An agent-based multi-channel customer analysis method and system

Abstract

The application relates to the technical field of customer analysis, and discloses a multi-channel customer analysis method and system based on an intelligent agent, which comprises the following steps: standardizing and performing cross-channel binding processing on multi-source behavior data to obtain a shared entry standardized touch sequence; calculating a single-group behavior heat value of the behavior data in the shared entry standardized touch sequence, dividing the shared entry standardized touch sequence into multiple diversion time windows, and calculating window features and heat features of the diversion time windows; calculating a pairing value between a stay record and the diversion time windows, sorting the diversion time windows according to the pairing value, and obtaining a stay link candidate sorting table of the stay record; and performing value verification and customer conversion rate analysis on the diversion time windows under the entry identifier. The application realizes unified integration of multi-channel diversion behavior data, extraction of diversion time windows, feature quantization and intelligent matching of stay clues, and completes multi-channel customer conversion efficiency analysis.

Description

Technical Field

[0001] This invention relates to the field of customer analysis using multi-agent systems, and more particularly to a multi-channel customer analysis method and system based on agents. Background Technology

[0002] With the widespread adoption of digital marketing touchpoints such as short videos, live streaming, WeChat Work, landing forms, and offline store QR codes, enterprise customer acquisition systems have shifted from traditional single-channel advertising to a multi-stage, multi-channel collaborative operation model encompassing public domain lead generation, live streaming engagement, and private domain lead generation. Short video content is responsible for identifying potential customers and cultivating user interest; live streaming scenarios leverage interactive explanations and promotional activities to strengthen customer intent; and shared entry points such as universal QR codes and private domain forms complete customer information retention and long-term operation. Under this business model, accurately tracing the upstream lead generation path corresponding to each lead and quantifying the conversion capabilities of each channel and touchpoint are core prerequisites for enterprises to assess channel value, optimize marketing campaigns, improve overall conversion efficiency, and control operating costs.

[0003] In real-world business scenarios where short videos and live streams are used to manage private domain leads, a large number of users browse public domain content anonymously and are subsequently redirected to the lead management page through public entry points such as universal QR codes for live streams, standardized posters in stores, and shared passwords in comment sections. These shared entry points are reused by a massive number of users, and the lead management process only records result information such as customer name, mobile phone number, and intended needs, failing to retain unique identifiers and link parameters that can be uniquely linked to prior public domain behavior.

[0004] Existing patent CN120634624B discloses a method for constructing customer profiles based on big data collection. This solution collects two types of data: website click logs and transaction records. It builds a customer feature matrix and performs data cleaning and standardization. By evaluating the importance of features, it generates a weight vector and combines weighted calculations to obtain a quantitative result of the customer profile. Then, it uses an adaptive threshold density clustering algorithm to complete user segmentation, combines profile tags to carry out differentiated marketing, and dynamically adjusts the clustering parameters based on the marketing conversion rate, forming a closed-loop system from data collection, analysis and modeling to marketing implementation, with good real-time performance and self-adaptability. However, this solution focuses on static customer profile construction and user segmentation operations, and does not perform customer analysis for multi-channel anonymity scenarios.

[0005] To address this issue, this invention proposes a multi-channel customer analysis method based on intelligent agents, which enhances the observability and interpretability of enterprises' collaborative traffic acquisition process across multiple channels such as short videos, live streaming, store posters, and private domain operations. It can intuitively locate shared traffic acquisition units with high or low conversion entry points, thereby enabling targeted traffic allocation and optimization of channel placement strategies. Summary of the Invention

[0006] This invention proposes a multi-channel customer analysis method and system based on intelligent agents. Step S1 collects end-to-end behavioral data from multiple channels, including short videos, live streams, and offline stores. It unifies field formats and encoding rules, and uses entry point identifiers to complete cross-channel binding, generating a standardized touchpoint sequence with shared entry points. This integrates scattered and heterogeneous data, establishes a unified data standard, and achieves orderly collection and correlation of user behavior data across all channels. Step S2 calculates the heat value of a single set of behaviors, divides the global time axis according to a fixed time scale to obtain a traffic-driving time window, and simultaneously extracts window features and heat features. This quantifies the traffic-driving contribution of each behavior, providing precise... Step S3 accurately delineates effective lead generation time windows and completes the structured expression of features; Step S4 constructs lead generation features by filtering lead generation form data, calculates matching value by combining time sequence, content, behavior, and popularity dimensions, and generates a candidate ranking table of lead generation links based on descending score, which can intelligently match lead generation records with corresponding lead generation time windows; Step S4 filters lead generation time windows by entry point identifier, completes value verification based on preset thresholds, and calculates the average lead generation conversion rate by overall and channel-specific criteria, which can filter out low-relevance invalid samples, quantify the customer conversion efficiency of each shared lead generation entry point and its affiliated channels, and complete a comprehensive judgment of multi-dimensional customer lead generation effect.

