A follow-up tracking data management method based on big data

Abstract

The application belongs to the technical field of analysis, application and management of revisit data, and specifically discloses a revisit tracking data management method based on big data, which comprises the following steps: based on user purchase attributes, clustering analysis is used to stratify users, and the priority of different users in revisit research is distinguished according to revisit application types, so as to ensure the adaptability and representativeness of users with different purchase attributes in various revisit research scenarios; by analyzing the historical research records of users in each user layer, the distribution effect of historical revisit methods is evaluated, and the revisit methods and revisit time of different users are differentially configured and targeted optimized accordingly, so as to maximize the response rate of users in the revisit process; by evaluating the structural data stability of different user layers, the encryption storage scheme of the user layer is dynamically optimized, so as to ensure the security and integrity of user data in the storage process, thereby meeting the technical requirements of data security protection.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of analysis, application and management of follow-up data, and relates to a follow-up tracking data management method based on big data. BACKGROUND

[0002] With the rapid development and wide application of big data technology, enterprises' demand for analyzing and mining user behavior data is increasing. In the field of user relationship management, follow-up tracking is an important means to improve user satisfaction and optimize product services. The scientificity and effectiveness of its data management method directly affect the operational efficiency and user stickiness of enterprises. However, the existing technology has obvious deficiencies in user stratification, follow-up strategy optimization and data security storage, and cannot meet the fine, intelligent and secure needs of enterprises for follow-up tracking data management. Therefore, it is of great practical significance and application value to develop a follow-up tracking data management method based on big data.

[0003] In the prior art, there are also some related solutions involving analysis, application and management of follow-up data. For example, a service evaluation method and system based on four-dimensional customer fit management is disclosed in Chinese Patent No. CN116739434A, which includes: constructing a customer service fit degree evaluation data collection channel, acquiring relevant data according to each collection channel, and supporting fit degree evaluation index calculation; combining customer service business scenarios, reasonably designing customer service follow-up questionnaires; developing customer service follow-up strategies; based on the collected customer service follow-up data, the unified acceptance message in the preset format is stored into the customer follow-up data center; constructing a customer satisfaction evaluation comprehensive index system from four dimensions of behavior, emotion, value and strategy. It can collect and analyze customer service information from multiple dimensions, construct a customer service data center, objectively evaluate customer service situation, and promote the continuous improvement and promotion of enterprise service work.

[0004] The above-mentioned scheme proposes some solutions for the analysis, application and management of follow-up data, but still has the following limitations: 1. The existing technology fails to finely stratify user purchase attributes according to different follow-up purposes of actual enterprise needs, resulting in low user response rate and insufficient resource utilization. Moreover, the analysis of user historical follow-up records is not deep enough, which cannot accurately evaluate the effect of different follow-up methods, and it is difficult to dynamically adjust the follow-up strategy, so that the follow-up method lacks pertinence, and the follow-up effect is unstable.

[0005] 2. When storing large-scale user data, the existing technology often uses fixed encryption storage scheme, which fails to dynamically optimize according to different data characteristics and stability, and there is a risk of data leakage or low storage efficiency. SUMMARY

[0006] In view of this, in order to solve the problems raised in the background art, a big data-based follow-up tracking data management method is proposed.

[0007] The object of the present application can be achieved by the following technical solutions: The present application provides a big data-based follow-up tracking data management method, comprising the following steps: in a follow-up application type setting state, the users are prioritized and layered according to historical purchase attributes by a clustering algorithm, and the follow-up research behaviors of different users corresponding to each user layer are obtained, wherein the follow-up application type includes product function improvement type, customer life cycle management type, and activity promotion management type.

[0008] The response rate of the follow-up research behaviors of different users corresponding to each user layer is verified to determine the distribution effect of the historical follow-up mode, and the distribution effect is divided into effective and ineffective.

[0009] In the case of effective historical follow-up mode, the response subject is marked, and the follow-up scheme is optimized.

[0010] In the case of ineffective historical follow-up mode, the loss subject is marked, and the dynamic optimization setting of the follow-up mode is performed.

