Insurance customer feature vector updating method based on dynamic adjustment of spatio-temporal correlation

Abstract

The application belongs to the technical field of feature updating in the insurance field, more specifically, relates to an insurance customer feature vector updating method based on dynamic adjustment of space-time correlation; the method collects customer core data, external space-time data and real-time data flow, establishes a space-time correlation matrix, and enhances the customer feature vector from the time and space dimensions through dynamic feature extraction; on this basis, the features are weighted using time decay and space decay functions, and the importance of the space-time dimensions is adjusted using a multi-head attention mechanism to ensure the timeliness and spatial adaptability of the features; further, through an event-driven and gradual updating mechanism, the updating frequency is controlled and a versioned feature vector is generated, and finally, the decay parameters are adjusted through an optimization strategy to improve the effect of feature updating and the accuracy of business decisions; the method effectively overcomes the problem of insufficient consideration of space-time correlation in traditional methods, and enhances the real-time and accuracy of insurance decisions.

Description

Technical Field

[0001] The present application belongs to the technical field of feature updating in the insurance field, and more specifically, to a method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation. Background Art

[0002] With the continuous advancement of technology, the insurance industry is gradually moving into a data-driven, intelligent era. Traditional insurance businesses rely on manual assessments and fixed rules, while modern insurance companies are increasingly relying on big data, artificial intelligence, and machine learning technologies to improve efficiency, reduce risks, optimize the customer experience, and promote refined management. The insurance industry faces a massive influx of information flows, customer data, and event data. Extracting effective information from this massive data to conduct accurate risk assessment and pricing is a key issue in this industry transformation.

[0003] Insurance companies often need to dynamically adjust policies, pricing, and claims processes based on multiple factors, including historical customer information, real-time dynamics, and the external environment. However, customer behavior, market conditions, and external conditions are constantly changing, making traditional analytical methods based on static data inadequate for refined management and timely decision-making. In particular, timeliness and spatial factors are becoming increasingly important in the insurance business. For example, changes in customer lifestyle, geographic location, and social events all have a profound impact on the insurance business.

[0004] Currently, the insurance industry has developed several feature extraction and updating methods based on data analysis. These methods typically focus on extracting fixed features from historical customer data and performing static analysis. For example, some models employ traditional machine learning algorithms (such as decision trees and support vector machines) to assess and classify customer risk. These models analyze past customer behavior data (such as insurance application and claims records) to predict future customer behavior or risk profiles.

[0005] However, traditional methods mostly focus on single-dimensional data analysis (for example, analysis based only on time or space), and fail to fully consider the temporal and spatial variability of customer behavior. In fact, the changes in customer behavior in time and space entropy are highly correlated. For example, customer consumption behavior and claim frequency can be significantly affected by external environments (such as holidays, economic changes, and geographical location). Summary of the Invention

[0006] The present invention provides an insurance customer feature vector updating method based on dynamic adjustment of spatiotemporal correlation, which aims to solve the technical problem that spatiotemporal correlation is not sufficiently considered in the insurance field.

[0007] The method for updating the insurance customer feature vector based on dynamic adjustment of spatiotemporal correlation includes the following steps:

[0008] Step 1: Collect core customer data, external spatiotemporal data, and real-time data streams, standardize the spatiotemporal coordinates of the collected data, and unify them into a standard grid system. Attach spatiotemporal tags to each business event to ensure that each event clearly corresponds to its time and spatial location. Also, set timeliness tags for the data to obtain a spatiotemporal correlation matrix between customers and events.

[0009] Step 2: Dynamic feature extraction is performed based on the spatiotemporal correlation matrix and the customer's historical feature vectors. Periodic feature analysis is performed from the time dimension. The event density change rate is calculated to quantify the trend within a short time window. The entropy of the mobility pattern and the migration of spatial hotspots are extracted from the spatial dimension to obtain a spatiotemporal enhanced feature tensor, which includes spatiotemporal features and corresponding timeliness labels.

[0010] Step 3: Based on the spatiotemporal enhanced feature tensor, time and space are attenuated. Time and space attenuation functions are used to weight timeliness and spatial distance respectively. Then, a multi-head attention mechanism is used to calculate the importance weights of the spatiotemporal dimensions. The feature vector is dynamically adjusted to obtain a feature vector with spatiotemporal weights.

[0011] Step 4: Based on the feature vectors with spatiotemporal weights, an event-driven and gradual update mechanism is constructed. A spatiotemporal attenuation threshold is set to control the update frequency. A combined update strategy is used to balance the changes in feature vectors and time information. A versioned feature vector is generated and an update log is recorded.

[0012] Step 5: Monitor the impact of feature updates on core business indicators, analyze the relationship between feature freshness and business indicators, and use the strategy optimizer to dynamically adjust the time decay parameters and update strategy to obtain the optimized update strategy parameters.