[0007] To achieve the above objectives, the present invention provides a multi-channel customer analysis method based on intelligent agents, comprising the following steps: S1: Collect multi-source behavioral data from the entire customer acquisition process across multiple channels, standardize the multi-source behavioral data and perform cross-channel binding processing to obtain a standardized touchpoint sequence for shared entry points; S2: Calculate the single-group behavior heat value of the behavior data in the standardized touch point sequence of the shared entry point, divide the standardized touch point sequence of the shared entry point into multiple traffic diversion time windows, and calculate the window characteristics and heat characteristics of the traffic diversion time windows; S3: Extract lead generation records from the lead generation form submission data of the standardized touchpoint sequence of the shared entry point and generate lead generation features. Based on the lead generation features and the window features and popularity features of the traffic generation time window, calculate the pairing value between the lead generation record and the traffic generation time window. Sort the traffic generation time windows according to the pairing value to obtain the lead generation link candidate sorting table of the lead generation record. S4: Based on the candidate ranking table of lead generation links, perform value verification and customer conversion rate analysis on the lead generation time window under the entry point identifier.

[0008] As a further improvement of the present invention: Furthermore, the multi-source behavioral data refers to the total amount of recordable interactive behavioral data generated by users in the entire customer acquisition process, including short video pre-heating, live streaming browsing, store outreach, private domain acquisition, and lead generation conversion, across public, offline, and private channels. This behavioral data includes short video placement and interaction data, live streaming content display and pop-up exposure data, store poster outreach data, shared redirect entry access data, WeChat addition data, landing page browsing data, and lead generation form submission data. Each piece of behavioral data includes the time of occurrence, duration, channel type, content identifier, entry identifier, and action type.

[0009] Furthermore, step S1, which standardizes and cross-channels the multi-source behavioral data, also includes: S11: Perform unified data format, coding standard, and null value pre-filling processing on the behavior occurrence time, channel type, content identifier, entry identifier, and action type in the multi-source behavior data to obtain standardized multi-source behavior data; S12: Using the entry identifier as the sole grouping criterion, filter behavioral data with the same entry identifier and bind them to form a set of behavioral data under that entry identifier; S13: Sort the behavior data in the behavior data set according to the order of the behavior occurrence time to obtain the standardized contact sequence of the shared entry point.

[0010] Further, in step S2, the single-group behavior heat value of the behavior data in the standardized touchpoint sequence of the shared entry point is calculated, and the standardized touchpoint sequence of the shared entry point is divided into multiple traffic-driving time windows, including: S21: Extract the global timeline formed by the occurrence times of all behaviors in the standardized touchpoint sequence of the shared entry point, divide the global timeline into continuous time scales, and count all behavior data under each time scale; S22: Calculate the single-group behavior heat value of any set of behavior data under the time scale, and sum up the single-group behavior heat values ​​of all behavior data under the time scale to obtain the behavior heat value of all behavior data under the time scale. Specifically, the formula for calculating the single-group behavior heat value of the time-scale behavior data is as follows: ; in, This represents the single-group behavior popularity value of the nth group of behavior data at the time scale, where N represents the total number of behavior data at the time scale. This indicates the channel type of the nth set of behavioral data. Indicates channel type Basic channel weight, This represents the action type of the nth set of behavioral data. Indicates action type Basic action weights, Each represents the global balance coefficient. This represents the duration of the action in the nth set of action data. This represents the interaction frequency of the promotional material associated with the content identifier of the nth set of behavioral data over a time scale. The interaction frequency includes the number of new likes, new comments, and new shares. Represents a logarithmic function with the natural constant as its base; S23: Perform proximity heat correction processing on the behavior heat values ​​of all behavior data under the time scale to obtain the corrected behavior heat values ​​of the time scale; Specifically, the formula for neighboring popularity correction of all behavioral data at the m-th time scale is as follows: ; in, This represents the corrected behavior heat value at the m-th time scale. This represents the behavioral heat value of all behavioral data at the m-th time scale. This represents the behavioral heat value of all behavioral data at the (m+1)th time scale. Let represent the behavioral popularity value of all behavioral data at the (m-1)th time scale, and represent the mean behavioral popularity value across all time scales. Indicates the correction factor; S24: Select the time scale in which the corrected behavior heat value is higher than the preset benchmark heat value as the effective traffic diversion interval, and use the behavior data sequence under the continuous effective traffic diversion interval as the traffic diversion time window in the shared entry standardized touch point sequence.

[0011] Furthermore, step S2, which calculates the window characteristics and heat characteristics of the drainage time window, also includes: S25: Calculate the window length of the traffic acquisition time window, the behavior occurrence time corresponding to the two ends of the window, the proportion of behavior data of each channel type, the proportion of behavior data of each action type, and the average behavior duration of the behavior data. Then, concatenate the calculation results as the window features of the traffic acquisition time window. S26: Calculate the average corrected behavioral heat value of each time scale within the drainage time window, and use it as the heat feature of the drainage time window.

[0012] Further, step S3, which involves extracting lead generation records and generating lead generation features from the lead generation form submission data of the standardized contact point sequence of the shared entry point, includes: The lead collection record includes the time of the lead collection form submission, the duration of the submission, the entry label, and the user's intended needs tags. Based on the entry identifier label of the data submitted by the lead generation form, filter the preceding behavior data of the same user with the same entry identifier label, and calculate the total duration of the preceding behavior data. The preceding behavior data includes store poster reach data with the entry identifier label, behavior occurrence time earlier than the time of the data submitted by the lead generation form, and shared jump entry access data. The lead generation characteristics are obtained by concatenating the time of occurrence of the data submission on the lead generation form, the duration of the behavior, the entry identifier label, the sum of the duration of the preceding behavior data, and the intention demand label.