[0011] The follow-up time is allocated based on the priority of the user layer where the response subject and the loss subject are located.

[0012] The current and historical follow-up data are collected, the time series characteristics of the follow-up data are evaluated by discrete degree, and the data structure stability of different user layers is determined.

[0013] The hierarchical encryption strategy is implemented according to the data structure stability of the user layer.

[0014] Compared with the prior art, the present application has the following advantages: (1) The present application is based on user purchase attributes, uses clustering analysis method to layer the users, and distinguishes the priority of different users in follow-up research according to follow-up application type, so as to ensure the adaptability and representativeness of users with different purchase attributes in various follow-up research scenes.

[0015] (2) The present application evaluates the distribution effect of the historical follow-up mode by analyzing the historical research records of the users in each user layer, and differentiates the configuration and targeted optimization of the follow-up mode and follow-up time of different users according to the distribution effect, so as to maximize the response rate of the users in the follow-up process.

[0016] (3) The present application evaluates the structural data stability of different user layers, dynamically optimizes the encryption storage scheme of the user layer, and ensures the security and integrity of the user data in the storage process, so as to meet the technical requirements of data security protection. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments. Obviously, the drawings described in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0018] Figure 1 The flowchart of the embodiment of the method of the present application is shown.

[0019] Figure 2 The schematic diagram of the priority layering of the user of the present application is shown.

[0020] Figure 3 The flowchart of the judgment of the response and non-response of the present application is shown. DETAILED DESCRIPTION

[0021] The technical solutions of the embodiments of the present application will be described clearly and completely in the following with reference to the drawings of the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments of the present application, all the other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0022] Please refer to Figure 1 、 2 The present application provides a big data-based follow-up tracking data management method, which comprises the following steps: Tp1, in a follow-up application type setting state, the users are priority-layered according to historical purchase attributes by a clustering algorithm, and the follow-up research behaviors of different users corresponding to each user layer are obtained from an enterprise follow-up tracking system, wherein the follow-up application types include product function improvement type, customer life cycle management type and activity promotion management type.

[0023] The historical purchase attributes include but are not limited to purchase records, consumption behaviors, promotion sensitivity and device usage of the users, and can be obtained by positioning the user login webpage through the enterprise follow-up tracking system.

[0024] In a preferred embodiment, the priority-layering of the users according to the historical purchase attributes by the clustering algorithm in the follow-up application type setting state comprises the following: B1, uploading the follow-up application types in the enterprise follow-up tracking system, and matching the follow-up application type corresponding follow-up research period and follow-up project weight distribution table through a system rule base.

[0025] The system rule base is used to store the follow-up application type corresponding follow-up research period, follow-up project weight distribution table, research interview theme and each question.

[0026] The revisit research cycle refers to the interval between the record time of the application record in the enterprise revisit tracking system and the current revisit time.

[0027] The revisit project weight distribution table is used to map the influence weight of different active behaviors on the application type of the revisit, and the active behaviors are composed of the purchase behavior, evaluation behavior, and historical participation in the revisit research behavior of the user, wherein the participation in the revisit research behavior includes the platform browsing record of the user in the research period, the research inquiry theme (such as the satisfaction survey theme, the effect evaluation theme, the improvement suggestion theme, and the problem follow-up theme), the reply voice record of the user in the historical research process, and each theme of the research inquiry theme corresponds to several different questions.

[0028] The product function improvement type is used to improve the functional experience and enhance the competitiveness of the product; the customer life cycle management type is used to improve the retention rate of the customer and increase the consumption amount and frequency of the customer; and the activity promotion management type is used to improve the activity effect, enhance the brand influence, or promote the user conversion (such as converting the activity participants into actual purchase users).

[0029] The user purchase record includes daily purchase and activity promotion purchase (such as the number of times of participating in the promotion activity and the proportion of the purchase promotion goods amount); and the evaluation record includes the evaluation text or complaint feedback text of the user on the purchased product and the communication text between the user and the background customer service.

[0030] B2, retrieve each user with active behavior in the revisit research cycle from the enterprise revisit tracking system, and mark as each active subject. For example, when a user has a purchase behavior or an evaluation behavior or a participation in the revisit research behavior in the revisit research cycle, the user is marked as an active subject.

