CN111861555A

CN111861555A - RFM-Session user modeling method, system and medium for behavior analysis

Info

Publication number: CN111861555A
Application number: CN202010681203.7A
Authority: CN
Inventors: 王翔; 叶凤; 张玉新; 周云龙
Original assignee: Shanghai Shijiu Information Technology Co ltd
Current assignee: Shanghai Shijiu Information Technology Co ltd
Priority date: 2020-07-15
Filing date: 2020-07-15
Publication date: 2020-10-30

Abstract

The invention provides an RFM-Session user modeling method, system and medium for behavior analysis, wherein behavior events of a user are collected at an OTT end in an event mode to form a behavior event set; selecting a starting event, a page browsing event and an application starting event representing a screen state to form a screen flow representing the screen state of a user; modeling the screen flow according to the definition of the OTT-Session, and calculating an index value; RFM-Session modeling is carried out on the users, and an RFM-Session vector related to each user is generated; and clustering after standardization to obtain user commonality and simultaneously obtain the relationship between the user value and the user behavior characteristics. The RFM-Session model applies the RFM model to the OTT scene, solves the problem that the value of the OTT equipment user is difficult to evaluate, and solves the problem that an operator cannot know the relationship between the value and the behavior by combining the RFM model and the Session model.

Description

RFM-Session user modeling method, system and medium for behavior analysis

Technical Field

The invention relates to the technical field of intelligent set-top boxes, in particular to an RFM-Session user modeling method, an RFM-Session user modeling system and a RFM-Session user modeling medium for behavior analysis, and particularly relates to an RFM-Session user modeling method for OTT equipment behavior analysis.

Background

An intelligent set-top box, also called a network set-top box or an OTT box (hereinafter referred to as a set-top box), generally refers to a set-top box loaded with an intelligent operating system such as Android, and a user can download and use various applications to obtain favorite contents. Millet box, tianmao magic box and the like which are popular in the market are all representative products, the volume is small and the price is low, a user can easily enjoy massive video contents on a television by only connecting a set top box with the television, and even an old television can be changed into a smart television. According to data display of an OvUygur cloud network OTT development prediction report in 2019, in 2018, the total number of activated set-top boxes reaches 3847 ten thousand, 1559 ten thousand daily active terminals, 2216 ten thousand weekly active terminals and 12606 ten thousand monthly active terminals. OTT user's behavior is stable, and the average is opened a machine 2 times daily, uses 4.8 hours, and stable activity and length of time of use make the STB become pay movie & TV and advertisement favoured terminal type.

Based on such magnitude of users and markets, research on user behaviors and user values of the users becomes very important, and operators of the set top box need to know differences of the users in use habits and attitudes of operation means of the users, so that products can be optimized, user liveness and stickiness are improved, and income in aspects of application distribution, advertisement and the like is increased.

Patent document CN109993582A discloses a multi-index customer segmentation method based on RFMCA model, which includes: acquiring network data and local data, constructing sample data, and preprocessing the sample data; analyzing the preprocessed data, and constructing a multi-index client subdivision model based on the RFMCA model; evaluating the result of the multi-index customer segmentation model to obtain segmented data, and comparing and analyzing the segmented data with the traditional segmentation indexes; and mining the intra-class association rule of the subdivided data. A complete customer segment index system suitable for retail industry is formed, and the multi-index customer segment model is high in accuracy and small in time complexity. The method is obviously superior to the traditional RFM in the aspects of distinguishing the behavior characteristics of the customers and subdividing the customers, can better distinguish different types of customers for enterprises, and makes a differentiated marketing strategy, so that the enterprises reasonably utilize limited resources to improve the satisfaction degree and loyalty degree of the customers, and the value of the enterprises is improved. But it is not suitable for application scenarios of intelligent set-top boxes.

Disclosure of Invention

Aiming at the defects in the prior art, the invention aims to provide an RFM-Session user modeling method, an RFM-Session user modeling system and an RFM-Session user modeling medium for behavior analysis.