[0013] Furthermore, step S3 calculates the pairing value between lead generation records and lead generation time windows, and sorts the lead generation time windows according to the pairing value. This also includes: S31: Obtain the lead retention characteristics of the lead retention records, as well as the window characteristics and popularity characteristics of the lead generation time window; S32: Based on the time of occurrence of the data submission behavior in the data retention form in the data retention feature and the time of occurrence of the behavior corresponding to the two ends of the window in the window feature, calculate the temporal matching degree between the data retention record and the traffic generation time window; S33: Based on the entry identifier label and intention demand tag of the data submitted by the lead retention form in the lead retention feature, as well as the entry identifier and content identifier set of the behavioral data in the traffic acquisition time window, calculate the content matching degree between the lead retention record and the traffic acquisition time window; S34: Based on the sum of the behavior duration of the preceding behavior data in the lead retention features and the behavior duration of the lead retention form submission data, calculate the behavior matching degree between the lead retention record and the lead generation time window; S35: Based on the window characteristics and popularity characteristics of the traffic acquisition time window, calculate the popularity weighting coefficient of the traffic acquisition time window, and weight the temporal matching degree, content matching degree and behavioral matching degree between the lead generation record and the traffic acquisition time window to obtain the pairing value between the lead generation record and the traffic acquisition time window.

[0014] Furthermore, in step S4, based on the candidate ranking table for lead generation links, the value verification and customer conversion rate analysis are performed on the lead generation time window under the entry identifier, including: S41: For any entry identifier L, filter the lead generation time window under the entry identifier L, and based on the lead retention link candidate sorting table, obtain the pairing value between the lead generation time window and any lead retention record, and retain the lead generation time window whose pairing value is higher than the preset value verification threshold. S42: Calculate the average lead generation conversion rate for the retained lead generation time window and the average lead generation conversion rate for each channel type; S43: The average lead generation conversion rate and the average lead generation conversion rate of each channel are used as the results of multi-channel customer acquisition analysis for the shared lead generation entry associated with the entry identifier L.

[0015] The present invention also provides an agent-based multi-channel customer analysis system, which includes a data acquisition agent, a sequence processing agent, a time period analysis agent, and a customer analysis agent, to implement the technical steps of the agent-based multi-channel customer analysis method described above.

[0016] Compared with existing technologies, this invention proposes a multi-channel customer analysis method and system based on intelligent agents, which has the following beneficial effects: First, this invention divides the global timeline into 5-minute intervals. By removing data from lead generation forms, it focuses on front-end traffic generation activities to calculate popularity, accurately identifying effective traffic generation intervals and generating traffic generation time windows. Specifically, this invention uses basic channel weights and basic action weights coupled with behavior duration and interaction frequency to construct a formula for a single set of behavior popularity values. It combines a logarithmic function to weaken the marginal effect of duration, comprehensively quantifying the traffic generation value of a single behavior. The calculation results closely match real user behavior characteristics and are corrected by using the popularity of adjacent intervals. It integrates traffic activity before and after time periods to avoid misjudgments caused by data deviations in a single interval, improving the objectivity and stability of popularity evaluation. Simultaneously, this invention relies on a unified benchmark popularity to screen effective traffic generation intervals and merges consecutive effective intervals to generate traffic generation time windows, automatically dividing traffic generation periods from massive amounts of continuous behavior data. The entire set of rules is logically coherent and computationally efficient, requiring no reliance on external historical data. This ensures the accuracy of multi-channel traffic generation behavior segmentation and has good engineering feasibility, providing reliable time period data support for subsequent lead generation attribution and channel performance customer analysis.

[0017] Meanwhile, this invention uses shared traffic entry points as independent analysis units, completing data collection and filtering according to entry point identifiers. It filters low-relevance traffic entry time windows based on preset value verification thresholds, eliminating invalid data interference and ensuring the validity of the analysis samples. Furthermore, this invention calculates the overall lead generation conversion rate and channel-specific lead generation conversion rate using behavioral data proportions, quantifying the overall traffic generation effectiveness and the contribution level of each channel under a single metric. By focusing on the operational differences of different shared traffic entry points, it clearly distinguishes the conversion capabilities of each channel within its corresponding entry point, intuitively identifying high-conversion or low-conversion shared traffic entry units. This provides quantitative evidence for operators to allocate traffic strategically, optimize channel placement strategies, and strengthen the operational capabilities of high-quality entry points. Simultaneously, it improves the full-link traffic conversion evaluation system, enhancing the comprehensiveness and practicality of data analysis and enabling multi-channel customer analysis. Attached Figure Description

[0018] Figure 1This is a flowchart illustrating a multi-channel customer analysis method based on intelligent agents, as provided in an embodiment of the present invention.

[0019] Figure 2 This is a structural diagram of a multi-channel customer analysis system based on intelligent agents provided in an embodiment of the present invention.

[0020] Figure 3 This is an experimental comparison diagram provided for one embodiment of the present invention.

[0021] In the diagram: 101, Data Acquisition Agent; 102, Sequence Processing Agent; 103, Time Period Analysis Agent; 104, Customer Analysis Agent. Detailed Implementation

[0022] The realization of the objectives, functional characteristics, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0023] This invention provides a multi-channel customer analysis method and system based on intelligent agents. The executing entity of the intelligent agent-based multi-channel customer analysis method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this invention: a server, a terminal, etc. In other words, the intelligent agent-based multi-channel customer analysis method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0024] Reference Figure 1 Embodiment 1 of the present invention is as follows: A multi-channel customer analytics method based on intelligent agents, the method comprising: S1: Collect multi-source behavioral data from the entire customer acquisition process across multiple channels, standardize the multi-source behavioral data and perform cross-channel binding processing to obtain a standardized touchpoint sequence for shared entry points.

