Customer portrait matching and positioning method and system

Abstract

The invention discloses a customer portrait matching and positioning method and system, and particularly relates to the technical field of portrait matching and positioning, and the method comprises the steps: carrying out the cleaning and time correction of multi-source behavior data under a unified time base, and constructing a behavior time sequence and a label time sequence; on this basis, identifying a tag high-intensity activation interval, and forming a multi-time scale portrait structure composed of a long-term structural portrait, a life cycle stage portrait and an event portrait; for business task analysis time sensitivity and risk tolerance, calculating a comprehensive analysis index to judge whether time-dependent dynamic attenuation is introduced, respectively obtaining scores of all portrait layers, fusing the scores according to weights, generating a comprehensive matching score, and outputting a target customer set; and combining recommendation, marketing or risk control feedback to construct a feedback event sequence, and adjusting the multi-time scale portrait and the fusion weight. According to the method, long-term features, stage behaviors and short-term events can be distinguished, and the accuracy and interpretability of customer portrait matching and positioning are improved.

Description

Technical Field

[0001] This invention relates to the field of profile matching and positioning technology, and more specifically, to a customer profile matching and positioning method and system. Background Technology

[0002] With the development of online businesses such as e-commerce transactions, content browsing, search services, and third-party advertising, enterprises can continuously collect multi-source behavioral data from multiple business systems and external cooperation channels. After time correction and cleaning, this data can be organized into customer behavior time series, and on this basis, a time series with timestamps can be built, providing a more refined input basis for customer profile modeling.

[0003] Existing customer profiling matching solutions mostly rely on statistical results of tags on a single time scale. This makes it difficult to distinguish between long-term structural characteristics, behaviors with phased concentrated outbreaks, and short-term single events with significant business significance. Furthermore, there is a lack of a systematic mechanism to combine environmental events and basic customer attributes to semantically classify behavioral fragments and abstract them into lifecycle stages. As a result, it is easy for short-term promotions and other phased behaviors to mask long-term preferences, or for phased states to be difficult to express clearly in the profile.

[0004] On the other hand, existing technologies typically employ fixed time windows or simple time decay rules based on experience when dealing with time factors. They have not yet introduced quantitative indicators such as the tag time instability index and the proportion of phased behavior index, making it difficult to determine whether to enable time-related dynamic decay based on the time sensitivity and risk tolerance of different tasks. At the same time, for feedback results generated by recommendations, marketing, or risk control, there is a lack of ability to incorporate the feedback event sequence into a unified time base for alignment and to iteratively adjust the multi-timescale profiles and the fusion weights of each profile layer accordingly. This results in shortcomings in the accuracy, time stability, and interpretability of profile matching results. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a customer profile matching and positioning method and system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A customer profile matching and positioning method includes the following steps: Under a unified time base, multi-source behavioral data is cleaned and time-corrected to construct behavioral time series and label time series. Based on the identification of high-intensity activation intervals of tags through tag time series, a multi-time scale profile structure is formed, which includes long-term structural profiles, life cycle stage profiles, and event profiles. The system analyzes the time sensitivity and risk tolerance of business tasks, and calculates comprehensive analysis indicators based on the tag time instability index and the proportion index of phased behavior. Based on the comprehensive analysis indicators, it determines whether to introduce time-related dynamic decay. The system calculates the long-term structural profile score, life cycle stage profile score, and event profile score respectively, and generates a comprehensive matching score by weighted summation of the scores of each profile layer. Based on the comprehensive matching score, the system outputs the target customer set. The system combines feedback from recommendations, marketing, or risk control to construct a sequence of feedback events, and then adjusts the multi-timescale profiles and fusion weights based on this sequence.

[0007] In a preferred embodiment, the process of identifying the high-intensity activation interval of a tag based on the tag time series includes: scanning along the time axis of the tag time series using a sliding time window of a preset length; comparing the tag activation intensity within the window with the historical baseline intensity of the corresponding customer; when the tag activation intensity in a certain continuous time interval shows a continuous and significant increase relative to the historical baseline, marking the time interval as the high-intensity activation interval of the tag, and recording its start and end times and peak intensity.

[0008] In a preferred embodiment, forming a multi-timescale profile structure includes: aligning the high-intensity activation intervals of customers across various tag dimensions along the timeline to identify candidate segments of phased events; semantically classifying the candidate segments of phased events by combining environmental event sequences and the customer's basic attributes to construct a lifecycle phase profile; removing high-intensity segments located within the lifecycle event impact window from the tag timeline and constructing a long-term structural profile based on the tag activation records of the remaining time period; and extracting single events that do not form a complete lifecycle phase but have significant business implications to construct an event profile.

[0009] In a preferred embodiment, constructing a lifecycle stage profile includes: for the identified lifecycle event types, recording the type, start time, end time, stage intensity, and related tag set of the corresponding lifecycle events at the customer level; when the same type of lifecycle event appears multiple times on the customer's behavior timeline, connecting these lifecycle events in chronological order to form a stage trajectory.

[0010] In a preferred embodiment, a comprehensive analysis index is calculated based on the tag time instability index and the phased behavior proportion index, including: weighting and fusing the tag time instability index and the phased behavior proportion index to obtain a comprehensive analysis index value; comparing the comprehensive analysis index value with a preset dynamic decay activation threshold, and determining whether to introduce time-related dynamic decay when calculating the score of each profile layer based on the comparison result.

[0011] In a preferred embodiment, the tag time instability index is used to reflect the average level of the difference in activation intensity between the recent observation window and the long-term historical window for the tags of interest in this task; the phased behavior proportion index is used to measure the proportion of behavior falling within the life cycle phase impact window or short-term event impact window in the overall behavior within the current task and candidate customer scope.