[0031] B3, list the active behaviors of each active subject and substitute into a clustering algorithm to generate a plurality of revisit feature similar active subject clusters, taking the revisit feature similar active subject cluster as a same user layer, and each revisit feature similar active subject cluster has similar revisit features, for example, a plurality of users with the same purchase behavior or evaluation behavior are divided into the same user layer.

[0032] In a further preferred embodiment, listing the active behaviors of each active subject can be systematically constructed from the consumption behavior feature system, the text sentiment feature system, the feedback quality feature system, the promotion sensitivity feature system, and the behavior sequence feature system.

[0033] The consumption behavior feature system is used to map the revisit features of the user in the purchase frequency, consumption amount, and product category preference dimensions.

[0034] The text sentiment feature system is used to map the user's return visit characteristics in dimensions such as the positive / negative review tendency value, complaint frequency, and customer service interaction frequency.

[0035] The feedback quality feature system is used to map user return visit characteristics in dimensions such as survey participation, feasibility score of function improvement suggestions, and push response time difference.

[0036] The promotion sensitivity feature system is used to map the user's return visit characteristics in dimensions such as promotion activity response rate, consumption growth during the promotion period, and promotion channel preference (such as the length of time spent on the activity page).

[0037] The behavioral sequence feature system is used to map the return visit characteristics of users in dimensions such as the purchase decision chain (such as the average number of steps from browsing to purchasing, the conversion rate from the promotion page to the checkout page), and the proportion of devices used to access promotional activities (such as mobile devices vs. PC devices).

[0038] B4. Priority weights are determined for each active subject cluster based on similar return visit features and the return visit item weight distribution table. Active subject clusters are sorted in descending order by priority weight to obtain sorted user layers.

[0039] Specifically, a number of active behaviors and their influence weights that match a number of similar return visit features corresponding to each active subject cluster are screened out from the return visit item weight distribution table. According to the one-to-one mapping relationship between active behaviors and similar return visit features, the weights of all similar return visit features corresponding to each active subject cluster are accumulated to generate the priority weight of each active subject cluster.

[0040] The present invention stratifies users based on their purchasing attributes using a cluster analysis method, and differentiates the priorities of different users in a follow-up survey according to the type of follow-up application, thereby ensuring the adaptability and representativeness of users with different purchasing attributes in various follow-up survey scenarios.

[0041] Tp2. Verify the response rate of the return visit survey behavior of different users in each user layer to determine the distribution effect of the historical return visit method, and the distribution effect is divided into effective and ineffective.

[0042] In a preferred embodiment, the response rate of the follow-up survey behavior of different users in each user layer is verified to determine the effect of the historical follow-up method distribution, including: obtaining the follow-up method and follow-up survey results of each active subject in each user layer during the follow-up survey cycle from the follow-up survey behavior, and the follow-up survey results are divided into response and non-response.

[0043] The follow-up methods include email follow-up, SMS follow-up, telephone follow-up, online questionnaires (communication texts between users and back-end customer service), social media follow-up (users log in to the official website of each platform), etc.

[0044] According to the visiting mode of each record and the visiting research result, the response rate of each active subject to each visiting mode in each user layer is determined, for example, the record weight value of the record with the research result of response is set as 1, and the record weight value of the record with the research result of non-response is set as 0; all the records of each active subject in each user layer corresponding to different visiting modes are clustered, and the comprehensive weight value of each active subject in each user layer corresponding to each visiting mode is obtained by accumulating the record weight values of all the records; and the proportion of the comprehensive weight value compared with the preset reference weight value is obtained to obtain the response rate of each active subject in each user layer to each visiting mode.

[0045] By comparing the response rate with the preset standard response rate, the distribution effect of each active subject in each user layer corresponding to each visiting mode is determined, specifically, if the response rate of a certain active subject in a certain user layer to a certain visiting mode exceeds the preset standard response rate, it is judged that the distribution effect of the active subject in the user layer to the visiting mode is effective; if the response rate of the active subject in the user layer to each visiting mode does not exceed the preset standard response rate, it is judged that the distribution effect of the active subject in the user layer to each visiting mode is ineffective.