[0025] Specifically, the multi-source behavioral data refers to all recordable interactive behavioral data generated by users in the entire customer acquisition process, including short video pre-heating, live streaming browsing, store outreach, private domain acquisition, and lead generation conversion, across public, offline, and private channels. This behavioral data includes short video placement and interaction data, live streaming content display and pop-up exposure data, store poster outreach data, shared redirect entry access data, WeChat addition data, landing page browsing data, and lead generation form submission data. Each piece of behavioral data includes the time of occurrence, duration, channel type, content identifier, entry identifier, and action type.

[0026] As an embodiment of the present invention, the customer acquisition process is divided into five stages: First, short video content is used to complete the pre-event promotion and attract user attention; then, users enter the live broadcast room to browse content and participate in interaction; at the same time, users can also receive promotional information in offline store scenarios to complete offline store reach; traffic from various channels is redirected to private domain channels through a unified shared entry QR code to realize traffic transfer; finally, users submit personal and intention information on the private domain page to complete lead generation conversion.

[0027] Specifically, the short video delivery and interaction data records the online delivery, distribution, and exposure of short video content, as well as relevant data on user interactions such as watching, liking, commenting, and forwarding, reflecting the reach and user engagement of short video channels; the live stream content display and pop-up exposure data includes playback records of live stream footage and explanation content, as well as data such as the display time, display location, and exposure frequency of QR codes and activity pop-ups within the live stream page, which are core records of live stream channel traffic generation behavior; the store poster reach data includes records of the placement location, effective display time, and user close-range contact of promotional posters posted and displayed in offline stores, representing the user reach in offline physical scenarios; the shared jump entrance access data includes user clicks, QR code scans, etc. Access logs generated from unified traffic entry points such as shared QR codes, exclusive passwords, and redirect links (using actions like entering passwords) are used to determine the traffic entry point to which the behavior belongs. The WeChat friend addition data includes records generated by users initiating WeChat friend addition operations through various traffic channels, including addition time and source entry information, representing the flow of public domain traffic to private domain traffic. The landing page browsing data includes data such as user visit duration, page browsing trajectory, and operational behavior after jumping to the event landing page or product details page, reflecting the user's level of intent towards the promoted content. The lead generation form submission data consists of form texts formed by users filling in and submitting information such as name, contact information, and intended needs on the landing page or private domain interface; this is a key indicator of completed customer lead conversion. Furthermore, the occurrence time of the behavior is the timestamp of the user operation corresponding to the behavior data; the channel type includes short video channels, live streaming channels, offline store channels, shared jump entry channels, and private domain access channels; the content identifier is a unique code assigned to specific promotional content under each channel, with short video channels corresponding to video content numbers, live streaming channels corresponding to live stream sessions and pop-up segment numbers, and offline store channels corresponding to store numbers, poster numbers, etc., used to accurately locate the specific promotional materials corresponding to a single piece of behavior data, achieving refined content-level effect attribution; the entry identifier is the unique identifier of the shared traffic entry point. One identification code corresponds to a unique QR code, traffic generation code, jump link, or other unified entry number. This is used to bind behavioral data from different traffic generation channels and at different times across channels, determining that all behavioral data belongs to the same traffic generation entry point and ensuring the integrity of data throughout the entire chain. The action type is used to mark the user's specific operation behavior, including video playback, content interaction, pop-up viewing, poster delivery, entry access, adding via WeChat, landing page browsing, form submission, etc. This is used to distinguish between ordinary browsing behavior and conversion behavior, and to quickly identify public domain traffic generation actions, private domain acquisition actions, and final lead generation conversion actions. As another embodiment of the present invention, the full-link behavior data of a user under a single shared entry point (entry point identifier: S001) is used as an example. Each piece of behavior data carries core fields such as behavior occurrence time, channel type, content identifier, entry point identifier, and action type, including: 1. Short video channel: 2025-04-25 09:12:05, behavior duration: 2 minutes, content identifier V028, entry point identifier S001, behavior is video playback; 2. Live streaming channel: 2025-04-25 09:15:18, behavior duration: 5 minutes, content identifier L106, entry point identifier S001, behavior is pop-up QR code exposure; 3. Lead generation conversion: 2025-04-25 09:22:16, behavior duration: 10 minutes, entry point identifier S001, behavior is form submission for lead generation, intended product is family SUV, intended store is M012.

[0028] Step S1, which standardizes and cross-channels the multi-source behavioral data, also includes: S11: Perform unified data format, coding standard, and null value pre-filling processing on the behavior occurrence time, channel type, content identifier, entry identifier, and action type in the multi-source behavior data to obtain standardized multi-source behavior data; Specifically, null values ​​in behavioral data are filled with a fixed fill value of null. S12: Using the entry identifier as the sole grouping criterion, filter behavioral data with the same entry identifier and bind them to form a set of behavioral data under that entry identifier; S13: Sort the behavior data in the behavior data set according to the order of the behavior occurrence time to obtain the standardized contact sequence of the shared entry point.

[0029] Specifically, the behavioral data in the shared entry standardized touchpoint sequence have the same entry identifier.

[0030] S2: Calculate the single-group behavior heat value of the behavior data in the standardized touch point sequence of the shared entry point, divide the standardized touch point sequence of the shared entry point into multiple traffic diversion time windows, and calculate the window characteristics and heat characteristics of the traffic diversion time windows.