[0012] In a preferred embodiment, the scores for each profile layer are calculated, including: calculating a long-term structural profile score, which is based on the similarity between the typical values ​​of the customer's tags over a long period and the target profile template; The lifecycle stage profile score is calculated based on the affinity between the customer's current lifecycle stage and the business task, and a timeliness weight is introduced when time-related dynamic decay is enabled. The event profile score is calculated based on the relevance of a single recent event to a customer's business tasks, and a time decay weight is introduced when time-related dynamic decay is enabled.

[0013] In a preferred embodiment, the feedback event sequence includes customer feedback data from recommendation, marketing, or risk control operations, which is anonymized and formatted and aligned with the behavioral time series.

[0014] In a preferred embodiment, the feedback event sequence adjusts the fusion weights, including: summarizing and analyzing the overall performance of several similar business tasks; and, based on the consistency between the overall matching score and the actual results, slowly adjusting the fusion weight ratio of the long-term structural profile, the life cycle stage profile, and the event profile through iterative learning.

[0015] In a preferred embodiment, a customer profile matching and positioning system includes the following modules: The data acquisition and processing module is used to clean and time-correct multi-source behavioral data under a unified time base, and to construct behavioral time series and label time series. The profile modeling module is used to identify high-intensity activation intervals of tags based on tag time series and form a multi-time scale profile structure, which includes long-term structural profiles, life cycle stage profiles and event profiles. The matching calculation module is used to analyze the time sensitivity and risk tolerance of business tasks, and calculate comprehensive analysis indicators based on the tag time instability index and the proportion index of phased behavior. Based on the comprehensive analysis indicators, it determines whether to introduce time-related dynamic decay. It calculates the long-term structural profile score, life cycle stage profile score and event profile score respectively, and generates a comprehensive matching score by weighted summation and fusion of the scores of each profile layer. Based on the comprehensive matching score, it outputs the target customer set. The feedback adjustment module is used to construct a feedback event sequence by combining recommendation, marketing or risk control feedback, and to adjust the multi-timescale profiles and fusion weights according to the feedback event sequence.

[0016] The technical effects and advantages of this invention are as follows: This invention performs time calibration and cleaning on multi-source behavioral data from e-commerce transaction systems, content browsing systems, search log systems, third-party advertising systems, and external cooperation channel interfaces under a unified time base. It constructs behavioral time series and timestamped tag time series, and forms multi-timescale profile layers such as long-term structural profiles, life cycle stage profiles, and event profiles based on the high-intensity activation intervals of tags. This allows for the differentiation of long-term stable characteristics, periodic concentrated bursts of behavior, and single events with significant business significance within the same profile framework. It avoids misjudging short-term activities or occasional events as long-term preferences, and significantly improves the accuracy and robustness of customer profile matching and positioning.

[0017] This invention introduces a tag time instability index and a phased behavior proportion index to quantitatively characterize the time stability of tags within the observation window and the proportion of phased behaviors. It also combines task parameters such as the time sensitivity of the business task, the observation window length, and risk tolerance to determine whether to enable a time-related dynamic decay strategy. When necessary, the contributions from expired behaviors or behaviors at the boundary of the influence window in the long-term structural profile score, lifecycle phase profile score, and event profile score are attenuated and corrected. When not necessary, the baseline score within the layer is maintained or only weak decay is implemented. This achieves differentiated control over time factors, ensuring that the matching results sensitively reflect recent changes in the time dimension while avoiding over-responding to short-term noise, thereby improving the system's time stability and adaptability.

[0018] This invention calculates long-term structural profile scores, lifecycle stage profile scores, and event profile scores separately, and then weights and fuses these three types of scores according to the weights set in the task profile template to obtain a comprehensive matching score. When making recommendation, marketing, or risk control decisions, the upper-level business system can directly use the comprehensive matching score for customer ranking and selection, or it can analyze the reasons for customer selection or exclusion based on the scores of each profile layer, significantly improving the interpretability of the matching results and facilitating strategy verification and decision explanation by operations and risk control personnel. Simultaneously, this invention constructs the results of recommendation, marketing, or risk control execution as a feedback event sequence, aligning it with behavioral time series and tag time series under a unified time base. Based on the consistency between the comprehensive matching score and actual feedback, it iteratively adjusts the multi-timescale profile structure and the fusion weight ratio of the three profile layers, enabling the system to automatically correct the influence of each profile layer as real business feedback is continuously introduced, forming a closed-loop optimization mechanism and reducing reliance on manual parameter tuning and empirical rules. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart illustrating a customer profile matching and positioning method according to the present invention. Figure 2 This is a schematic diagram of the structure of a customer profile matching and positioning system according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: A customer profile matching and positioning method of the present invention, such as... Figure 1 As shown, it includes the following steps: Step 1: Constructing the labeled time series; In one embodiment of the customer profile matching and positioning method of the present invention, the purpose of step one is to organize the customer behaviors scattered in different business systems into a behavior time series on a unified time axis, and on this basis, construct a label time series to provide structured input for subsequent multi-time scale profile modeling.

[0022] Specifically, within a pre-set data collection period, without infringing on customer privacy, behavioral and transaction data related to customers are continuously collected from e-commerce transaction systems, content browsing systems, search log systems, third-party advertising systems, and external cooperation channel interfaces. The aforementioned behavioral data may include customer search keywords on the platform and their trigger times, exposure records and dwell time of browsing pages or content, adding products to the shopping cart, order and payment behavior, ratings and evaluations of products or services, after-sales and refund applications, and other business operation records. At the same time, basic attribute information and terminal device environment information associated with customer accounts are also collected.