[0013] The present invention establishes a spatiotemporal correlation matrix by collecting customer core data, external spatiotemporal data and real-time data streams, and enhances customer feature vectors from the time and space dimensions through dynamic feature extraction; on this basis, time decay and space decay functions are used to weight features, and the importance of spatiotemporal dimensions is adjusted using a multi-head attention mechanism to ensure the timeliness and spatial adaptability of features; further, through event-driven and gradual update mechanisms, the update frequency is controlled and versioned feature vectors are generated; finally, the attenuation parameters are adjusted through optimization strategies to improve the effect of feature updates and the accuracy of business decisions; it effectively overcomes the problem of insufficient consideration of spatiotemporal correlation in traditional methods, and enhances the real-time and accuracy of insurance decisions.

[0014] Preferably, the step 1 comprises the following steps:

[0015] Data acquisition and preprocessing: Collect customer core data, external spatiotemporal data, and real-time data, and clean and preprocess the collected data;

[0016] Standardization and gridding of spatiotemporal coordinates: Convert each data point into H3 code:

[0017] Adding spatiotemporal labels: Obtain the timestamp of the event and convert the geographic location of the event into a unique identifier through H3 grid encoding to obtain the spatiotemporal label of the event;

[0018] Set timeliness labels: Add timeliness labels to each piece of data and weight them according to the time decay function:

[0019] ;

[0020] Where: Indicates the time-dependent decay value; Indicates the current time; Indicates the timestamp of the data record; represents the attenuation parameter;

[0021] The types of timeliness tags are set to three categories according to the timeliness decay value, including static data, quasi-real-time data, and real-time data;

[0022] Generate a spatiotemporal association matrix: Combine the spatiotemporal label, timeliness label, and customer history feature vector of each event to construct a spatiotemporal association matrix between customers and events.

[0023] Preferably, the step 2 comprises the following steps:

[0024] Time dimension feature extraction: Fourier transform is used to process the time series data in the spatiotemporal correlation matrix. Based on the Fourier transform results, frequency components above the preset threshold are extracted to obtain periodic features, including daily, weekly, and monthly periodic features.

[0025] Event density change rate: Based on the event occurrence time of the spatiotemporal correlation matrix, the density change of events in a short time window is calculated to quantify the fluctuation trend of the event frequency:

[0026] ;

[0027] Where: Indicates time Number of events in the time period; Indicates time Number of events within a segment; Indicates the length of the time window; represents the rate of change of the density of the event;

[0028] Spatial entropy feature extraction: Based on the event space labels in the spatiotemporal correlation matrix, the spatial distribution is divided into multiple spatial units and the spatial entropy is calculated. :

[0029] ;

[0030] Where: represents the probability of an event occurring in a spatial unit, ,in represents the number of events in the i-th spatial unit; represents the total number of events in all spatial units; n represents the number of spatial units;

[0031] Spatial hotspot migration feature extraction: Use hotspot detection methods to identify hotspot areas in space and calculate the migration trajectory of hotspot areas over time:

[0032] ;

[0033] Where: Indicates time Coordinates of the center of the hotspot area at that moment; Indicates time Coordinates of the center of the hotspot area at that moment; represents the spatial migration distance of the hotspot center in time period t;

[0034] By calculating the migration distance between different time points, the migration pattern of spatial hot spots is identified;

[0035] A spatiotemporal enhanced feature tensor is obtained based on time dimension feature extraction, event density change rate extraction, spatial entropy feature extraction and spatial hotspot migration feature extraction.

[0036] Preferably, step 2 further includes spatiotemporal interaction feature extraction, and the features extracted from the spatiotemporal interaction feature are used as part of the spatiotemporal enhanced feature tensor, wherein the spatiotemporal interaction feature extraction includes spatiotemporal co-occurrence pattern extraction and cross-regional correlation strength extraction;

[0037] Spatiotemporal co-occurrence pattern extraction: Calculate the co-occurrence of events under different spatiotemporal conditions and identify the spatiotemporal interaction relationship of events. That is, for events at different times and spatial locations, calculate the spatiotemporal co-occurrence:

[0038] ;

[0039] Where: Representing an event and events In time and space Whether co-occurrence occurs, if so, it is 1, otherwise 0; N represents the total number of time windows calculated; Representing an event and events correlations in space and time;

[0040] Cross-region correlation strength extraction: for events in different regions A and B and events , calculate the cross-regional correlation strength:

[0041] ;

[0042] Where: and represents the number of events in region A and region B; Representing an event and whether they occur simultaneously; represents the spatiotemporal correlation strength across regions A and B.

[0043] Preferably, step 3 comprises the following steps:

[0044] Time decay function:

[0045] ;

[0046] Where: Indicates the time when the event occurs; t indicates the current time; represents the time decay factor; represents the base of natural logarithms;

[0047] Spatial attenuation function:

[0048] ;

[0049] Where: d represents the spatial distance between events; represents the spatial attenuation factor;

[0050] Multi-head attention mechanism: setting spatiotemporal enhanced tensor It is divided into h subspaces, that is, the head processing of each multi-head attention mechanism The calculation process is as follows:

[0051] Input query, key and value, calculate attention score based on the input query, key and key dimension, and then obtain weighted representation based on the calculated attention score weight value matrix; the multi-head output of the multi-head attention mechanism is obtained by concatenating the results of each head to obtain the overall feature representation;

[0052] Spatiotemporal weight adjustment: Based on the overall feature representation output by the multi-head attention mechanism and the values ​​of the time decay function and the spatial decay function, a feature vector with spatiotemporal weights is obtained:

[0053] ;

[0054] Where: Represents the overall feature representation of the multi-head attention mechanism output; Represents a feature vector with spatiotemporal weights.