[0031] Specifically, step S2 calculates the single-group behavior heat value of the behavior data in the standardized touchpoint sequence of the shared entry point, and divides the standardized touchpoint sequence of the shared entry point into multiple traffic-driving time windows, including: S21: Extract the global timeline formed by the occurrence times of all behaviors in the standardized touchpoint sequence of the shared entry point, divide the global timeline into continuous time scales, and count all behavior data under each time scale; Specifically, by obtaining the earliest and latest behavior occurrence times of the behavior data in the standardized touchpoint sequence of the shared entry point, the earliest and latest behavior occurrence times are used as the time points at both ends of the global time axis, and the length of the time scale is 5 minutes. Furthermore, the behavioral data collected at each time scale does not include data submitted through the lead generation form; S22: Calculate the single-group behavior heat value of any set of behavior data under the time scale, and sum up the single-group behavior heat values ​​of all behavior data under the time scale to obtain the behavior heat value of all behavior data under the time scale. Specifically, the formula for calculating the single-group behavior heat value of the time-scale behavior data is as follows: ; in, This represents the single-group behavior popularity value of the nth group of behavior data at the time scale, where N represents the total number of behavior data at the time scale. This indicates the channel type of the nth set of behavioral data. Indicates channel type Basic channel weight, This represents the action type of the nth set of behavioral data. Indicates action type Basic action weights, All represent the global balance coefficient (default setting). It is 2. (1.5) This represents the duration of the action in the nth set of action data. This represents the interaction frequency of the promotional material associated with the content identifier of the nth set of behavioral data over a time scale. The interaction frequency includes the number of new likes, new comments, and new shares. Represents a logarithmic function with the natural constant as its base; Furthermore, the basic channel weights for short video channels, live streaming channels, offline store channels, shared redirect entry channels, and private domain access channels are set to 1.2, 1.3, 1, 1.1, and 0.4 respectively by default; the basic action weights for video playback, content interaction, pop-up viewing, poster delivery, entry access, adding via WeChat, and landing page browsing are set to 1.4, 1.2, 1.1, 1.0, 1.2, 0.7, and 0.7 respectively by default. S23: Perform proximity heat correction processing on the behavior heat values ​​of all behavior data under the time scale to obtain the corrected behavior heat values ​​of the time scale; Specifically, the formula for neighboring popularity correction of all behavioral data at the m-th time scale is as follows: ; in, This represents the corrected behavior heat value at the m-th time scale. This represents the behavioral heat value of all behavioral data at the m-th time scale. This represents the behavioral heat value of all behavioral data at the (m+1)th time scale. This represents the behavioral heat value of all behavioral data at the (m-1)th time scale. This represents the average behavioral heat value across all time scales. Indicates the correction factor; S24: Select the time scale in which the corrected behavior heat value is higher than the preset benchmark heat value as the effective traffic diversion interval, and use the behavior data sequence under the continuous effective traffic diversion interval as the traffic diversion time window in the shared entry standardized touch point sequence.

[0032] Specifically, the default preset baseline heat value is set to 18; Furthermore, the single-group behavior heat value is used to quantify the comprehensive traffic-driving contribution intensity of a single behavior data. It combines channel attributes, user operation type, behavior dwell time and interaction performance multi-dimensional indicators to calculate the quantified heat value. The higher the heat value, the stronger the attraction of the promotional content and touch action corresponding to the behavior data to users and the more prominent the traffic-driving effect. At the same time, the logarithmic function is used to weaken the distortion of heat value caused by long dwell time, which can objectively distinguish the traffic-driving value of different behavior data and provide a basic quantitative basis for the overall heat statistics of the time period. The corrected behavior heat value is based on the behavior heat value and is adjusted by combining the behavior heat values ​​of adjacent time scales to eliminate the evaluation bias caused by data fluctuations and random traffic at a single time scale.

[0033] Step S2, which calculates the window characteristics and heat characteristics of the drainage time window, also includes: S25: Calculate the window length of the traffic acquisition time window, the behavior occurrence time corresponding to the two ends of the window, the proportion of behavior data of each channel type, the proportion of behavior data of each action type, and the average behavior duration of the behavior data. Then, concatenate the calculation results as the window features of the traffic acquisition time window. S26: Calculate the average corrected behavioral heat value of each time scale within the drainage time window, and use it as the heat feature of the drainage time window.

[0034] S3: Extract lead generation records from the lead generation form submission data of the standardized touchpoint sequence of the shared entry point and generate lead generation features. Based on the lead generation features and the window features and popularity features of the traffic generation time window, calculate the pairing value between the lead generation record and the traffic generation time window. Sort the traffic generation time windows according to the pairing value to obtain the lead generation link candidate sorting table of the lead generation record.

[0035] Specifically, step S3, which involves extracting lead generation records and generating lead generation features from the lead generation form submission data of the standardized contact point sequence of the shared entry point, includes: The lead collection record includes the time of the lead collection form submission, the duration of the submission, the entry label, and the user's intended needs tags. Based on the entry identifier label of the data submitted by the lead generation form, filter the preceding behavior data of the same user with the same entry identifier label, and calculate the total duration of the preceding behavior data. The preceding behavior data includes store poster reach data with the entry identifier label, behavior occurrence time earlier than the time of the data submitted by the lead generation form, and shared jump entry access data. Specifically, the store poster reach data and shared redirect entry access data are users' private domain behavior data, which can be matched based on user ID; The lead generation characteristics are obtained by concatenating the time of occurrence of the data submission on the lead generation form, the duration of the behavior, the entry identifier label, the sum of the duration of the preceding behavior data, and the intention demand label.