[0023] Since the aforementioned behavioral data often originates from different systems, covers different regions, and may be located in different time zones, this step first uniformly corrects the timestamps of all behavioral records. The system, according to a preset standard time base, converts the local times recorded in each business system into a time representation under a unified time base through time zone conversion and format standardization. Based on this, using the customer's primary identifier as the aggregation key, it merges multi-source behavioral records of the same customer from the e-commerce transaction system, content browsing system, search log system, and third-party advertising system, and sorts them from earliest to latest according to the corrected timestamps. Let the behavioral time series of a certain customer be denoted as... ; Where u represents the customer's primary identifier. This represents the timestamp of the k-th action record under a unified time base. This indicates the business content or operation type related to this action. This represents the number of actions taken by this customer within the data collection period. This step is part of constructing the sequence. During the process, obviously abnormal timestamps, missing time fields, and duplicate records are cleaned and corrected to ensure that the position of each behavior on a unified time base is clear and traceable, thereby forming a standardized behavior dataset with customer ID plus behavior time series as the basic unit. The customer ID can be set as a virtual number, etc., without infringing on customer privacy.

[0024] Information on environmental events is collected from the operations management system and external public data sources. Environmental events mainly include large-scale promotional activities within the platform, cross-brand joint activities, statutory holidays, and major social events. For each type of environmental event, the system records its effective start time, effective end time, and event type, and converts it to a unified time base to form an environmental event sequence parallel to the behavioral time series.

[0025] After obtaining standardized behavioral time series and environmental event sequences, this step maps the original behavioral records to activation records at the tag level, based on a pre-designed tagging system. The tagging system can be divided into different dimensions, such as interest preference tags, consumer category tags, risk characteristic tags, and environment-related tags, with each dimension containing several specific tags. For each record in the behavioral time series... The system searches for and matches corresponding tag codes based on feature fields such as product category, brand affiliation, price range, payment method, transaction channel, and content theme. It then marks the behavior as activated on one or more related tags, while retaining the original behavior timestamp. This generates a tag activation record with time information. For cases where a single action triggers multiple tags, this step does not compress or merge them, but rather preserves the correspondence between the multiple tag activations for subsequent analysis of tag co-occurrence patterns and tag combination characteristics.

[0026] After completing the label mapping, this step constructs a label time series under a unified time base, using the customer as the granularity and a single label as the observation object. For any customer u and label... The system accumulates all records where the tag is activated in chronological order along its behavior timeline, forming a sequence. ,in Indicates label The j-th activation time in the customer u dimension, This refers to the number of times the tag is activated within the collection period. Based on the above sequence, this step calculates and saves the statistical characteristics of the tag's first activation time, most recent activation time, cumulative activation count, adjacent activation time intervals, and activation intensity within a preset time slice for that customer dimension. Activation intensity can be obtained by comprehensively considering factors such as the number of times the tag is activated within the time slice, the amount of related behavior, and the duration of stay, and is used to characterize the tag's activity level at different times. After obtaining the tag's time series, the system uses a preset-length sliding time window to scan along the time axis, comparing the activation intensity within the window with the customer's historical baseline intensity. When the activation intensity within a certain continuous time interval shows a sustained and significant increase relative to the baseline, that time interval is marked as the tag's high-intensity activation interval, and its start and end times and peak intensity are recorded in the tag's time series.

[0027] Step Two: Stage Behavioral Profile Modeling; Step two involves further identifying behavioral segments with phased characteristics based on the tagged time series view obtained in step one. On this basis, a multi-timescale hierarchical profile structure is constructed, allowing long-term preferences, phased states, and short-term events to be expressed separately in the profile, rather than simply being superimposed. For ease of understanding, step two can be understood as first identifying which behaviors on the timeline are concentrated in a particular phase, and then modeling these phases separately from daily, relatively stable behaviors.

[0028] Specifically, this step first aligns the high-intensity activation intervals across various tag dimensions for each individual customer. For the same customer, if multiple semantically related tags simultaneously exhibit a significant increase in intensity within a consecutive time window, or if they are closely sequential in time, then the behavioral segment corresponding to this time window is considered a candidate segment of a phased event. To quantify the degree of this concentrated outbreak, under a unified time base, the activation intervals of a customer u and a tag set L within a time range can be analyzed. The intensity of each stage within the process can be calculated, for example, by introducing a simple intensity index. ; in, and These represent the start and end times of the segment, respectively. This indicates the base intensity corresponding to a specific tag activation within that time interval; for example, for browsing behavior, It can be the normalized value of the duration of the visit; For different tags or different behavior types, the weight coefficient can be set. For example, the weight of successful payment behavior can be set to 1.0, the weight of adding to cart behavior can be set to 0.7, and the weight of browsing product details page behavior can be set to 0.3. This refers to the duration of the segment. (By comparison...) By comparing the baseline intensity of the customer over a longer historical window, it is possible to more intuitively identify which segments are abnormally active in a phase, and thus mark them as candidates for phased events.

[0029] Combining environmental event sequences and customer basic attributes, candidate fragments are semantically categorized. Here, semantic categorization means not only looking at which category or interest the tag itself belongs to, but also whether certain environmental factors are superimposed when it occurs. Based on a pre-defined pattern library, this step categorizes fragments that meet the characteristics into different life cycle event types, such as the parenting stage, renovation stage, exam preparation stage, and treatment stage, and configures a reasonable impact window and decay period for each type of life cycle event.