[0055] Preferably, the event-driven update mechanism is as follows:

[0056] Set the time trigger update condition to , when a significant event occurs, the feature vector is updated:

[0057] ;

[0058] Where: represents the updated feature vector with spatiotemporal weights; Represents the feature vector with spatiotemporal weights before updating; represents the update intensity factor; Indicates the event trigger flag function, which is 1 if a major event occurs, otherwise 0; Indicates the characteristic change of spatiotemporal weight brought about by the event;

[0059] The judgment of the major event is based on business rules, and the characteristic change amount is calculated based on business rules.

[0060] Preferably, the gradual update mechanism is as follows:

[0061] ;

[0062] Where: Represents the feature vector with spatiotemporal weights after gradual update; represents the gradient factor, which is calculated based on the product of the time decay function and the space decay function; Represents the changes in features with spatiotemporal weights caused by small changes, calculated based on the cumulative effects of small events.

[0063] Preferably, the specific steps of controlling the update frequency based on the spatiotemporal attenuation threshold are as follows:

[0064] For each feature , calculate the time-dependent decay value and compare it with the spatiotemporal threshold:

[0065] ;

[0066] Where: Represents the feature representation of the multi-head attention mechanism output; represents the time decay function; represents the spatial attenuation function; Represents features with spatiotemporal weights The attenuation value;

[0067] when Less than threshold The update is triggered when the update is triggered, and the logic of triggering the update is as follows:

[0068] If there is a major event, the event-driven update mechanism is executed; if there is no major event, the gradual update mechanism is executed.

[0069] Preferably, step 4 also includes feature versioning and historical tracing; a version control mechanism is used to record changes in feature vectors with spatiotemporal weights, each feature update generates a new feature version number, and records the updated audit log, and through version control, historical backtracking and grayscale release strategies are implemented.

[0070] Preferably, step 5 uses Bayesian optimization to optimize the time decay parameters and update strategy, wherein Bayesian optimization aims to maximize the business effect of the model and the freshness of the features, and comprehensively optimizes the time decay factor, space decay factor and feature update frequency.

[0071] The beneficial effects of the present invention include:

[0072] The present invention establishes a spatiotemporal correlation matrix by collecting customer core data, external spatiotemporal data and real-time data streams, and enhances customer feature vectors from the time and space dimensions through dynamic feature extraction; on this basis, time decay and space decay functions are used to weight features, and the importance of spatiotemporal dimensions is adjusted using a multi-head attention mechanism to ensure the timeliness and spatial adaptability of features; further, through event-driven and gradual update mechanisms, the update frequency is controlled and versioned feature vectors are generated; finally, the attenuation parameters are adjusted through optimization strategies to improve the effect of feature updates and the accuracy of business decisions; it effectively overcomes the problem of insufficient consideration of spatiotemporal correlation in traditional methods, and enhances the real-time and accuracy of insurance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0074] Figure 1 This is a flowchart of the overall steps provided by an embodiment of the present invention.

[0075] Figure 2 This is a block diagram of the specific steps of step 2 provided in an embodiment of the present invention.

[0076] Figure 3 This is a simplified schematic diagram of step 3 provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the technical problems, technical solutions and beneficial effects to be solved by this application more clearly understood, this application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0078] See also Figure 1 As shown in FIG, the multi-factor integrated dynamic risk pricing optimization method for insurance products includes the following steps:

[0079] The method for updating the insurance customer feature vector based on dynamic adjustment of spatiotemporal correlation includes the following steps:

[0080] Step 1: Collect core customer data, external spatiotemporal data, and real-time data streams, standardize the spatiotemporal coordinates of the collected data, and unify them into a standard grid system. Attach spatiotemporal tags to each business event to ensure that each event clearly corresponds to its time and spatial location. Also, set timeliness tags for the data to obtain a spatiotemporal correlation matrix between customers and events.

[0081] The step 1 comprises the following steps:

[0082] Data acquisition and preprocessing: Collect customer core data, external spatiotemporal data, and real-time data, and clean and preprocess the collected data;

[0083] Core data includes customer policy information, claims records, and customer activity trajectories (such as location changes, device login logs, etc.);

[0084] External spatiotemporal data include weather data, traffic flow information, social event data (such as natural disasters, holidays, etc.), and geographic information system data;

[0085] Real-time data streams include IoT device data, mobile behavioral event data, location tracking data, etc.