[0036] Step S3 calculates the pairing value between lead generation records and lead generation time windows, sorts the lead generation time windows according to the pairing value, and also includes: S31: Obtain the lead retention characteristics of the lead retention records, as well as the window characteristics and popularity characteristics of the lead generation time window; S32: Based on the time of occurrence of the data submission behavior in the data retention form in the data retention feature and the time of occurrence of the behavior corresponding to the two ends of the window in the window feature, calculate the temporal matching degree between the data retention record and the traffic generation time window; Specifically, the formula for calculating the time-series matching degree between the lead retention records and the lead generation time window is as follows: ; in, This indicates the temporal matching degree between the lead generation record E and the lead generation time window G. This represents an exponential function with the natural constant as its base. Indicates the time when the data was submitted on the contact form. The time of occurrence of the behavior corresponding to the two endpoints of the window in the window features. This indicates the maximum time difference threshold (default setting is 30 minutes). S33: Based on the entry identifier label and intention demand tag of the data submitted by the lead retention form in the lead retention feature, as well as the entry identifier and content identifier set of the behavioral data in the traffic acquisition time window, calculate the content matching degree between the lead retention record and the traffic acquisition time window; Specifically, the formula for calculating the content matching degree between the lead generation records and the lead generation time window is as follows: ; in, This indicates the content matching degree between the lead generation record E and the lead generation time window G. This indicates the entry point for behavioral data within the traffic generation time window. Tags indicating intended demand This represents the set of content identifiers for behavioral data within the traffic acquisition time window. If they match, then the discriminant coefficient is identified. =1, otherwise =0, This is the content weighting coefficient, set by default. The values ​​are 0.6 and 0.4 respectively. This is achieved by converting the intended demand tags into content identifiers and selecting a set of content identifiers. Tags related to intended demand A consistent content identifier D is used, and the proportion of behavioral data with the content identifier D within the traffic acquisition time window is calculated as the content tag matching coefficient. ; S34: Based on the sum of the behavior duration of the preceding behavior data in the lead retention features and the behavior duration of the lead retention form submission data, calculate the behavior matching degree between the lead retention record and the lead generation time window; Specifically, the formula for calculating the behavioral matching degree between the lead generation records and the lead generation time window is as follows: ; in, This indicates the behavioral match between the lead generation record E and the lead generation time window G. This represents the total duration of actions based on preceding behavioral data. This indicates the window length of the drainage time window. This represents the average duration of behavioral data within the traffic generation time window. All represent time control parameters, default settings. It lasts for 10 minutes. It lasts for 5 minutes; S35: Based on the window characteristics and popularity characteristics of the traffic acquisition time window, calculate the popularity weighting coefficient of the traffic acquisition time window, and weight the temporal matching degree, content matching degree and behavioral matching degree between the lead generation record and the traffic acquisition time window to obtain the pairing value between the lead generation record and the traffic acquisition time window.

[0037] Specifically, by calculating the standard deviation of the proportion of behavioral data for each channel type and the standard deviation of the proportion of behavioral data for each action type, the product of the two standard deviations and (mean of corrected behavioral popularity value / preset maximum corrected behavioral popularity value) is calculated as the popularity weighting coefficient for the traffic acquisition time window, where the preset maximum corrected behavioral popularity value is set to 20 by default. A linear weighting method is used to weight the time-series matching degree, content matching degree, and behavior matching degree, and a heat-weighting coefficient is used to perform a second weighting on the weighting result to obtain the pairing value between lead generation records and traffic generation time windows.

[0038] It should be noted that this invention relies on an exponential function to construct a time-series matching degree, combined with the maximum time difference threshold to constrain the time correlation range, accurately depicting the time coupling relationship between lead generation behavior and traffic acquisition period. The content matching degree integrates entry identifier consistency verification and intention tag ratio calculation, taking into account traffic attribution and promotion content matching, effectively distinguishing between same-source and different-source traffic acquisition links. The behavior matching degree combines the user's total pre-processing time, form filling time, overall window duration, and average behavior duration to perform similarity calculations, deeply mining the correlation between user behavior and window behavior of traffic acquisition time window.

[0039] Furthermore, a heat-weighted coefficient is constructed by combining the standard deviation of channel and action proportions with the mean heat value. This achieves secondary weighting based on window traffic distribution characteristics and lead generation activity, weakening the influence of single-dimensional bias and improving the objectivity and robustness of matching value calculation. The entire set of calculation rules has configurable parameters, stable function value range, and strong engineering feasibility. It can automatically complete the association and matching of massive lead generation records with lead generation time periods, significantly reducing the cost of manual traceability. By accurately quantifying the correlation between lead generation records and lead generation time windows, it achieves automated and accurate matching of lead generation leads and lead generation time windows.

[0040] S4: Based on the candidate ranking table of lead generation links, perform value verification and customer conversion rate analysis on the lead generation time window under the entry point identifier.