[0046] Please refer to Figure 3 In a further preferred embodiment, the judging mode of response and non-response includes: setting an access response rule according to the visiting mode of each record, the access response rule including a voice response rule and an information response rule, for example, for the telephone visiting mode, the access response rule is set as the voice response rule; for non-telephone visiting modes such as short message visiting and email visiting, the access response rule is set as the information response rule.

[0047] The voice response rule is to track the user's reply audio, when it is identified that there is voice reply content after a certain visiting telephone is dialed, such as identifying that there is reply content of the user to one or more questions of a certain visiting telephone, it is judged that the visiting research result of the user in the record is response; otherwise, it is judged that the visiting research result of the user in the record is non-response.

[0048] The information response rule is to set an access response period, for example, the access response period can be set as 3 hours, 10 hours, 24 hours, etc., when it is tracked that the user logs in the application platform to browse the related product within the access response period after a certain visiting information is published, it is judged that the visiting research result of the user in the record is response; otherwise, it is judged that the visiting research result of the user in the record is non-response.

[0049] Tp3, mark the response subject in the case of effective historical visiting mode, and optimize the visiting scheme.

[0050] In a preferred embodiment, the marking of the response subject in the case of effective historical revisit mode and the optimization of the revisit scheme include: counting the active subjects in each user layer that have one or more effective revisit modes, marking each active subject in each user layer as the corresponding response subject of each user layer, and selecting the item with the highest response rate in the one or more effective revisit modes as the current revisit mode of the corresponding response subject of each user layer. For example, when a telephone revisit and a short message revisit exist for an active subject in a user layer, the item with the highest response rate of the telephone revisit and the short message revisit of the active subject in the user layer is obtained as the current revisit mode of the active subject in the user layer.

[0051] The investigation and inquiry topics and the problems corresponding to the revisit application type are matched through the system rule base, and the first response rate of the problems corresponding to the investigation and inquiry topics belonging to the revisit application type in all users is identified through natural language processing (NLP) technology. Specifically, based on the above-mentioned judgment of response and non-response, the number of records of response and non-response of each problem in the corresponding revisit investigation behavior of all users is counted, and the first response rate of each problem corresponding to the investigation and inquiry topics belonging to the revisit application type in all users is obtained according to the formula: first response rate = number of records of response / (number of records of response + number of records of non-response).

[0052] The revisit investigation behavior of each response subject corresponding to each user layer is obtained, the problems belonging to the investigation and inquiry topics of each revisit are screened out, and the second response rate of each user layer corresponding to each response subject to the problems corresponding to the investigation and inquiry topics belonging to the revisit application type is determined. Specifically, based on the above-mentioned judgment of response and non-response, the number of comprehensive investigation records and the number of records of response of each user layer corresponding to each response subject to the problems corresponding to the investigation and inquiry topics belonging to the revisit application type are counted, and the second response rate of each user layer corresponding to each response subject to the problems corresponding to the investigation and inquiry topics belonging to the revisit application type is obtained according to the formula: second response rate = number of records of response / number of comprehensive investigation records.

[0053] The first response rate and the second response rate are accumulated to generate the overall response rate of each user layer corresponding to each response subject to the problems corresponding to the investigation and inquiry topics belonging to the revisit application type, and the problems with an overall response rate lower than a pre-set standard response rate are replaced.

[0054] Tp4, marking the lost subject in the case of ineffective historical revisit mode and performing dynamic optimization setting of the revisit mode.

[0055] In a preferred embodiment, the marking of the lost subject in the case of ineffective historical revisit mode and the dynamic optimization setting of the revisit mode include: marking each active subject in each user layer except the response subject as the corresponding lost subject of each user layer, and integrating the multiple revisit modes recorded by each lost subject corresponding to each user layer.

[0056] For each loss subject in each user layer that does not record a return visit mode, any one of one or more return visit modes not recorded is selected as the current return visit mode. For example, for a loss subject in a certain user layer that does not record a mail return visit mode and a social media return visit mode, one of the mail return visit mode and the social media return visit mode is randomly selected as the current return visit mode of the loss subject in the user layer.