The RFM-Session user modeling method for OTT equipment behavior analysis provided by the invention comprises the following steps:

step S1: collecting behavior events of a user at a set-top box end in an event mode, and storing the behavior events in a database to form a behavior event set;

step S2: selecting a starting event, a page browsing event and an application starting event representing a screen state from the behavior event set to form a screen flow representing a user screen state;

step S3: modeling the screen flow according to the definition of the OTT-Session, and calculating an index value;

step S4: performing RFM-Session modeling on users, giving six model indexes for each user, and generating an RFM-Session vector related to each user after modeling, wherein the six model indexes can represent the RFM-Session vector;

step S5: Z-Score standardization is carried out on model indexes in RFM-Session vectors of all users, then K-Means clustering is carried out, the users are divided into K classes, K is a positive integer, all the users in each class of the K classes have commonality in value, the commonality of the users in the class is obtained by calculating the mean value of the RFM-Session six model indexes in each class, and meanwhile, the relation between the user value and the user behavior characteristics is obtained.

Preferably, in the modeling according to the definition of OTT-Session, any one or more of a Session start event, a Session end event, a Session duration, a Session depth, and a Session achievement is defined to form and model the data structure.

Preferably, the six model indexes include an RFM index and a session index; wherein R in the RFM index is the time of the last session from the expiration date, F is the number of sessions in a period of time, and M is the time length of the user using the OTT equipment; the duration of the session index is the average session duration of the user within a period of time, the depth is the average session depth of the user within a period of time, and the achievement is the session achievement rate of the user within a period of time.

Preferably, the users are divided into 5 classes, namely high-value users, maintenance users, development users, saving users and loss users.

The invention provides an RFM-Session user modeling system for OTT equipment behavior analysis, which comprises:

module S1: collecting behavior events of a user at a set-top box end in an event mode, and storing the behavior events in a database to form a behavior event set;

module S2: selecting a starting event, a page browsing event and an application starting event representing a screen state from the behavior event set to form a screen flow representing a user screen state;

Module S3: modeling the screen flow according to the definition of the OTT-Session, and calculating an index value;

module S4: performing RFM-Session modeling on users, giving six model indexes for each user, and generating an RFM-Session vector related to each user after modeling, wherein the six model indexes can represent the RFM-Session vector;

module S5: Z-Score standardization is carried out on model indexes in RFM-Session vectors of all users, then K-Means clustering is carried out, the users are divided into K classes, K is a positive integer, all the users in each class of the K classes have commonality in value, the commonality of the users in the class is obtained by calculating the mean value of the RFM-Session six model indexes in each class, and meanwhile, the relation between the user value and the user behavior characteristics is obtained.

According to the present invention, a computer-readable storage medium is provided, in which a computer program is stored, which, when being executed by a processor, carries out the steps of the method as set forth in the claims.

Compared with the prior art, the invention has the following beneficial effects:

1. through the proposed OTT-Session modeling mode, a method for judging the achievement of the purpose through the property of an end event is added into a Session analysis method, and the technical effect of quickly judging whether the purpose is achieved after a series of operations of a user is achieved;

2. Through the proposed RFM-Session modeling mode, six indexes used by a user are described, and clustering is performed through a K-Means clustering method, so that the effects of evaluating OTT user value and classifying according to the value are achieved;

3. by comparing the centroid values of the RFM-Session index clustering results, the effect of knowing the reason of low user value and giving different operation means to users with different values is achieved.

Drawings

Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments with reference to the following drawings:

FIG. 1 is a schematic process flow diagram of the present invention;

FIG. 2 is a data flow diagram of the present invention;

FIG. 3 is a diagram illustrating a clustering result according to the present invention.

Detailed Description

The present invention will be described in detail with reference to specific examples. The following examples will assist those skilled in the art in further understanding the invention, but are not intended to limit the invention in any way. It should be noted that it would be obvious to those skilled in the art that various changes and modifications can be made without departing from the spirit of the invention. All falling within the scope of the present invention.

Example 1

The invention solves the problem that the user value of the OTT equipment is difficult to evaluate by applying the RFM model to the OTT scene, and solves the problem that an operator cannot know the relation between the value and the behavior by combining the RFM model and the Session model. The OTT-Session model based on the purpose is combined with the RFM-Session model which reflects the relationship between the user behavior and the value, and the two models are suitable for OTT equipment such as an intelligent television and an intelligent set top box. The method comprises the steps of performing simple extension splicing on an RFM model and the OTT-Session model, changing the three indexes of the RFM into six indexes of the RFM-Session, and reflecting the relation between behaviors and values.

As shown in fig. 1, the method of the present invention comprises the steps of:

step 1: and collecting the behavior events of the user by an event mode at the set-top box end, and storing the behavior events in a database.

An event model is a modeling method for data collection, which abstracts a behavior of a user into an event, such as a boot event, a page browsing event, an application start event, a click event, and the like.