[0030] After identifying lifecycle events, a lifecycle stage profile is constructed at the customer level. For the same customer, if the same type of lifecycle event occurs multiple times on their behavioral timeline, these events are connected chronologically to form a stage trajectory. Each stage trajectory typically includes information such as the lifecycle event type, the start and end times of the stage, the stage intensity calculated based on the aforementioned intensity indicators, the radius of influence inferred from business experience or models, and a set of tags significantly related to that stage. By maintaining such stage trajectories, this invention can clearly record the customer's life scenarios and stage states at different times at the profile level, so that subsequent matching and positioning no longer depends solely on what the person likes, but can simultaneously answer what stage the person is currently in. The lifecycle stage profile, as a relatively independent layer, is interconnected with the tag time series and environmental event series, providing a structure reflecting the medium- to long-term state for multi-timescale profiling.

[0031] High-intensity segments marked as lifecycle events within the impact window are removed from the tag timeline, retaining only tag activation records for the remaining time periods. For these remaining records, this step no longer emphasizes bursts within a short period, but focuses on average levels and relative rankings over a longer time span. Typical values ​​and activity levels for each tag over the long term are calculated using time-weighted averaging, quantile statistics, or other robust statistical methods.

[0032] On the other hand, for single events that, while not forming a complete lifecycle phase, have significant business implications in a short period, this step extracts them and constructs a separate event profile layer. For these events, this step records the event occurrence time, event type, event intensity, and the set of tags associated with the event, and configures appropriate decay strategies for the relevant tags, so that they can influence the matching calculation within a reasonable time window after the event occurs, and gradually fade out after that window.

[0033] Step 3: Calculate the task matching score; Step three, based on the completed multi-timescale profiling, generates a target customer set that can be directly used for execution, centered around specific business needs. To this end, when the upper-level business system initiates a matching and location request, the system does not simply perform a static screening of existing profiles, but first meticulously analyzes the business information carried by the request. The request typically includes the type of business objective, the time sensitivity requirements, the acceptable risk level, and a profile template describing the characteristics of the ideal target customer group. The type of business objective can take various forms, such as short-term promotional activities, phased benefit operations, annual value tiering assessments, or risk screening; time sensitivity reflects whether the current task prioritizes recent behavioral performance or relies more on long-term stable characteristics; risk tolerance constrains the acceptable fluctuation range of the matching results in terms of conversion rate, bad debt rate, or complaint rate; and the profile template, in the form of tags and feature ranges, provides an approximate outline of the type of customers desired. Based on this information, the system derives the basic weight ratio of the three profile layers—long-term structural profile, lifecycle stage profile, and event profile—in subsequent calculations for this task. For example, the weight of the event profile and the current lifecycle stage is appropriately increased during short-term events before holidays, while the weight of the long-term structural profile is increased in annual customer segmentation scenarios.

[0034] After completing the above task analysis, the system selects a candidate customer set from the profile storage; the scope of the candidate set can be determined based on the constraints set in advance by the business party. After the candidate set is determined, the matching degree between each candidate customer and the target profile template needs to be calculated at each of the three profile levels. For each profile level, the long-term structural profile score, lifecycle stage profile score, and event profile score are calculated.

[0035] In one optional implementation of the present invention, whether a time-related dynamic decay mechanism needs to be introduced during the matching score calculation process of each image layer depends on the already calculated tag time instability index. and the percentage of phased behaviors index Based on this, by merging the two indices in a weighted manner, a comprehensive analytical index J for decision-making is constructed, which can be defined as follows: The weighting coefficients are among them. and , + =1, for example =0.4, =0.6; The higher the value, the greater the degree of short-term hot spot heat and the higher the coverage of phased behaviors, indicating that the short-term behaviors and phased events in the current task have a more significant impact on the overall profile.

[0036] After obtaining the comprehensive analysis index J, this step introduces a dynamic decay activation threshold that is related to the business scenario. This threshold also takes a value of [0,1], and can be set through offline simulation, historical task backtracking analysis, and operational experience. It can also be adjusted appropriately based on feedback during long-term system operation. For task types with high time sensitivity, It can be set relatively low so that time-related dynamic decay can be activated promptly once a certain degree of temporal instability and periodic concentration of behavior are detected; for tasks that are mainly based on annual value assessment and long-term profile stratification and are not sensitive to short-term fluctuations, It can be set relatively high, so that static images are the primary focus in most cases, and attenuation is only activated when short-term disturbances are extremely significant. During actual operation, the system compares J calculated for the current task with the threshold. Compare; when J≥ When it is determined that the impact of short-term hotspots and phased behaviors on the label distribution in the current task has reached a level requiring close attention, and there is a high risk of mistakenly solidifying short-term behaviors into long-term preferences, the contribution of labels that are expired or at the edge of the influence window is attenuated according to the preset time decay curve in the in-layer score calculation; when J < When the label is relatively stable over time, the coverage of phased behaviors and short-term events within the observation window is limited. The current task can be safely regarded as being dominated by long-term structural features. In this case, the system can directly use the intra-layer baseline score without time decay, or implement weak decay only in the event profile layer, thereby avoiding the introduction of unnecessary time modeling overhead while ensuring matching accuracy.

[0037] Among them, the label time instability index This average level reflects the difference in activation intensity between the recent observation window and the long-term historical window for the tags of interest in this task. By comparing the overall intensity of recent and historical tag activation frequency, associated monetary value, or dwell time, when most key tags show a significant recent increase in activity while remaining at a low baseline for an extended period, this level is considered appropriate. The value of the index tends to increase; when recent behavior is basically consistent with or slightly decreases from the long-term distribution, the value of the index is smaller. Standardize to the [0,1] range. The closer the value is to 1, the more significant the short-term hotspots are to the label distribution, and the higher the risk that long-term structural characteristics will be masked by short-term behavior.