[0041] Specifically, step S4 involves verifying the value of the lead generation time window under the entry identifier and analyzing the customer conversion rate based on the candidate ranking table of the lead generation link, including: S41: For any entry identifier L, filter the lead generation time window under the entry identifier L, and based on the lead retention link candidate sorting table, obtain the pairing value between the lead generation time window and any lead retention record, and retain the lead generation time window whose pairing value is higher than the preset value verification threshold. Specifically, the default value verification threshold is set to 0.45; S42: Calculate the average lead generation conversion rate for the retained lead generation time window and the average lead generation conversion rate for each channel type; Specifically, the formula for calculating the lead generation conversion rate is: the proportion of data submitted in the lead generation form during the lead generation time window; the formula for calculating the lead generation conversion rate for each channel type is: the product of the proportion of behavioral data under the channel type during the lead generation time window and the lead generation conversion rate. S43: The average lead generation conversion rate and the average lead generation conversion rate of each channel are used as the results of multi-channel customer acquisition analysis for the shared lead generation entry associated with the entry identifier L.

[0042] Example 2: As another embodiment of the present invention, this embodiment provides a multi-channel customer analysis system based on intelligent agents, as shown below. Figure 2 The diagram shows the structure of a multi-channel customer analysis system based on intelligent agents. The system adopts a hierarchical multi-agent architecture, which is divided into four functional agents: data acquisition agent 101, sequence processing agent 102, time period analysis agent 103, and customer analysis agent 104. Each agent performs its duties independently and works in concert to fully implement the business logic of the entire process from S1 to S4 in Example 1. The architecture is lightweight and has low coupling, making it suitable for full-link traffic acquisition data analysis scenarios.

[0043] The data acquisition intelligent agent 101 includes a multi-source data acquisition unit and a standardization processing unit. The multi-source data acquisition unit is used to collect multi-source behavioral data from the entire customer acquisition process across multiple channels. The standardization processing unit is used to standardize the multi-source behavioral data and perform cross-channel binding processing to obtain a standardized touchpoint sequence for a shared entry point. The sequence processing agent 102 is used to calculate the single-group behavior heat value of behavior data in the shared entry standardized touch point sequence, divide the shared entry standardized touch point sequence into multiple traffic diversion time windows, and calculate the window features and heat features of the traffic diversion time windows. The time-period analysis agent 103 is used to extract lead-in records and generate lead-in features from the lead-in form submission data of the standardized touchpoint sequence of the shared entry point. Based on the lead-in features and the window features and popularity features of the traffic-driving time window, the pairing value between the lead-in records and the traffic-driving time window is calculated. The traffic-driving time windows are sorted according to the pairing value to obtain the lead-in link candidate sorting table of the lead-in records. Customer analytics agent 104 is used to verify the value of traffic generation time windows under the entry point and to analyze customer conversion rates.

[0044] Example 3: This invention collects behavioral data from 1.2 million real users across five major channel types (covering short videos, live streams, offline stores, shared redirects, and private domain integration scenarios, including behavioral outreach, content interaction, entry point access, WeChat account addition, and lead submission). It compares the performance of traditional manual statistical analysis methods with the intelligent agent-based multi-channel customer analysis method described in this invention in four core indicators: accuracy of lead generation matching, lead tracking efficiency, conversion rate calculation error, and average system processing time. See below for reference. Figure 3 As shown in the experimental comparison chart, the method described in this invention improves the accuracy of lead generation matching by 28.6%, leads tracking efficiency by 42.3%, conversion rate calculation error by 67.9%, and average system processing time by 75.4%. It has significant technical advantages and practical value in scenarios such as multi-channel lead generation behavior analysis, lead attribution, and customer conversion evaluation.

[0045] It should be noted that the terms "comprising," "including," or any other variations thereof used herein are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0046] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0047] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. A multi-channel customer analysis method based on intelligent agents, characterized in that, The method includes: S1: Collect multi-source behavioral data from the entire customer acquisition process across multiple channels, standardize the multi-source behavioral data and perform cross-channel binding processing to obtain a standardized touchpoint sequence for shared entry points; S2: Calculate the single-group behavior heat value of the behavior data in the standardized touch point sequence of the shared entry point, divide the standardized touch point sequence of the shared entry point into multiple traffic diversion time windows, and calculate the window characteristics and heat characteristics of the traffic diversion time windows; S3: Extract lead generation records from the lead generation form submission data of the standardized touchpoint sequence of the shared entry point and generate lead generation features. Based on the lead generation features and the window features and popularity features of the traffic generation time window, calculate the pairing value between the lead generation record and the traffic generation time window. Sort the traffic generation time windows according to the pairing value to obtain the lead generation link candidate sorting table of the lead generation record. S4: Based on the candidate ranking table of lead generation links, perform value verification and customer conversion rate analysis on the lead generation time window under the entry point identifier.

2. The multi-channel customer analysis method based on intelligent agents as described in claim 1, characterized in that, The multi-source behavioral data refers to all recordable interactive behavioral data generated by users in the entire customer acquisition process, including short video pre-heating, live streaming browsing, store outreach, private domain acquisition, and lead generation conversion, across public, offline, and private channels. This behavioral data includes short video placement and interaction data, live streaming content display and pop-up exposure data, store poster outreach data, shared redirect entry access data, WeChat addition data, landing page browsing data, and lead generation form submission data. Each piece of behavioral data includes the time of occurrence, duration, channel type, content identifier, entry identifier, and action type.

