CN-121724670-B - Marketing intervention method and system based on customer loss prediction

CN121724670BCN 121724670 BCN121724670 BCN 121724670BCN-121724670-B

Abstract

The invention relates to the technical field of loss prediction and discloses a marketing intervention method and a marketing intervention system based on client loss prediction, wherein the method comprises the steps of carrying out abnormal fluctuation detection on interactive behavior data of clients to obtain loss symptom data; the method comprises the steps of carrying out state evolution analysis on flow loss symptom data to obtain an active decay characteristic, carrying out time sequence track fitting on the flow loss symptom data to obtain an active decay track graph, carrying out phase matching on the active decay track graph and a life cycle of a client, merging static attribute data of the client to obtain a loss risk multidimensional portrait, carrying out decision factor analysis on loss risk driving factors of the client, carrying out risk comprehensive research on the analyzed root cause characteristics to obtain a loss risk level, mapping the root cause characteristics and the loss risk level to a preset strategy intervention library, and combining the static attribute data to generate a personalized intervention strategy.

Inventors

WANG GANG
YANG XI
WANG XIANLI
ZHANG JUN
YANG XINYI

Assignees

贵州商学院

Dates

Publication Date: 20260508
Application Date: 20260212

Claims (6)

1.A method of marketing intervention based on customer churn prediction, the method comprising: s1, carrying out abnormal fluctuation detection on interactive behavior data of a client to obtain loss symptom data of the client; S2, carrying out state evolution analysis on the loss symptom data to obtain activity decay characteristics of the loss symptom data, wherein the method comprises the following steps: Heterogeneous feature extraction is carried out on the loss symptom data to obtain frequency features, interval features and channel diversity features of the loss symptom data; according to the time sequence, multidimensional sequencing is carried out on the frequency characteristic, the interval characteristic and the channel diversity characteristic to obtain a characteristic time sequence of the client; analyzing the change situation of the characteristic time sequence to obtain the frequency fading rate, interval fading rate and channel fading rate of the characteristic time sequence; Integrating the frequency decay rate, the interval decay rate, and the channel decay rate into active decay features of the loss symptom data; S3, based on the activity decay characteristics, performing time sequence track fitting on the loss symptom data to obtain an activity decay track graph of the client, wherein the method comprises the following steps: taking time as a horizontal axis and taking a multidimensional feature vector of the activity decay feature as a vertical axis to construct a multidimensional coordinate system of the client; mapping the activity decay feature to the multidimensional coordinate system to obtain a discrete time sequence state point of the activity decay feature; nonlinear fitting is carried out on the discrete time sequence state points to obtain a continuous fading track curve of the discrete time sequence state points; Performing smoothness optimization on the continuous fading track curve, and performing key information annotation on the optimized continuous fading track curve to obtain an activity fading track diagram of the client; s4, performing phase matching on the activity decay trajectory graph and the life cycle of the client, and fusing static attribute data of the client to obtain a loss risk multidimensional image of the client; S5, based on the loss risk multidimensional image, resolving decision factors of loss risk driving factors of the clients, and comprehensively studying and judging risks of resolved root cause features to obtain loss risk grades of the clients, wherein the method comprises the following steps: based on a preset decision factor weight rule base, carrying out importance analysis on the multidimensional features of the loss risk multidimensional image to obtain loss risk contribution of the multidimensional features; according to the loss risk contribution degree, conducting dominant factor analysis on loss risk driving factors of the clients to obtain root factor characteristics of the clients, wherein the loss risk driving factors are extracted from the loss risk multidimensional portrait and comprise client static attribute characteristics and client activity decay dynamic characteristics; carrying out risk quantification evaluation on the root cause characteristics to obtain loss risk degree of the root cause characteristics; performing association matching on the loss risk degree and a preset risk level rule base to obtain the loss risk level of the client; S6, mapping the root cause characteristics and the loss risk level to a preset strategy intervention library, and generating a personalized intervention strategy of the client by combining the static attribute data, wherein the personalized intervention strategy comprises the following steps: Performing logic association analysis on a preset strategy intervention library to obtain an intervention matching rule of the strategy intervention library; Based on the intervention matching rule, carrying out association mapping on the root cause characteristics and the loss risk level to obtain a basic intervention strategy of the client; And correcting key parameters of the basic intervention strategy according to the client grade and industry attribute in the static attribute data to obtain the personalized intervention strategy of the client.
2. The marketing intervention method based on customer churn prediction according to claim 1, wherein the performing abnormal fluctuation detection on the interactive behavior data of the customer to obtain churn symptom data of the customer comprises: acquiring historical behavior data and interactive behavior data of a client; carrying out normal behavior analysis on the historical behavior data to obtain a historical statistical baseline of the client; Based on the historical statistical baseline, performing difference comparison analysis on the interactive behavior data to obtain behavior deviation indexes of the interactive behavior data; And removing conventional behavior data in the interactive behavior data based on the behavior deviation index to obtain loss symptom data of the client.