[0057] For each loss subject in each user layer that records all return visit modes, a preferential policy is formulated according to the priority weight to promote user return visit response. For example, for a loss subject in a first user layer, multiple return visit modes are set in cooperation, and when setting the corresponding access response rules of these return visit modes, a certain amount of coupons are issued.

[0058] Tp5, return visit time allocation based on the priority of the user layer where the response subject and the loss subject are located.

[0059] In a preferred embodiment, the return visit time allocation based on the priority of the user layer where the response subject and the loss subject are located includes: performing time alignment processing on the historical purchase attributes of each active subject in each user layer in a unified time period to determine the response probability of each active subject in each user layer in each time period. Specifically, each working day is divided into different time stages, such as 2:00-5:00 and 5:00-8:00, etc. The comprehensive record number of each active subject in each user layer logging in the platform in different time stages is identified through historical purchase attributes, the proportion of the comprehensive record number compared to the pre-set reference comprehensive record number is obtained, and the response probability of each active subject in each user layer in each time period is obtained.

[0060] The time period with the highest response probability is selected as the return visit information publishing time period of each response subject and each loss subject in each user layer.

[0061] The current return visit mode of each response subject and each loss subject in each user layer is counted, and the subjects whose current return visit mode is telephone return visit among each response subject and each loss subject are marked as each inquiry subject in each user layer.

[0062] Different subjects in the same return visit information publishing time period are counted, and the different subjects in the same return visit information publishing time period are prioritized for telephone return visit according to the priority of the user layer, and then the different subjects in different return visit information publishing time periods are sequentially telephone return visited according to the telephone return visit priority order.

[0063] The application evaluates the allocation effect of historical follow-up methods by analyzing the historical investigation records of users in each user layer, and differentiates the follow-up methods and follow-up time of different users for targeted optimization to maximize the response rate of users in the follow-up process.

[0064] Tp6, collects current and historical follow-up data, evaluates the time series characteristics of the follow-up data through dispersion degree, and determines the data structure stability of different user layers.

[0065] The follow-up data refers to the data of the change of corresponding users in each user layer and the change of historical purchase attributes in different follow-up investigation behaviors.

[0066] In a preferred embodiment, the collection of current and historical follow-up data, the evaluation of the time series characteristics of the follow-up data through dispersion degree, and the determination of the data structure stability of different user layers include: using statistical analysis tools (such as pandas, NumPy library in Python, and R language environment, etc.) to arrange the follow-up data of each user layer into time series data set according to user identification and time dimension, and carrying out dispersion degree evaluation on the time series data set of each user layer, the time series data set is composed of multiple features of active behaviors recorded by users at different times, specifically, comparing the proportion of the corresponding comprehensive change amplitude of multiple features recorded by different user layers at different times compared with the preset reference amplitude, the greater the change amplitude, the greater the dispersion degree.

[0067] The change amplitude includes but is not limited to: for the same user, the corresponding change amplitude of the cumulative comprehensive purchase amount of different products; for different users, the cumulative comprehensive purchase amount of the two or more users is compared to generate the product comprehensive change amplitude between different users.

[0068] The user identification is, for example, identity name number, etc.

[0069] The stability level is divided according to different dispersion degree intervals, such as high stability for the dispersion degree in the first dispersion degree interval, medium stability for the dispersion degree in the second dispersion degree interval, and low stability for the dispersion degree in the third dispersion degree interval, matching the stability of the corresponding dispersion degree of the time series data set of each user layer, the stability is divided into high stability, medium stability and low stability, and the stability level is used to map the data structure relationship of the user layer.

[0070] Tp7, implementing hierarchical encryption strategy according to the data structure stability of the user layer.

[0071] In a preferred embodiment, the hierarchical encryption strategy is implemented according to the data structure stability of the user layer, specifically, lightweight encryption is used for high-stability user layers, and high-intensity encryption protection is implemented for low-stability user layers.