Step 2: of the collected events, three events representing the screen state are selected: a startup event, a page browsing event, and an application starting event (the startup event represents the user screen opening, the page browsing event represents the user screen state as a page inside the operating system, and the application starting event represents the screen state as a page inside the application). These three events may constitute a "screen flow" representing the state of the user's screen.

And step 3: and modeling the screen flow according to the definition of the OTT-Session, and calculating an index value. The relevant definitions are as follows:

the session is defined as: if a user in a continuously generated event sequence E ═ { E1, … eN, … eN }, the time interval Δ t between every two events is less than the time μ, then E is defined as a session.

session start event (start _ session): if the event sequence E ═ { E1, … eN, … eN } constitutes a session, then event E1 is the start event of this session.

session end event (end _ session): if the event sequence E ═ { E1, … eN, … eN } constitutes a session, the event eN is the end event of this session.

session duration (duration): if the event sequence E ═ { E1, … eN, … eN } constitutes a session, the time difference tN-t1 between the session end event and the session start event is the session duration.

session depth (depth): if the event sequence E ═ { E1, … eN, … eN } constitutes a session, then the total number of events N in E is the session depth.

session achievement (achievement): if the session ending event is an event representing that the user's purpose is completed, it is called session achievement, the ending event is called achievement event, the session duration is called achievement duration, the session depth is called achievement depth, otherwise, it is called session non-achievement. The ratio of the number of session achievements to the total number of session is the session achievement rate.

According to the definition, the data structure after session modeling is shown as table one.

Data structure after table session modeling

In the related definition of session achievement, an achievement event representing that a user accomplishes the purpose needs to be determined, a starting event app _ startup is taken as the session achievement event, which represents that the user starts a certain application and continuously watches for a period of time to achieve the purpose that the user uses the set top box for entertainment, and the other events represent that the event is not achieved. The achievement rate of session is very helpful for judging the satisfaction degree of the user, and if the achievement rate is low, the user is not interested in the content of the set top box, so that loss is easily caused. The session duration and depth are also important, the too long duration and depth represent that the user is in a process of not knowing what to see or the design of the set-top box system is complicated, so that the user is difficult to find the content, while the too short duration and depth represent that the user can easily achieve the purpose, but from the perspective of an operator, the too short duration and depth represent the solidification of the user behavior pattern, so that the development of other interests of the user is not facilitated.

And 4, step 4: RFM-Session modeling is carried out on users, and the modeling method gives six indexes (R, F, M, duration, depth, achievent) for each user. The six indexes comprise an RFM index and a session index: RFM represents value, R is the time of the last session from the expiration date, F is the number of sessions in a period of time, and M is the duration of the OTT equipment used by the user, such as the application use duration in a period of time. The Session model represents behaviors, duration is the average Session duration of the user within a period of time, depth is the average Session depth of the user within a period of time, achievement is the Session achievement rate of the user within a period of time, and the calculation method is (number of achieved sessions)/(total number of sessions). After modeling, the RFM-Session vector x for each user will be produced (R, F, M, duration, depth, achievement)

And 5: Z-Score standardization is carried out on R, F and M values in RFM-Session vectors of all users, K-Means clustering is carried out, the users are divided into K classes, all the users in each class of the K classes have certain commonality in value, the commonality of the users can be obtained by solving the mean value of the RFM-Session six indexes in each class, and the relationship between the user value and the user behavior characteristics can also be obtained.

Specifically, the RFM vector r of each user over a period of time is determined_vCalculating (R, F and M), and then performing K-Means clustering to obtain K clusters, wherein the data in the clusters have certain similarity in RFM value, and the mass center of the clusters can represent the RFM value of users in the clusters.

The Z-Score is a common data standardization method, data with different magnitude levels are uniformly converted into the same magnitude level, the magnitude levels among the data can be unified by measuring the calculated Z-Score value, and comparability is guaranteed. It is calculated by the formula

At publicWhere μ is the mean value of the data, σ is the standard deviation of the data, and the normalized vector is denoted as r_v ^*＝(R^*，F^*，M^*)。

The present invention also provides a computer readable storage medium having stored thereon a computer program, which when executed by a processor, performs the steps of the RFM-Session user modeling method for behavioral analysis.

Example 2

Embodiment 2 can be regarded as a preferable example of embodiment 1. The system for RFM-Session user modeling for behavioral analysis described in embodiment 2 utilizes the steps of the RFM-Session user modeling method for behavioral analysis described in embodiment 1.