3. The method of claim 1, wherein said stage matching the activity decay trajectory graph to the lifecycle of the customer comprises: dividing the life cycle of the client into an introduction period, a growth period, a maturity period and a decay period according to a preset enterprise client cycle relationship; analyzing the change trend of the activity decay trajectory graph, and identifying key inflection points in the activity decay trajectory graph; and based on the occurrence time and the change amplitude characteristics of the key inflection points, carrying out characteristic association comparison on the activity decay trajectory graph and the introduction period, the growth period, the maturation period and the decay period, and confirming the current life cycle stage of the client.
4. The method for marketing intervention based on customer churn prediction according to claim 3, wherein said fusing said customer's static attribute data to obtain said customer's churn risk multidimensional image comprises: acquiring industry attribute, collaboration age, client grade and historical value contribution data of the client; Performing feature tensor synthesis on the industry attribute, the cooperation period, the client level and the historical value contribution data to obtain a static attribute feature vector of the client; carrying out multi-source feature fusion on the static attribute feature vector and the stage feature vector of the current life cycle stage to obtain a fusion feature vector of the client; And reconstructing the multidimensional view of the activity decay trajectory graph based on the fusion feature vector to obtain the loss risk multidimensional image of the client.
5. The method of claim 1, wherein the risk of churn is calculated as follows: ; in the formula, Represent the first Risk of loss of individual root cause features, Represent the first The risk contribution of loss of individual root cause features, Representing the total number of root cause features, Representing a natural logarithmic function, The open square operation is represented as such, Representing a summation operation.
6. A customer churn prediction based marketing intervention system for implementing the customer churn prediction based marketing intervention method of claim 1, the system comprising: the abnormal fluctuation detection module is used for carrying out abnormal fluctuation detection on the interactive behavior data of the client to obtain loss symptom data of the client; The state evolution analysis module is used for carrying out state evolution analysis on the loss symptom data to obtain the activity decay characteristics of the loss symptom data, and comprises the following steps: Heterogeneous feature extraction is carried out on the loss symptom data to obtain frequency features, interval features and channel diversity features of the loss symptom data; according to the time sequence, multidimensional sequencing is carried out on the frequency characteristic, the interval characteristic and the channel diversity characteristic to obtain a characteristic time sequence of the client; analyzing the change situation of the characteristic time sequence to obtain the frequency fading rate, interval fading rate and channel fading rate of the characteristic time sequence; Integrating the frequency decay rate, the interval decay rate, and the channel decay rate into active decay features of the loss symptom data; The time sequence track fitting module is used for performing time sequence track fitting on the loss symptom data based on the activity decay characteristics to obtain an activity decay track graph of the client, and comprises the following steps: taking time as a horizontal axis and taking a multidimensional feature vector of the activity decay feature as a vertical axis to construct a multidimensional coordinate system of the client; mapping the activity decay feature to the multidimensional coordinate system to obtain a discrete time sequence state point of the activity decay feature; nonlinear fitting is carried out on the discrete time sequence state points to obtain a continuous fading track curve of the discrete time sequence state points; Performing smoothness optimization on the continuous fading track curve, and performing key information annotation on the optimized continuous fading track curve to obtain an activity fading track diagram of the client; the risk portrait construction module is used for carrying out phase matching on the activity decay trajectory graph and the life cycle of the client, and fusing static attribute data of the client to obtain a loss risk multidimensional image of the client; The risk research and judgment and determination module is used for analyzing decision factors of the loss risk driving factors of the clients based on the loss risk multidimensional image, and comprehensively researching and judging the risks of the analyzed root cause characteristics to obtain the loss risk grade of the clients, and comprises the following steps: based on a preset decision factor weight rule base, carrying out importance analysis on the multidimensional features of the loss risk multidimensional image to obtain loss risk contribution of the multidimensional features; according to the loss risk contribution degree, conducting dominant factor analysis on loss risk driving factors of the clients to obtain root factor characteristics of the clients, wherein the loss risk driving factors are extracted from the loss risk multidimensional portrait and comprise client static attribute characteristics and client activity decay dynamic characteristics; carrying out risk quantification evaluation on the root cause characteristics to obtain loss risk degree of the root cause characteristics; performing association matching on the loss risk degree and a preset risk level rule base to obtain the loss risk level of the client; The intervention policy generation module is configured to map the root cause feature and the loss risk level to a preset policy intervention library, and combine the static attribute data to generate a personalized intervention policy of the client, where the personalized intervention policy comprises: Performing logic association analysis on a preset strategy intervention library to obtain an intervention matching rule of the strategy intervention library; Based on the intervention matching rule, carrying out association mapping on the root cause characteristics and the loss risk level to obtain a basic intervention strategy of the client; And correcting key parameters of the basic intervention strategy according to the client grade and industry attribute in the static attribute data to obtain the personalized intervention strategy of the client.