[0072] For example, for user layers with high stability, AES-128 encryption algorithm is used for encrypted storage. AES-128 encryption algorithm has high encryption efficiency, which can quickly complete encryption and decryption operations while ensuring data security.

[0073] For user layers with medium stability, AES-256 encryption algorithm with higher security is used. Although the encryption and decryption speed of this algorithm is slightly slower than that of AES-256 algorithm, it can provide stronger encryption strength to cope with possible security risks in the medium stability state of the data structure.

[0074] For user layers with low stability, in addition to using AES-256 encryption algorithm, an additional data confusion processing is added. Data confusion further increases the complexity of data before encryption by performing specific transformation and recombination on the original data, so as to improve the security of data.

[0075] The present application dynamically optimizes the encryption storage scheme of the user layer by evaluating the structure data stability of different user layers, ensures the security and integrity of user data in the storage process, and meets the technical requirements of data security protection.

[0076] The above content is only an example and description of the concept of the present application. Those skilled in the art can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, as long as they do not deviate from the concept of the present application or exceed the scope defined by the present application, which should belong to the protection scope of the present application.

Claims

1. A return visit tracking data management method based on big data, characterized in that: The steps include: When the return visit application type is set, users are prioritized and stratified according to their historical purchase attributes through a clustering algorithm to obtain the return visit research behavior of different users in each user layer. The return visit application types include product function improvement type, customer life cycle management type, and activity promotion management type. Verify the response rate of the return survey behavior of different users in each user layer to determine the distribution effect of the historical return method, and the distribution effect is divided into effective and ineffective; Mark the responding subject when the historical return visit method is effective, and optimize the return visit plan; The method of marking the responding subjects and optimizing the responding plan when the historical responding methods are effective includes: counting the active subjects in each user layer who have one or more effective responding methods, marking them as the responding subjects corresponding to each user layer, and selecting the one with the highest response rate among the one or more effective responding methods as the current responding method of the responding subjects corresponding to each user layer; matching the research inquiry topics and questions corresponding to the responding application type through the system rule library, and identifying the first-level response rate of all users for the questions belonging to the research inquiry topics corresponding to the responding application type through natural language processing technology; obtaining the responding research behavior of the responding subjects corresponding to each user layer, screening out the questions belonging to the research inquiry topics of each responding return, and determining the second-level response rate of the responding subjects corresponding to each user layer to the questions belonging to the research inquiry topics corresponding to the responding application type; accumulating the first-level response rate and the second-level response rate to generate the comprehensive response rate of the responding subjects corresponding to each user layer to the questions belonging to the research inquiry topics corresponding to the responding application type, and replacing the questions with a comprehensive response rate lower than the preset standard response rate; The method of marking the churn subjects and dynamically optimizing the revisit methods when the historical revisit methods are invalid includes: marking each active subject except each responding subject in each user layer as a churn subject corresponding to each user layer, integrating multiple revisit methods recorded for each churn subject in each user layer; for each churn subject in each user layer with no recorded revisit method, selecting one or more unrecorded revisit methods as the current revisit method; for each churn subject in each user layer with all revisit methods recorded, formulating preferential policies according to priority weights to promote user revisit responses; Mark the churned subjects when the historical revisit method is ineffective, and dynamically optimize the revisit method; Allocate return visit time based on the priority of the user layer where the responding subject and the churned subject are located; Collect current and historical return visit data, and evaluate the time series characteristics of the return visit data through the degree of dispersion to determine the stability of the data structure of different user layers; Implement a hierarchical encryption strategy based on the data structure stability of the user layer.