[0086] Data cleaning and preprocessing include abnormal data processing, data standardization, etc., which are conventional data processing methods and will not be described in detail in the embodiment;

[0087] Standardization and gridding of spatiotemporal coordinates: The spatiotemporal coordinates of the data come from different sources, so all spatiotemporal coordinates need to be unified into the same standard grid system. Considering the accuracy and real-time requirements of geographic location, the H3 grid system is adopted. The H3 grid system is divided into multiple hexagonal grid cells of different sizes to facilitate the multi-level and high-precision representation of spatial data. We convert each data point (such as a geographic location point) into H3 code:

[0088] ;

[0089] in: Represents the original geographic coordinates; Indicates the level of the H3 grid; The unique H3 network identifier corresponding to each geographic location;

[0090] Attaching spatiotemporal tags: Obtain the timestamp of the event (for example, assume that an insurance claim occurs at time T, and this time T represents the timestamp of the event). Then, use H3 grid coding to convert the geographic location of the event into a unique identifier to obtain the spatiotemporal tag of the event. The spatiotemporal identifier of the event includes the timestamp of the event and the H3 network identifier.

[0091] Set timeliness labels: Add timeliness labels to each piece of data and weight them according to the time decay function:

[0092] ;

[0093] Where: Indicates the time-dependent decay value; Indicates the current time; Indicates the timestamp of the data record; represents the attenuation parameter;

[0094] The types of timeliness tags are set to three categories according to the timeliness decay value, including static data, quasi-real-time data, and real-time data;

[0095] It should be noted that here we filter out useful data by timeliness, such as static data, which can be defined as Smaller, the data no longer has significant timeliness, that is, we can set a threshold interval, within which all data are static data, which can be ; It can also be 0.1-0.2; within this range, we classify it as static data; similarly, the calibration method of punctual real-time data and real-time data is also carried out in accordance with the calibration method of static data;

[0096] Generate a spatiotemporal association matrix: Through the above steps, we obtain spatiotemporal labels, timeliness labels, and customer history feature vectors, which are combined to form a spatiotemporal association matrix between customers and events. For example, the spatiotemporal association matrix is ​​represented as follows:

[0097] ;

[0098] in: represents the spatiotemporal association between the i-th customer and the j-th event; Represents the historical feature vector of the i-th customer (such as age, gender, purchase history, etc.); Represents the business characteristics of the j-th event (such as insurance type, amount, etc.); represents the time label of event j; represents the spatial label (i.e., network identifier) ​​of event j; Represents the timeliness label of event j (the validity is calculated by the timeliness decay function);

[0099] In this embodiment, geographic data is unified into a high-precision grid system through H3 gridding processing. By combining the historical characteristics of customers with the spatiotemporal characteristics of events, a spatiotemporal correlation matrix between customers and events is constructed. This not only effectively integrates data from different sources, but also ensures the freshness and timeliness of the data, laying the foundation for subsequent spatiotemporal feature extraction and dynamic feature modeling.

[0100] Step 2: Dynamic feature extraction is performed based on the spatiotemporal correlation matrix and the customer's historical feature vectors. Periodic feature analysis is performed from the time dimension. The event density change rate is calculated to quantify the trend within a short time window. The entropy of the mobility pattern and the migration of spatial hotspots are extracted from the spatial dimension to obtain a spatiotemporal enhanced feature tensor, which includes spatiotemporal features and corresponding timeliness labels.

[0101] See also Figure 2 As shown, step 2 includes the following steps:

[0102] Time dimension feature extraction: Fourier transform is used to process the time series data in the spatiotemporal correlation matrix. Based on the Fourier transform results, frequency components above the preset threshold are extracted to obtain periodic features, including daily, weekly, and monthly periodic features.

[0103] The details are as follows: For time series data Apply Fourier transform (event time) to get frequency domain representation :

[0104] ;

[0105] Where: represents the frequency component in the frequency domain, which is the intensity of the time signal at different frequencies; f represents the frequency, which represents the periodicity of the event time; T represents the total length of the event time series; represents the complex exponential function, e represents the base of the natural logarithm; t represents the time variable, which is a certain moment in the time series;

[0106] Based on the results of Fourier transform, frequency components above a preset threshold are extracted to obtain periodic features, including daily periodicity, weekly periodicity, and monthly periodicity features;

[0107] Event density change rate: Based on the event occurrence time of the spatiotemporal correlation matrix, the density change of events in a short time window is calculated to quantify the fluctuation trend of the event frequency:

[0108] ;

[0109] Where: Indicates time Number of events in the time period; Indicates time Number of events within a segment; Indicates the length of the time window; represents the rate of change of the density of the event;

[0110] Spatial entropy feature extraction: event spatial labels based on spatiotemporal correlation matrix , divide the spatial distribution into multiple spatial units and calculate the spatial entropy :

[0111] ;

[0112] Where: represents the probability of an event occurring in a spatial unit, ,in represents the number of events in the i-th spatial unit; represents the total number of events in all spatial units; n represents the number of spatial units;

[0113] Spatial hotspot migration feature extraction: Use hotspot detection methods to identify hotspot areas in space and calculate the migration trajectory of hotspot areas over time:

[0114] ;

[0115] Where: Indicates time Coordinates of the center of the hotspot area at that moment; Indicates time Coordinates of the center of the hotspot area at that moment; represents the spatial migration distance of the hotspot center in time period t;

[0116] In this embodiment, the calculation of spatial entropy can reveal the regularity of spatial distribution, while the hotspot migration pattern can capture the dynamic interaction between regions, thereby enhancing the timeliness of spatial dimension features.