[0038] Phased behavior proportion index This index measures the proportion of behavior falling within the lifecycle stage influence window or short-term event influence window within the current task and candidate customer scope, relative to overall behavior. Based on identified lifecycle stage profiles and event profiles, the system categorizes behavior within the observation window into staged and non-staged behaviors, and calculates the proportion of staged behavior in the total behavior volume accordingly. This index can also be normalized to the [0,1] range; a higher value indicates that current business occurs more frequently within a specific stage or timeframe dominated by stage events, and that staged behavior has a stronger influence on the current task. A lower value indicates that most behavior occurs during a stable period, and the impact of lifecycle stages and short-term events is relatively limited.

[0039] Long-term structural profile score The goal of this assessment is to measure the consistency between a customer's steady-state characteristics and the target profile template. This score is calculated by comparing the typical values ​​of the customer's tags over a long period with the template requirements. Specifically, the long-term structural profile layer contains a set of tags. Each of the tags There is a long-term typical value in the customer u dimension, obtained through time-weighted average or quantile statistics. This value reflects the steady-state level after removing short-term fluctuations; the target profile template is for each label. Define an expected value or reasonable range .Score The calculation formula is: ; in, It is a tag The weighting coefficients, The system administrator can preset the tags based on their importance in the business. This is a similarity function used to quantify the closeness between a customer's value and a target value. As a preferred implementation, the similarity function is calculated based on the normalized absolute difference, and its specific mathematical form is as follows: ;in, Indicates customer u on the label Long-term typical values; This indicates that the target profile template is a tag. Defined expected value or benchmark value; For example, a very small positive number =0.00000001, used to avoid the case where the denominator is zero.

[0040] Life cycle stage profile score The data is used to assess the dynamic fit between a customer's current lifecycle stage and business tasks, and its calculation significantly reflects the timeliness of the stage. Let customer u be currently in a set of lifecycle stages. In each stage Record its type and start time. End time (i.e., the impact window), stage intensity and related tag collection The formula for calculating the stage strength H is: ; in, Indicates the time interval The base strength of a particular tag activation. These are weighting coefficients for different behavior types. Scores reflect the dynamic influence of each stage. The calculation introduces a timeliness weight, when J≥ When, the formula is: When J < When; its formula is: ; in, It represents the affinity between stage p and the current business task. It can be pre-configured in a mapping table; for example, for the childcare stage and infant formula promotion tasks, the affinity can be set to 0.9; for the renovation stage and the same task, the affinity can be set to 0.1. Is stage p at the current calculation time? The timeliness weight can be calculated, for example, using a linear decay function: ;in It is the preset decay buffer duration, such as 30 days.

[0041] Event profile score The purpose of this data acquisition is to measure the relevance of a single, recent salient event to a customer's business objectives, and to strictly control its short-term impact through a decay mechanism. Suppose customer u experienced a set of events recently. Each event Includes event type and occurrence time Event intensity and related tag collection When J≥ At that time, the score was... The calculation formula is: When J < At that time, the score was... The calculation formula is: ; in, This is the weighting coefficient for event e, which is usually related to the event intensity. Positive correlation; It is the relevance of event e to the current business task. Similar to stage affinity, it can be obtained through a predefined mapping table; It is a time decay weighting function, which can be set as follows: ; in, Indicates the current time and the time the event occurred. The interval between This is the attenuation coefficient, used to control the rate at which the weights decay, for example... This formula is typically used to represent the decay constant in an exponential decay process. With half-life Relationship, If half-life The decay constant is calculated for each unit of time (e.g., day, hour). This mechanism ensures that recent events contribute significantly to the score, while their impact diminishes rapidly over time, thus effectively preventing short-term behavior from being misjudged as long-term preferences.

[0042] After time decay correction within each profile layer, the matching scores from the three layers can be fused to obtain a comprehensive matching score tailored to the current business scenario. For ease of description, an optional implementation method defines the following fusion relationship for each candidate customer u: ; in, This indicates the overall matching score of customer u under the current task. The score is calculated based on a long-term structural profile. The score is obtained based on the life cycle stage profile. The score is obtained based on the event profile; , , The fusion weights for long-term profiles, phase profiles, and event profiles in this task should meet the following requirements. + + =1, for example , =0.4、 =0.4.

[0043] The candidate set is sorted from highest to lowest score. Combining the pre-set score thresholds by the business stakeholders, the tiered strategies corresponding to different score ranges, and resource constraints, the system sequentially selects customers from the front of the sorted results to form the final target customer set.

[0044] Step 4: Feedback-driven weight adjustment; Step four aims to realign the calculated results with actual business events, ensuring that the previously established multi-timescale profiles are no longer static snapshots but dynamic structures that can be gradually revised based on subsequent customer behavior. To this end, after Step three outputs the target customer set and the upper-level system completes recommendations, marketing, benefit distribution, or risk control, this step first collects and organizes the execution results of these actions under a unified time base, extracting information related to customer feedback to form an independent feedback event sequence. Feedback events can include whether a customer saw, clicked, or ignored a recommendation; whether they placed an order or made a payment within a specified time; whether they unsubscribed from or complained about marketing outreach; whether they used the distributed benefits; whether credit approval was granted; whether they repaid on time after using the credit; whether they defaulted; and whether manual review was triggered. After anonymization and format standardization, this data is reordered according to the customer's primary identifier and standard time base, ensuring a one-to-one correspondence between it and the aforementioned behavior time series and tag time series on the timeline.