2. A return visit tracking data management method based on big data according to claim 1, characterized in that: In the state of setting the return application type, the user priority is stratified according to the historical purchase attributes through a clustering algorithm, including: uploading the return application type in the enterprise return tracking system, matching the return application type with the corresponding return survey cycle and return project weight distribution table through the system rule library; The return item weight allocation table is used to map the influence weights of different active behaviors on the return application types. The active behaviors are composed of the user's purchase behavior, evaluation behavior, and historical return survey participation behavior. The return survey participation behavior includes the user's platform reading records during the survey period, the survey inquiry topics, and the user's voice response records during the historical survey process. Retrieve users who have active behaviors during the return survey period from the enterprise return visit tracking system and mark them as active subjects; The active behaviors of each active subject are characterized and substituted into the clustering algorithm to generate multiple clusters of active subjects with similar return visit characteristics. The clusters of active subjects with similar return visit characteristics are considered to be in the same user layer, and each cluster of active subjects with similar return visit characteristics has similar return visit characteristics. According to the similar return visit characteristics and the return visit item weight distribution table, the priority weight of each active subject cluster is judged, and the active subject clusters are sorted in descending order according to the priority weight to obtain the sorted user layers.

3. A return visit tracking data management method based on big data according to claim 2, characterized in that: The characteristic listing of the active behaviors of each active subject includes systematic construction from the dimensions of consumption behavior characteristic system, text emotion characteristic system, feedback quality characteristic system, promotion sensitivity characteristic system and behavior sequence characteristic system.

4. The method for managing return visit tracking data based on big data according to claim 2, characterized in that: The response rate verification of the return survey behavior of different users in each user layer is performed to determine the effect of the historical return method distribution, including: obtaining the return method and return survey results of each active subject in each user layer during the return survey cycle from the return survey behavior, and the return survey results are divided into response and non-response; Determine the response rate of each active subject in each user layer to each return visit method based on the return visit method and return visit survey results of each record; By comparing the response rate with the preset standard response rate, the distribution effect of each return visit method for each active subject in each user layer is determined.

5. A return visit tracking data management method based on big data according to claim 4, characterized in that: The judgment method of whether there is a response or not includes: setting access response rules according to the return visit method of each record, and the access response rules include voice response rules and information response rules; The voice response rule is to track the user's reply audio. When it is recognized that the user has a voice reply content after a certain callback call is dialed, the user is judged to have responded to the callback survey result recorded at that time; otherwise, the user is judged to have not responded to the callback survey result recorded at that time; The information response rule is to set an access response period. When it is tracked that the user logs in to the application platform to browse related products within the access response period after a certain return visit information is released, it is judged that the user's return visit survey result in this record is responsive; otherwise, it is judged that the user's return visit survey result in this record is unresponsive.

6. A return visit tracking data management method based on big data according to claim 5, characterized in that: The revisit time allocation based on the user layer priority of the responding subject and the churned subject includes: using a unified time period to time-align the historical purchase attributes of each active subject corresponding to each user layer to determine the response probability of each active subject corresponding to each user layer in each time period; The time period with the highest response probability is selected as the return information release period for each responding subject and each churn subject corresponding to each user layer; Count the current return visit methods of each responding subject and each churn subject corresponding to each user layer, filter out the subjects whose current return visit method is telephone return visit among each responding subject and each churn subject, and mark them as the corresponding inquiry subjects of each user layer; Count the different subjects in the same follow-up information release period, prioritize the different subjects in the same follow-up information release period for telephone follow-up according to the user layer priority, and then conduct telephone follow-up on the different subjects in different follow-up information release periods in turn according to the telephone follow-up priority ranking order.

7. The method for managing return visit tracking data based on big data according to claim 3, characterized in that: The present invention collects current and historical return visit data and evaluates the time series characteristics of the return visit data by using the degree of dispersion to determine the data structure stability of different user layers. The content includes: using statistical analysis tools to organize the return visit data of each user layer into a time series data set according to user identification and time dimension, and performing a degree of dispersion evaluation on the time series data set of each user layer. The time series data set is composed of multiple features of the user's active behavior recorded at different times; The stability level is divided into different discrete degree intervals to match the stability of the corresponding discrete degree of the time series data set of each user layer. The stability is divided into high stability, medium stability, and low stability. The stability level is used to map the data structure relationship of the user layer.

8. The method for managing return visit tracking data based on big data according to claim 7, characterized in that: The hierarchical encryption strategy is implemented according to the stability of the data structure of the user layer, specifically: lightweight encryption is used for the user layer with high stability, and high-intensity encryption protection is implemented for the user layer with low stability.

Images