[0117] Spatiotemporal co-occurrence pattern extraction: Calculate the co-occurrence of events under different spatiotemporal conditions and identify the spatiotemporal interaction relationship of events. That is, for events at different times and spatial locations, calculate the spatiotemporal co-occurrence:

[0118] ;

[0119] Where: Representing an event and events In time and space Whether co-occurrence occurs, if so, it is 1, otherwise 0; N represents the total number of time windows calculated; Representing an event and events correlations in space and time;

[0120] Cross-region correlation strength extraction: for events in different regions A and B and events , calculate the cross-regional correlation strength:

[0121] ;

[0122] Where: and represents the number of events in region A and region B; Representing an event and whether they occur simultaneously; represents the strength of spatiotemporal correlation across regions A and B;

[0123] A spatiotemporal enhanced feature tensor T is constructed based on periodic characteristics, event density changes, spatial entropy, hotspot migration patterns, spatiotemporal co-occurrence, and cross-regional correlation strength; for example:

[0124] The spatiotemporal enhanced feature tensor T has the dimension ,in Indicates the number of customers; Indicates the number of events; Indicates the number of spatiotemporal features associated with each event and the customer;

[0125] The spatiotemporal features of each customer and event will form a three-dimensional tensor, where each dimension represents the customer, event, and spatiotemporal features, for example:

[0126] ;

[0127] In the formula: represents the spatio-temporal feature vector of the i-th customer and the j-th event; represents the time periodicity feature;

[0128] In this embodiment, the spatio-temporal correlation between events is revealed by extracting spatio-temporal co-occurrence features, which helps to identify potential business correlations; secondly, by extracting cross-regional correlation strength features, potential correlations between different geographical regions are identified, which is very important to understand how events in different regions affect each other, migrate, and how business activities expand across regions. Cross-regional correlation strength not only reveals the dependency between different regions, but also helps to build a business prediction model with better spatio-temporal perception ability.

[0129] Step 3: Perform attenuation processing on time and space based on the spatio-temporal enhanced feature tensor, use time attenuation and space attenuation functions to weight the timeliness and spatial distance respectively, then use the multi-head attention mechanism to calculate the importance weight of the spatio-temporal dimension, and dynamically adjust the feature vector to obtain the feature vector with spatio-temporal weight;

[0130] As shown in Figure 3 , the step 3 includes the following steps:

[0131] Time attenuation function: assuming that the time of a certain event is , the timeliness of the event decreases exponentially with time, so the time attenuation function is represented as:

[0132] ;

[0133] In the formula: represents the time of the event; t represents the current time; represents the time attenuation factor; represents the base of natural logarithm;

[0134] Space attenuation function: assuming that the spatial distance between two events is d, the influence of the event decreases with the increase of distance, and the space attenuation function is represented as:

[0135] ;

[0136] In the formula: d represents the spatial distance between events; represents the space attenuation factor;

[0137] Multi-head attention mechanism: assuming that the spatio-temporal enhanced tensor is divided into h subspaces, that is, each head of the multi-head attention mechanism processes part of it, and the calculation process is as follows:

[0138] Input query, key and value, calculate attention score based on the input query, key and key dimension, and then obtain weighted representation based on the calculated attention score weight value matrix; the multi-head output of the multi-head attention mechanism is obtained by concatenating the results of each head to obtain the overall feature representation;

[0139] Exemplary:

[0140] Query matrix: ,in is the query weight matrix for each head;

[0141] Bond Matrix: ,in is the key weight matrix for each head;

[0142] Value matrix: ,in is the value weight matrix for each head;

[0143] For each head h, calculate the attention score :

[0144] ;

[0145] Where: Represents the dimension of the key; represents the transpose of the bond matrix; The function converts the relevance score of each event to between 0 and 1;

[0146] Weighted sum: weighted matrix according to attention score , and obtain the weighted representation :

[0147] ;

[0148] Multi-head output merging: The final multi-head output channel is connected to the results of each head to obtain the overall feature representation:

[0149] ;

[0150] Where: represents the combined output weight matrix;

[0151] Spatiotemporal weight adjustment: Based on the overall feature representation output by the multi-head attention mechanism and the values ​​of the time decay function and the spatial decay function, a feature vector with spatiotemporal weights is obtained:

[0152] ;

[0153] Where: Represents the overall feature representation of the multi-head attention mechanism output; Represents a feature vector with spatiotemporal weights.

[0154] In this embodiment, the introduction of time decay and space decay functions enhances the model's adaptability to temporal and spatial changes, so that the influence of timeliness and spatial distance can be dynamically reflected in the weighting of feature vectors. In addition, a multi-head attention mechanism is used to process spatiotemporal features, which can dynamically adjust the weights of spatiotemporal features from multiple angles, thereby improving the model's ability to express complex spatiotemporal interactions. Finally, by combining timeliness and spatial decay with multi-head attention output, the model ensures a more refined modeling of spatiotemporal features, making dynamic modeling more accurate.

[0155] Step 4: Based on the feature vectors with spatiotemporal weights, an event-driven and gradual update mechanism is constructed. A spatiotemporal attenuation threshold is set to control the update frequency. A combined update strategy is used to balance the changes in feature vectors and time information. A versioned feature vector is generated and an update log is recorded.