[0045] After obtaining the feedback event sequence, this step does not directly recalculate the entire profile. Instead, it extracts several quantifiable performance metrics from the feedback to measure the reliability of the profile-driven matching decisions. For example, for a specific matching task, metrics such as click-through rate, conversion rate, cancellation rate, and complaint rate can be statistically analyzed at the customer level. These results can then be stratified and compared according to score ranges, tag combinations, or lifecycle stages. To facilitate automatic use in the algorithm, a simplified feedback score can be constructed. For example, the comprehensive feedback for a customer u in a certain task can be defined as: ; in, This can be understood as conversion-related metrics, such as whether the expected behavior was achieved or its probability value. This indicates the intensity of negative feedback, such as unsubscribing or complaints. , Positive weights are used to reflect the relative importance of positive and negative results in the overall evaluation, for example... , .

[0046] In the long-term structural profiling layer, this step focuses more on the stability and reliability of certain long-term tags across multiple tasks. This includes tags that have been used multiple times in different scenarios to screen customers and generate feedback scores. Labels or combinations of labels consistently exceeding the preset baseline can be considered to genuinely reflect the customer's long-term characteristics. In this case, this step will appropriately extend the remaining validity period of these labels and slow down the natural decay rate of their weights over time. Conversely, for some labels, although they were assigned high weights during the modeling phase, if the corresponding customer group feedback is weak or even negative in multiple rounds of tasks, a label aging strategy will be triggered. This strategy will gradually weaken their influence in the long-term structural profile by reducing their weights and shortening their validity period. If necessary, they may be downgraded from strong preference labels to weak preference labels, or even marked as temporarily excluded from matching calculations. By strengthening well-performing long-term characteristics and weakening repeatedly ineffective long-term characteristics, the long-term structural profile can gradually converge into a more accurate depiction of the customer's intrinsic attributes during continuous operation, rather than remaining at the static assumptions of the initial definition.

[0047] In the lifecycle stage profiling layer, this step focuses on addressing when a stage should end and whether its impact is overestimated or underestimated. For a customer currently in a particular lifecycle stage, the system aligns the corresponding stage trajectory with feedback events, observing whether the matching strategies designed around that stage continue to receive positive feedback within the stage's impact window. If the relevant behaviors and feedback have significantly weakened before the theoretical impact window ends, the customer can be considered to have prematurely exited the stage. This step will accordingly shorten the stage trajectory's termination time and reduce the stage's weight in subsequent matching. Conversely, if behaviors around the stage still occur frequently near the originally set end date, and recommendations or marketing triggered based on the stage profile still perform well, the impact window of the lifecycle event can be extended accordingly, or its stage intensity increased, allowing it to continue playing a role in subsequent matching and targeting. In short, the division of lifecycle stages is not fixed once determined, but needs to be dynamically adjusted based on the customer's subsequent actual behavior to determine stage boundaries and intensity.

[0048] For the event profiling layer, this step deals with single, significant events that are very noticeable in the short term but may not have a long-term, lasting impact on customer behavior. On one hand, if after a large purchase or major medical expense, within the system's preset observation window, the frequency of subsequent behavior related to the product or service rapidly declines, and targeted outreach based on the event profile does not bring sustained significant benefits, it indicates that the event was more of a one-off impulse or sporadic event. This step will accelerate the reduction of the corresponding tag's weight according to a rapid decay curve, allowing it to quickly exit decision-making after its short-term effect, avoiding misinterpretation as a long-term preference. On the other hand, if, for a period of time after an event, behavior related to the event theme remains at a high level, and the feedback score when using the event profile for matching... If the event remains above the baseline, this step can moderately slow down its decay rate, or even aggregate the event into a new life cycle stage trajectory and transfer it to the life cycle stage profile layer for management, thereby reflecting that the short-term event has evolved into a relatively stable life stage.

[0049] After making feedback-based local adjustments to the three profiling layers, this step also requires a holistic review to determine whether the current multi-timescale weight configuration remains suitable for subsequent similar tasks. To this end, the overall performance of several similar business tasks can be summarized and analyzed over an observation period, comparing the consistency between the overall matching score and the actual results under different weight combinations. For example, an ideal weight vector for a certain task type can be set as follows: The weight vector in actual operation is When it is found that the event profiling layer contributes too much to a certain type of task but generates more complaints, or that the long-term structural profiling layer has too little weight, leading to long-term value prediction bias, adjustments can be made slowly in the following manner: ;in, The step size is determined by a value between 0 and 1. A smaller value indicates a smoother adjustment, avoiding drastic fluctuations in weights in the short term. Through repeated iterations, the weight vector will gradually approach a more suitable configuration under the guidance of historical results, making subsequent matching tasks of the same type more stable in the long run.

[0050] Example 2: The design of a customer profile matching and positioning system according to the present invention is based on the method in Example 1, specifically as follows... Figure 2 The following modules are shown: The data acquisition and processing module is used to clean and time-correct multi-source behavioral data under a unified time base, and to construct behavioral time series and label time series.

[0051] Specifically, within a pre-defined data collection period, the data collection and processing module continuously collects customer-related behavioral and transaction data from e-commerce transaction systems, content browsing systems, search log systems, third-party advertising systems, and external cooperation channel interfaces. This includes search keywords and their trigger times, exposure records and dwell time, adding items to cart, order and payment behaviors, ratings and reviews, after-sales and refund applications, as well as basic account attribute information and terminal device environment information. For behavioral records from different systems, regions, and time zones, the data collection and processing module converts the local time recorded by each business system to a unified time base. Through time zone conversion and format normalization, the data is converted into a time representation under a unified time base. Using the customer's primary identifier as the aggregation key, multi-source behavioral records of the same customer are merged and sorted to form a standardized behavioral time series. Based on this, according to the tag system pre-designed in Example 1, the original behavioral records in the behavioral time series are mapped to activation records at the tag level. A tag time series is constructed under a unified time base, and the statistical characteristics of the tag's first activation time, most recent activation time, cumulative activation count, adjacent activation time interval, and activation intensity in the preset time slice are calculated to provide input for the profile modeling module to identify high-intensity activation intervals of tags.

