggestions are combined and arranged with core data blocks; based on the reading habits of business roles, a service handover summary with a complete structure and highlighting key points is obtained; Standardize the format of the service handover summary to obtain a structured service handover summary that meets the requirements of the business system.

8. A system for processing historical information summaries of orphan policyholders, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The settings module is used to set the policy monitoring status flow control point if a policy status change to an orphan policy is detected. The process control points are the initial trigger node for policy status change, the instant node for confirming the status of orphan policies, and the initiation node for the service handover process; Based on the matching analysis between process control points and monitoring time sequence nodes, the geometric control region is obtained, and the geometric control region is divided into grids. Data aggregation adjustment values ​​are obtained based on the distribution characteristics of the grid; The cross-source heterogeneous historical data were optimized and adjusted based on the adjustment values; The processing module, based on the adjusted cross-source heterogeneous historical data, obtains the customer interaction timeline through semantic alignment and temporal reconstruction. The reasoning module is used to perform in-depth analysis and reasoning based on the customer interaction timeline using a large language model to obtain analysis results that include a set of key service events and customer profile data. The calculation module is used to cluster key customer service events based on customer profile data in the analysis results; map different service events into discrete points on a two-dimensional plane according to their occurrence time and interaction channels; and identify the geometric boundaries of dense areas and boundary ranges of customer service interactions in the spatiotemporal dimension by calculating the convex hull boundary of the discrete points. Based on geometric boundaries, customer active periods and core service modes are quantified to obtain quantitative analysis results; The summary module is used to generate a structured service handover summary for business roles based on the quantitative analysis results. The structured service handover summary covers basic customer information, historical interaction timeline, customer insights and personalized service suggestions.

9. A computing device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a program that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

Images