[0156] The event-driven update mechanism is as follows:

[0157] Set the time trigger update condition to , when a significant event occurs, the feature vector is updated:

[0158] ;

[0159] Where: represents the updated feature vector with spatiotemporal weights; Represents the feature vector with spatiotemporal weights before updating; represents the update intensity factor; Indicates the event trigger flag function, which is 1 if a major event occurs, otherwise 0; Indicates the characteristic change of spatiotemporal weight brought about by the event;

[0160] The determination of major events is based on business rules (for example, in the insurance industry, possible major events include: major claims, accidents, large transactions, etc.);

[0161] The feature change is calculated based on business rules, for example:

[0162] For example, if a large claim occurs, the feature value associated with the claim amount may change significantly; the amount of change can be calculated as follows:

[0163] ;

[0164] It should be noted that f is a function that calculates the change according to business logic. For each feature, the change can be an increment, a proportional change, or the new value after the feature is updated.

[0165] The gradual update mechanism is as follows:

[0166] ;

[0167] Where: Represents the feature vector with spatiotemporal weights after gradual update; represents the gradient factor, which is calculated based on the product of the time decay function and the space decay function; Represents the changes in features with spatiotemporal weights caused by small changes, calculated based on the cumulative effects of small events.

[0168] The specific steps of controlling the update frequency based on the spatiotemporal attenuation threshold are as follows:

[0169] For each feature , calculate the time-dependent decay value and compare it with the spatiotemporal threshold:

[0170] ;

[0171] Where: Represents the feature representation of the multi-head attention mechanism output; represents the time decay function; represents the spatial attenuation function; Represents features with spatiotemporal weights The attenuation value;

[0172] when Less than threshold The update is triggered when the update is triggered, and the logic of triggering the update is as follows:

[0173] If there is a major event, the event-driven update mechanism is executed; if there is no major event, the gradual update mechanism is executed.

[0174] Step 4 also includes feature versioning and historical tracing. A version control mechanism is used to record changes in feature vectors with spatiotemporal weights. Each feature update generates a new feature version number and records an audit log of the update. Through version control, historical tracing and grayscale release strategies are implemented.

[0175] For example, each feature update generates a new feature version number. , and record updated audit logs, including update time, update events, attenuation factors, and other information. Through version control, it can support historical backtracking and grayscale release strategies;

[0176] For each feature version , the historical state is represented by the following formula:

[0177] ;

[0178] In the formula: represents the nth version of the feature; represents the change to the feature in this update; represents the timestamp of the update;

[0179] The audit log is generated at each feature update, recording information such as the timeliness of the update, the spatial impact of the update strategy, etc. The audit log can help analyze the impact of feature changes on the model later, especially in the case of feature drift, enabling timely backtracking and taking corrective measures;

[0180] The audit log record is represented as follows:

[0181] ;

[0182] Where: represents the nth audit log; is the update intensity factor; represents the time decay factor; represents the spatial decay factor; represents the trigger reason for this update;

[0183] In this embodiment, through the combination of event-driven and gradual update mechanisms, together with the control of time and space decay thresholds, the flexibility and real-time performance of dynamic feature updates are achieved; version control and audit log recording ensure that the system can trace back to historical features, support gray release and rollback operations; in this way, the timeliness of the feature vector is guaranteed, and excessive frequent updates are avoided, improving the stability and performance of the system

[0184] Step 5: Monitor the impact of feature updates on core business indicators, analyze the relationship between feature freshness and business indicators, and dynamically adjust the time decay parameter and update strategy using the strategy optimizer to obtain optimized update strategy parameters;

[0185] The step 5 uses Bayesian optimization to optimize the time decay parameter and the update strategy, where the Bayesian optimization aims to maximize the business effect of the model and the freshness of the feature, and comprehensively optimizes the time decay factor, the spatial decay factor, and the feature update frequency;

[0186] Exemplary:

[0187] Based on the time decay factor, the spatial decay factor, and the feature update frequency, a target function is constructed;

[0188] Assuming the goal is to maximize the model's business performance (for example, underwriting accuracy, claims approval efficiency, etc.) and the freshness of features, the objective function is as follows:

[0189] ;

[0190] Where: and denote the time attenuation factor and the space attenuation factor respectively; Indicates the feature update frequency; Indicates the improvement in business performance after feature updates. A specific indicator may be an increase in underwriting accuracy. Indicates that the freshness of the feature has increased, reflecting the improvement in model effectiveness after the feature update; Indicates the penalty for excessive or insufficient updates of special certificates; represents the penalty factor;

[0191] Exemplary: Consider the following factors:

[0192] Data novelty: This measures the similarity between newly collected data and existing data. New feature data can cover more unseen customer behaviors or events.

[0193] Model response: Observe the model's prediction performance on new data after using the new features, such as the improvement in the model's prediction accuracy, precision, recall, and other indicators;

[0194] Time window: Regularly evaluate the impact of updated features on model performance within a certain time window to ensure that features are updated in a timely manner and reflect the latest trends;

[0195] The calculation formula is:

[0196] ;

[0197] Where: represents the feature freshness score calculated based on the above factors; and Represent the timestamps of new feature data and old feature data respectively;

[0198] described It is measured based on the following business indicators:

[0199] Underwriting accuracy: Does the accuracy of underwriting decisions improve after the introduction of new features?