3. The multi-channel customer analysis method based on intelligent agents as described in claim 2, characterized in that, Step S1, which standardizes and cross-channels the multi-source behavioral data, also includes: S11: Perform unified data format, coding standard, and null value pre-filling processing on the behavior occurrence time, channel type, content identifier, entry identifier, and action type in the multi-source behavior data to obtain standardized multi-source behavior data; S12: Using the entry identifier as the sole grouping criterion, filter behavioral data with the same entry identifier and bind them to form a set of behavioral data under that entry identifier; S13: Sort the behavior data in the behavior data set according to the order of the behavior occurrence time to obtain the standardized contact sequence of the shared entry point.

4. The multi-channel customer analysis method based on intelligent agents as described in claim 1, characterized in that, In step S2, the single-group behavior heat value of the behavior data in the standardized touchpoint sequence of the shared entry point is calculated, and the standardized touchpoint sequence of the shared entry point is divided into multiple traffic-driving time windows, including: S21: Extract the global timeline formed by the occurrence times of all behaviors in the standardized touchpoint sequence of the shared entry point, divide the global timeline into continuous time scales, and count all behavior data under each time scale; S22: Calculate the single-group behavior heat value of any set of behavior data under the time scale, and sum up the single-group behavior heat values ​​of all behavior data under the time scale to obtain the behavior heat value of all behavior data under the time scale. S23: Perform proximity heat correction processing on the behavior heat values ​​of all behavior data under the time scale to obtain the corrected behavior heat values ​​of the time scale; S24: Select the time scale in which the corrected behavior heat value is higher than the preset benchmark heat value as the effective traffic diversion interval, and use the behavior data sequence under the continuous effective traffic diversion interval as the traffic diversion time window in the shared entry standardized touch point sequence.

5. The multi-channel customer analysis method based on intelligent agents as described in claim 4, characterized in that, Step S2, which calculates the window characteristics and heat characteristics of the drainage time window, also includes: S25: Calculate the window length of the traffic acquisition time window, the behavior occurrence time corresponding to the two ends of the window, the proportion of behavior data of each channel type, the proportion of behavior data of each action type, and the average behavior duration of the behavior data. Then, concatenate the calculation results as the window features of the traffic acquisition time window. S26: Calculate the average corrected behavioral heat value of each time scale within the drainage time window, and use it as the heat feature of the drainage time window.

6. The multi-channel customer analysis method based on intelligent agents as described in claim 1, characterized in that, Step S3 involves extracting lead generation records and generating lead generation features from the lead generation form submission data of the standardized contact point sequence of the shared entry point, including: The lead collection record includes the time of the lead collection form submission, the duration of the submission, the entry label, and the user's intended needs tags. Based on the entry identifier label of the data submitted by the lead generation form, filter the preceding behavior data of the same user with the same entry identifier label, and calculate the total duration of the preceding behavior data. The preceding behavior data includes store poster reach data with the entry identifier label, behavior occurrence time earlier than the time of the data submitted by the lead generation form, and shared jump entry access data. The lead generation characteristics are obtained by concatenating the time of occurrence of the data submission on the lead generation form, the duration of the behavior, the entry identifier label, the sum of the duration of the preceding behavior data, and the intention demand label.

7. The multi-channel customer analysis method based on intelligent agents as described in claim 6, characterized in that, Step S3 calculates the pairing value between lead generation records and lead generation time windows, sorts the lead generation time windows according to the pairing value, and also includes: S31: Obtain the lead retention characteristics of the lead retention records, as well as the window characteristics and popularity characteristics of the lead generation time window; S32: Based on the time of occurrence of the data submission behavior in the data retention form in the data retention feature and the time of occurrence of the behavior corresponding to the two ends of the window in the window feature, calculate the temporal matching degree between the data retention record and the traffic generation time window; S33: Based on the entry identifier label and intention demand tag of the data submitted by the lead retention form in the lead retention feature, as well as the entry identifier and content identifier set of the behavioral data in the traffic acquisition time window, calculate the content matching degree between the lead retention record and the traffic acquisition time window; S34: Based on the sum of the behavior duration of the preceding behavior data in the lead retention features and the behavior duration of the lead retention form submission data, calculate the behavior matching degree between the lead retention record and the lead generation time window; S35: Based on the window characteristics and popularity characteristics of the traffic acquisition time window, calculate the popularity weighting coefficient of the traffic acquisition time window, and weight the temporal matching degree, content matching degree and behavioral matching degree between the lead generation record and the traffic acquisition time window to obtain the pairing value between the lead generation record and the traffic acquisition time window.

8. The multi-channel customer analysis method based on intelligent agents as described in claim 1, characterized in that, In step S4, based on the candidate ranking table for lead generation links, the value of the lead generation time window under the entry identifier and customer conversion rate analysis are performed, including: S41: For any entry identifier L, filter the lead generation time window under the entry identifier L, and based on the lead retention link candidate sorting table, obtain the pairing value between the lead generation time window and any lead retention record, and retain the lead generation time window whose pairing value is higher than the preset value verification threshold. S42: Calculate the average lead generation conversion rate for the retained lead generation time window and the average lead generation conversion rate for each channel type; S43: The average lead generation conversion rate and the average lead generation conversion rate of each channel are used as the results of multi-channel customer acquisition analysis for the shared lead generation entry associated with the entry identifier L.

9. A multi-channel customer analysis system based on intelligent agents as described in claim 1, characterized in that, The multi-channel customer analysis system includes a data acquisition agent, a sequence processing agent, a time period analysis agent, and a customer analysis agent to implement the technical steps of the agent-based multi-channel customer analysis method as described in any one of claims 1-8.

Images