[0052] The profile modeling module is used to identify high-intensity activation intervals of tags based on tag time series, forming a multi-time-scale profile structure, which includes long-term structural profiles, life-cycle stage profiles, and event profiles.

[0053] Specifically, the profile modeling module scans the tag time series along the time axis using a preset-length sliding time window under a unified time base. It compares the tag activation intensity within the window with the customer's historical baseline intensity. When the activation intensity shows a sustained and significant increase relative to the baseline within a certain continuous time interval, that time interval is marked as a high-intensity activation interval for that tag. Based on this, the profile modeling module aligns the high-intensity activation intervals for each tag dimension along the time axis, taking a single customer as the unit. It identifies candidate fragments of phased events formed by the concentrated outbreak of multiple semantically related tags within a specific time window. Combining environmental event sequences and customer basic attributes, and according to the pre-defined pattern library in Example 1, the candidate fragments are categorized into different lifecycle event types, constructing a record. The system creates a lifecycle stage profile and trajectory, including the event type, start time, end time, stage intensity, impact window, and related tag set. Simultaneously, it removes high-intensity segments already marked as lifecycle event impact windows from the tag timeline, and calculates long-term typical values ​​and activity levels for the remaining tag activation records using time-weighted averaging, quantile statistics, or other robust statistical methods to form a long-term structural profile. For single events that do not form a complete lifecycle stage but have significant business implications in a short period, the profile modeling module extracts the event occurrence time, event type, event intensity, and related tag set to construct an event profile layer. This completes the modeling of a multi-timescale profile structure consisting of a long-term structural profile, a lifecycle stage profile, and an event profile.

[0054] The matching calculation module is used to analyze the time sensitivity and risk tolerance of business tasks, and calculate comprehensive analysis indicators based on the tag time instability index and the proportion index of phased behavior. Based on the comprehensive analysis indicators, it determines whether to introduce time-related dynamic decay. It calculates the long-term structural profile score, life cycle stage profile score and event profile score respectively, and generates a comprehensive matching score by weighted summation and fusion of the scores of each profile layer. Based on the comprehensive matching score, it outputs the target customer set.

[0055] Specifically, after receiving a matching and positioning request from the upper-layer business system, the matching calculation module parses configuration information such as business target type, time sensitivity, risk tolerance, and profile template from the request. Based on the business target type and time sensitivity, it derives the basic weight ratio of the three profile layers—long-term structured profile, lifecycle stage profile, and event profile—for this task. Within the candidate customer set, the matching calculation module calculates the tag time instability index and the stage behavior proportion index based on the tag time series and multi-timescale profile structure. The tag time instability index reflects the average level of the difference in activation intensity between the tags of interest in this task and the long-term historical window. The stage behavior proportion index measures the proportion of behavior falling within the lifecycle stage influence window or short-term event influence window in the overall behavior. The two indices are then integrated into a comprehensive analysis indicator for decision-making as described in Example 1. The matching calculation module compares the comprehensive analysis index with the dynamic decay activation threshold. When the comprehensive analysis index reaches or exceeds the threshold, time-related dynamic decay is introduced into the matching score calculation process of each profile layer. The contribution of tags that are expired or at the edge of the influence window, the influence of the life cycle stage, and the influence of short-term events are corrected according to the preset time decay curve. When the comprehensive analysis index is lower than the threshold, the intra-layer benchmark score without time decay is used or weak decay is only implemented in the event profile layer. On this basis, the matching calculation module calculates the long-term structural profile score, the life cycle stage profile score, and the event profile score respectively, and performs a weighted summation of the scores of the three profile layers according to the fusion weight set in Example 1 to obtain the comprehensive matching score of the candidate customer. The candidate set is sorted from high to low according to the comprehensive matching score. Combining the score threshold, the gradient strategy corresponding to different score intervals, and resource constraints, the target customer set is selected from the sorting results and output.

[0056] The feedback adjustment module is used to construct a feedback event sequence by combining recommendation, marketing or risk control feedback, and to adjust the multi-timescale profile and fusion weight according to the feedback event sequence.

[0057] Specifically, after the matching calculation module outputs the target customer set and the upper-level system completes the recommendation, marketing, benefits distribution, or risk control processing, the feedback adjustment module collects information related to customer feedback under a unified time base. This includes whether the customer saw, clicked, or ignored the recommended content; whether they completed the order and payment within the specified time; whether they unsubscribed from or complained about the marketing outreach; whether the credit approval was approved; whether they repaid on time after using the credit; whether there were any overdue payments; and whether manual review was triggered. The feedback information is anonymized and formatted, and sorted according to the customer's primary identifier and the unified time base to construct a feedback event sequence. The feedback adjustment module then uses this feedback event sequence as a basis for further processing. The system extracts performance metrics such as click-through rate, conversion rate, unsubscription rate, and complaint rate. Feedback scores are constructed at the customer level and across dimensions such as score ranges, tag combinations, and lifecycle stages. Based on these scores, dynamic adjustments are made to the weight and aging strategy of long-term tags in the long-term structured profile layer, the influence window and stage intensity of lifecycle stages in the lifecycle stage profile layer, and the decay strategy of events in the event profile layer. Simultaneously, the feedback adjustment module summarizes the overall performance of multiple similar business tasks over an observation period. Based on the consistency between the overall matching score and the actual results, the fusion weights of the long-term structured profile, lifecycle stage profile, and event profile are slowly updated.