[0200] Claims approval efficiency: Does the new feature speed up the claims approval process and reduce approval time?

[0201] Customer satisfaction: Evaluate the impact of new features on customer experience through customer feedback or satisfaction scores;

[0202] ;

[0203] Where: and Represent the new accuracy score and old accuracy frequency score calculated based on business indicators respectively;

[0204] described The specific calculation formula is as follows:

[0205] ;

[0206] Where: 、 and Both represent weight factors; Indicates the update frequency; Indicates model stability; represents the update cost;

[0207] Use Gaussian processes to build a surrogate model to approximate the target function , learn the underlying laws of the objective function through historical data, and give the prediction results and uncertainties for each set of parameters. The proxy model is as follows:

[0208]

[0209] in: represents the mean function; represents the covariance function; For current selection Neighboring points of the parameter point;

[0210] Using expected improvement as the acquisition function to guide the exploration process of Bayesian optimization, the goal is to maximize:

[0211] ;

[0212] Where: Indicates the current optimal target value; represents the predicted value of the surrogate model;

[0213] At each iteration, Bayesian optimization selects a new set of parameters , and by calculating the objective function to update the proxy model, the optimal parameter configuration can be intelligently selected in the high-dimensional parameter space.

[0214] In this embodiment, through Bayesian optimization, the spatiotemporal attenuation factor and the feature update frequency can be intelligently adjusted to maximize the timeliness, freshness and business effect of the features.

[0215] The above are only preferred embodiments of the present application and are not intended to limit the present application. Any modifications, equivalent replacements, and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.

Claims

1. A method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation, characterized in that: The following steps are involved: Step 1: Collect core customer data, external spatiotemporal data, and real-time data streams, standardize the spatiotemporal coordinates of the collected data, and unify them into a standard grid system. Attach spatiotemporal tags to each business event to ensure that each event clearly corresponds to its time and spatial location. Timeliness tags are also set for the core customer data, external spatiotemporal data, and real-time data streams to obtain a spatiotemporal correlation matrix between customers and events. Core data includes customer policy information, claims records, and customer activity trajectories; external spatiotemporal data includes weather data, traffic flow information, social event data, and geographic information system data; real-time data streams include IoT device data, mobile behavioral event data, and location tracking data; Step 2: Dynamic feature extraction is performed based on the spatiotemporal correlation matrix and the customer's historical feature vectors. Periodic feature analysis is performed from the time dimension. The event density change rate is calculated to quantify the trend within a short time window. The entropy of the mobility pattern and the migration of spatial hotspots are extracted from the spatial dimension to obtain a spatiotemporal enhanced feature tensor, which includes spatiotemporal features and corresponding timeliness labels. Extracting the entropy value of the mobility pattern from the spatial dimension is to divide the space into a plurality of spatial units based on the event space labels in the spatiotemporal correlation matrix and calculate the spatial entropy; Step 3: Based on the spatiotemporal enhanced feature tensor, time and space are attenuated. Time and space attenuation functions are used to weight timeliness and spatial distance respectively. Then, a multi-head attention mechanism is used to calculate the importance weights of the spatiotemporal dimensions. The feature vector is dynamically adjusted to obtain a feature vector with spatiotemporal weights. Step 4: Based on the feature vector with spatiotemporal weights, an event-driven and gradual update mechanism is constructed. A spatiotemporal attenuation threshold is set to control the update frequency. A combined update strategy is used to balance the changes in feature vectors and time information, generate versioned feature vectors, and record update logs. The gradual update mechanism is as follows: ； Where: Represents the feature vector with spatiotemporal weights after gradual update; represents the gradient factor, which is calculated based on the product of the time decay function and the space decay function; Represents the characteristic changes with spatiotemporal weights caused by small changes, calculated based on the cumulative effects of small events; Step 5: Monitor the impact of feature updates on core business indicators, analyze the relationship between feature freshness and business indicators, and use the strategy optimizer to dynamically adjust the time decay parameters and update strategy to obtain the optimized update strategy parameters.

2. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: The step 1 comprises the following steps: Data acquisition and preprocessing: Collect customer core data, external spatiotemporal data, and real-time data, and clean and preprocess the collected data; Standardization and gridding of spatiotemporal coordinates: Convert each data point into H3 code: Adding spatiotemporal labels: Obtain the timestamp of the event and convert the geographic location of the event into a unique identifier through H3 grid encoding to obtain the spatiotemporal label of the event; Set timeliness labels: Add timeliness labels to each piece of data and weight them according to the time decay function: ； Where: Indicates the time-dependent decay value; Indicates the current time; Indicates the timestamp of the data record; represents the attenuation parameter; The types of timeliness tags are set to three categories according to the timeliness decay value, including static data, quasi-real-time data, and real-time data; Generate a spatiotemporal association matrix: Combine the spatiotemporal label, timeliness label, and customer history feature vector of each event to construct a spatiotemporal association matrix between customers and events.

3. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: The step 2 comprises the following steps: Time dimension feature extraction: Fourier transform is used to process the time series data in the spatiotemporal correlation matrix. Based on the Fourier transform results, frequency components above the preset threshold are extracted to obtain periodic features, including daily, weekly, and monthly periodic features. Event density change rate: Based on the event occurrence time of the spatiotemporal correlation matrix, the density change of events in a short time window is calculated to quantify the fluctuation trend of the event frequency: ； Where: Indicates time Number of events in the time period; Indicates time Number of events within a segment; Indicates the length of the time window; represents the rate of change of the density of the event; Spatial entropy feature extraction: Based on the event space labels in the spatiotemporal correlation matrix, the spatial distribution is divided into multiple spatial units and the spatial entropy is calculated. : ； Where: represents the probability of an event occurring in a spatial unit, ,in represents the number of events in the i-th spatial unit; represents the total number of events in all spatial units; n represents the number of spatial units; Spatial hotspot migration feature extraction: Use hotspot detection methods to identify hotspot areas in space and calculate the migration trajectory of hotspot areas over time: ； Where: Indicates time Coordinates of the center of the hotspot area at that moment; Indicates time Coordinates of the center of the hotspot area at that moment; represents the spatial migration distance of the hotspot center in time period t; By calculating the migration distance between different time points, the migration pattern of spatial hot spots is identified; A spatiotemporal enhanced feature tensor is obtained based on time dimension feature extraction, event density change rate extraction, spatial entropy feature extraction and spatial hotspot migration feature extraction.

4. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: Step 2 also includes spatiotemporal interaction feature extraction, where the features extracted from the spatiotemporal interaction feature are used as part of the spatiotemporal enhanced feature tensor. The spatiotemporal interaction feature extraction includes spatiotemporal co-occurrence pattern extraction and cross-regional correlation strength extraction. Spatiotemporal co-occurrence pattern extraction: Calculate the co-occurrence of events under different spatiotemporal conditions and identify the spatiotemporal interaction relationship of events. That is, for events at different times and spatial locations, calculate the spatiotemporal co-occurrence: ； Where: Representing an event and events In time and space Whether co-occurrence occurs, if so, it is 1, otherwise 0; N represents the total number of time windows calculated; Representing an event and events correlations in space and time; Cross-region correlation strength extraction: for events in different regions A and B and events , calculate the cross-regional correlation strength: ； Where: and represents the number of events in region A and region B; Representing an event and whether they occur simultaneously; represents the spatiotemporal correlation strength across regions A and B.

5. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: The step 3 comprises the following steps: Time decay function: ； Where: Indicates the time when the event occurs; t indicates the current time; represents the time decay factor; represents the base of natural logarithms; Spatial attenuation function: ； Where: d represents the spatial distance between events; represents the spatial attenuation factor; Multi-head attention mechanism: setting spatiotemporal enhanced tensor It is divided into h subspaces, that is, the head processing of each multi-head attention mechanism The calculation process is as follows: Input query, key and value, calculate attention score based on the input query, key and key dimension, and then obtain weighted representation based on the calculated attention score weight value matrix; the multi-head output of the multi-head attention mechanism is obtained by concatenating the results of each head to obtain the overall feature representation; Spatiotemporal weight adjustment: Based on the overall feature representation output by the multi-head attention mechanism and the values ​​of the time decay function and the spatial decay function, a feature vector with spatiotemporal weights is obtained: ； Where: Represents the overall feature representation of the multi-head attention mechanism output; Represents a feature vector with spatiotemporal weights.

6. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: The event-driven update mechanism is as follows: Set the time trigger update condition to , when a significant event occurs, the feature vector is updated: ； Where: represents the updated feature vector with spatiotemporal weights; Represents the feature vector with spatiotemporal weights before updating; represents the update intensity factor; Indicates the event trigger flag function, which is 1 if a major event occurs, otherwise 0; Indicates the characteristic change of spatiotemporal weight brought about by the event; The judgment of the major event is based on business rules, and the characteristic change amount is calculated based on business rules.

7. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: The specific steps of controlling the update frequency based on the spatiotemporal attenuation threshold are as follows: For each feature , calculate the time-dependent decay value and compare it with the spatiotemporal threshold: ； Where: Represents the feature representation of the multi-head attention mechanism output; represents the time decay function; represents the spatial attenuation function; Represents features with spatiotemporal weights The attenuation value; when Less than threshold The update is triggered when the update is triggered, and the logic of triggering the update is as follows: If there is a major event, the event-driven update mechanism is executed; if there is no major event, the gradual update mechanism is executed.

8. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: Step 4 also includes feature versioning and historical tracing; a version control mechanism is used to record changes in feature vectors with spatiotemporal weights. Each feature update generates a new feature version number and records the updated audit log. Through version control, historical tracing and grayscale release strategies are implemented.

9. The method for updating insurance customer feature vectors based on dynamic adjustment of spatiotemporal correlation according to claim 1, characterized in that: The step 5 uses Bayesian optimization to optimize the time decay parameters and update strategy, wherein Bayesian optimization aims to maximize the business effect of the model and the freshness of the features, and comprehensively optimizes the time decay factor, space decay factor and feature update frequency.

Images