Description

Marketing intervention method and system based on customer loss prediction Technical Field The invention relates to the technical field of loss prediction, in particular to a marketing intervention method and a marketing intervention system based on customer loss prediction. Background The existing customer loss prediction related technology has limitation on the data processing level, lacks a systematic method for detecting abnormal fluctuation of customer interaction behavior data, fails to construct an accurate normal baseline through historical behavior data, and is difficult to effectively reject conventional behavior data and accurately extract loss symptom information. Meanwhile, state evolution analysis of flow loss related data is not deep enough, and only the simple extraction of surface features is stopped, so that the fading situation of multidimensional features such as frequency, interval, channel and the like cannot be comprehensively captured, so that the characterization of the active fading features is not comprehensive enough, and the accuracy of subsequent prediction is affected. In the link of risk assessment and intervention strategy generation, the prior art fails to effectively correlate and match a customer activity decay track with a life cycle stage, and fusion of static attribute data and dynamic decay characteristics is insufficient when a loss risk portrait is constructed, so that risk level research and judgment lacks scientificity and accuracy. In addition, most of the existing intervention strategies are designed in a universal way, targeted adaptation is not performed based on the loss root characteristics and the personalized attributes of the clients, so that the targeting of marketing intervention is insufficient, and the loss of the clients is difficult to effectively inhibit. Therefore, how to improve the accuracy of customer loss prediction and the pertinence and effectiveness of marketing intervention becomes a problem to be solved urgently. Disclosure of Invention The invention provides a marketing intervention method and a marketing intervention system based on customer loss prediction, which are used for solving the problems in the background technology. In order to achieve the above object, the present invention provides a marketing intervention method based on customer churn prediction, comprising: s1, carrying out abnormal fluctuation detection on interactive behavior data of a client to obtain loss symptom data of the client; S2, carrying out state evolution analysis on the loss symptom data to obtain activity decay characteristics of the loss symptom data; S3, based on the activity decay characteristics, performing time sequence track fitting on the loss symptom data to obtain an activity decay track graph of the client; s4, performing phase matching on the activity decay trajectory graph and the life cycle of the client, and fusing static attribute data of the client to obtain a loss risk multidimensional image of the client; S5, analyzing decision factors of the loss risk driving factors of the clients based on the loss risk multidimensional image, and comprehensively researching and judging risks of the analyzed root cause characteristics to obtain loss risk grades of the clients; s6, mapping the root cause characteristics and the loss risk level to a preset strategy intervention library, and generating a personalized intervention strategy of the client by combining the static attribute data. In a preferred embodiment, the detecting abnormal fluctuation of the interactive behavior data of the client to obtain the loss symptom data of the client includes: acquiring historical behavior data and interactive behavior data of a client; carrying out normal behavior analysis on the historical behavior data to obtain a historical statistical baseline of the client; Based on the historical statistical baseline, performing difference comparison analysis on the interactive behavior data to obtain behavior deviation indexes of the interactive behavior data; And removing conventional behavior data in the interactive behavior data based on the behavior deviation index to obtain loss symptom data of the client. In a preferred embodiment, the performing state evolution analysis on the loss symptom data to obtain an activity decay characteristic of the loss symptom data includes: Heterogeneous feature extraction is carried out on the loss symptom data to obtain frequency features, interval features and channel diversity features of the loss symptom data; according to the time sequence, multidimensional sequencing is carried out on the frequency characteristic, the interval characteristic and the channel diversity characteristic to obtain a characteristic time sequence of the client; analyzing the change situation of the characteristic time sequence to obtain the frequency fading rate, interval fading rate and channel fading rate of the characteristic time sequence; integra