[0058] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, and a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0059] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0060] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0061] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0062] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A customer profile matching positioning method, characterized in that, The method comprises the following steps: cleaning and time correction of multi-source behavior data under a unified time base, construction of behavior time series and label time series; identification of label high-intensity activation intervals based on the label time series, formation of a multi-time scale portrait structure, and the multi-time scale portrait structure comprising a long-term structural portrait, a life cycle stage portrait, and an event portrait; analysis of time sensitivity and risk tolerance for business tasks, calculation of a comprehensive analysis index based on a label time instability index and a stage behavior proportion index, and judgment of whether to introduce time-related dynamic attenuation based on the comprehensive analysis index; calculation of long-term structural portrait scores, life cycle stage portrait scores, and event portrait scores, and generation of a comprehensive matching score based on weighted summation of the portrait layer scores; output of a target customer set based on the comprehensive matching score; construction of a feedback event sequence in combination with feedback from recommendation, marketing, or risk control, and adjustment of the multi-time scale portrait and fusion weights based on the feedback event sequence.

2. The customer profiling and targeting method of claim 1, wherein: The process of identifying label high-intensity activation intervals based on the label time series comprises: scanning along the time axis of the label time series using a preset length of a sliding time window; comparing the label activation intensity within the window with the historical baseline intensity of the corresponding customer; when the label activation intensity in a certain continuous time interval is continuously and significantly raised relative to the historical baseline, marking the time interval as a high-intensity activation interval of the label and recording its start and end times and peak intensity.

3. The method of claim 1, wherein: The formation of the multi-time scale portrait structure comprises: time axis alignment of the high-intensity activation intervals of the customer in each label dimension, identification of stage event candidate segments; semantic classification of the stage event candidate segments in combination with the environmental event sequence and the basic attributes of the customer, construction of the life cycle stage portrait; exclusion of high-intensity segments within the life cycle event influence window from the label time axis, construction of the long-term structural portrait based on the label activation records of the remaining time period; extraction of single events that do not form complete life cycle stages but have significant business significance, construction of the event portrait.

4. The customer profiling and matching positioning method of claim 3, wherein: The construction of the life cycle stage portrait comprises: for the identified life cycle event types, recording the type, start time, end time, stage intensity, and related label set of the corresponding life cycle event for the customer; when the same type of life cycle event occurs multiple times on the behavior time axis of the customer, connecting these life cycle events in chronological order to form a stage trajectory.

5. The method of claim 1, wherein: Calculation of the comprehensive analysis index based on the label time instability index and the stage behavior proportion index comprises: weighted fusion of the label time instability index and the stage behavior proportion index to obtain a comprehensive analysis index value; comparison of the comprehensive analysis index value with a preset dynamic attenuation enabling threshold, and judgment of whether to introduce time-related dynamic attenuation when calculating the portrait layer scores based on the comparison result.

6. The customer profiling and matching positioning method of claim 5, wherein: The label time instability index is used to reflect the average level of the difference in activation intensity between the recent observation window and the long-term historical window of the label concerned in the task; and the phased behavior proportion index is used to measure the proportion of behaviors falling into the life cycle stage influence window or the short-term event influence window in the overall behaviors within the current task and the candidate customer range.

7. The method of claim 1, wherein: The scores of the respective portrait layers are calculated, including: calculating a long-term structural portrait score based on the similarity between the typical value of the label of the customer in a long-term range and the target portrait template; calculating a life cycle stage portrait score based on the affinity between the current life cycle stage of the customer and the business task, and introducing a timeliness weight when time-related dynamic attenuation is enabled; calculating an event portrait score based on the correlation between the single event recently occurred by the customer and the business task, and introducing a time attenuation weight when time-related dynamic attenuation is enabled.

8. The method of claim 1, wherein: The feedback event sequence includes customer feedback data from recommendation, marketing or risk control operations, which is aligned with the behavior time sequence after desensitization and formatting processing.

9. The method of claim 1, wherein: The feedback event sequence adjusts the fusion weight, including: analyzing the comprehensive performance of several similar business tasks; based on the consistency between the comprehensive matching score and the actual result, the fusion weight ratio of the long-term structural portrait, the life cycle stage portrait and the event portrait is slowly adjusted through iterative learning.

10. A client profiling and matching positioning system, characterized by, The positioning system is used to implement the method of any one of claims 1-9, comprising the following modules: a data acquisition and processing module for cleaning and time correction of multi-source behavior data under a unified time base, constructing a behavior time sequence and a label time sequence; a portrait modeling module for identifying a label high-intensity activation interval based on the label time sequence, forming a multi-time scale portrait structure, and the multi-time scale portrait structure including a long-term structural portrait, a life cycle stage portrait and an event portrait; a matching calculation module for analyzing time sensitivity and risk tolerance for a business task, and calculating a comprehensive analysis index based on a label time instability index and a phased behavior proportion index, and determining whether to introduce time-related dynamic attenuation based on the comprehensive analysis index; long-term structural portrait score, life cycle stage portrait score and event portrait score are calculated respectively, and the scores of the respective portrait layers are fused according to weighted summation to generate a comprehensive matching score; outputting a target customer set based on the comprehensive matching score; a feedback adjustment module for constructing a feedback event sequence in combination with recommendation, marketing or risk control feedback, and adjusting the multi-time scale portrait and the fusion weight according to the feedback event sequence.

Images