Module S1: collecting behavior events of a user at an OTT end in an event mode, and storing the behavior events in a database to form a behavior event set; the OTT end comprises a set top box and a television.

In the modeling according to the definition of the OTT-Session, any one or more of a Session start event, a Session end event, a Session duration, a Session depth and a Session achievement are defined to form and model the data structure.

The six model indexes comprise an RFM index and a session index; wherein R in the RFM index is the time of the last session from the expiration date, F is the number of sessions in a period of time, and M is the time length of the user using the OTT equipment; the duration of the session index is the average session duration of the user within a period of time, the depth is the average session depth of the user within a period of time, and the achievement is the session achievement rate of the user within a period of time.

The user is divided into K classes and 5 classes, namely a high-value user, a maintenance user, a development user, a saving user and a loss user. And analyzing the common differences of the five types of users in the behaviors by combining the session indexes, thereby giving operation strategy suggestions for the five types of users.

High value users belong to heavily dependent users of set-top boxes, who have plenty of time to use them compared to others. From the achievement rate, the average achievement rate index of the users is the highest of the five types of users, and the satisfaction degree of the users on the content of the set top box is higher. Therefore, more operations are not needed for the part of users, the current situation is kept, and the user can pay attention to the retention.

The maintenance users belong to general value users who keep relatively high duration and frequency. From the viewpoint of the duration, depth and achievement rate of sessions, they are very close to high-value users. They are also more satisfied with the set-top box, but may be less than the high value user in the time available for distribution to the television, such as a user whose habitual time pattern is evening-type, and therefore maintain the existing strategy for this portion of users as well.

The developing users belong to low-value users, and the frequency and the duration are low. From the viewpoint of the achievement rate, the achievement rate of the part of users is relatively low, the satisfaction degree of the content of the set top box is not high, and from the viewpoint of the duration and the depth, the part of users has simpler behavior in the set top box. Therefore, for the part of users, the interest needs to be mined, and the activity and the satisfaction degree of the part of users are improved through an excellent recommendation strategy. Meanwhile, the part of users occupies the most of all people, the lifting space is the largest, and more emphasis should be placed on the whole resource.

The saving users also belong to low-value users, and the difference between them and the developing users is that their R value is higher, and the set-top box is not used for a period of time. From the perspective of session indexes, the part of users is similar to the developing users, and operation strategies similar to the developing users can be applied. Meanwhile, because the users have high loss risk, special operation strategies need to be performed, for example, when the users log in next time, a certain time of movie and television VIP can be given to stimulate the users to use, and the purpose of saving is achieved.

The most important characteristic of the lost users is that the R value is large, and in the data set of the last month, the lost users have not used the set top box for 24 days on average, and the loss probability is very large. It is found that the session duration of the user is relatively long, and the achievement rate is relatively low, which indicates that the user has little interest in the content and has low satisfaction. The strategy of this part of users is the same as that of the retained users, and it is assumed that they can log in next time and should be stimulated by some to achieve the aim of recall.

By applying the RFM model to the OTT scene, the problem that the value of the OTT equipment user is difficult to evaluate is solved, and the problem that an operator cannot know the relation between the value and the behavior is solved by combining the RFM model and the Session model. The index definition of OTT-Session is used to replace the definition of RFM three indexes in the traditional RFM model, so that the RFM can be applied to the OTT field, and the traditional RFM is originally applied to a scene similar to shopping. The RFM model and the OTT-Session model are simply expanded and spliced, three indexes of the RFM are changed into six indexes of the RFM-Session, and the method has the advantage of reflecting the relation between behaviors and values.

Those skilled in the art will appreciate that, in addition to implementing the systems, apparatus, and various modules thereof provided by the present invention in purely computer readable program code, the same procedures can be implemented entirely by logically programming method steps such that the systems, apparatus, and various modules thereof are provided in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Therefore, the system, the device and the modules thereof provided by the present invention can be considered as a hardware component, and the modules included in the system, the device and the modules thereof for implementing various programs can also be considered as structures in the hardware component; modules for performing various functions may also be considered to be both software programs for performing the methods and structures within hardware components.

The foregoing description of specific embodiments of the present invention has been presented. It is to be understood that the present invention is not limited to the specific embodiments described above, and that various changes or modifications may be made by one skilled in the art within the scope of the appended claims without departing from the spirit of the invention. The embodiments and features of the embodiments of the present application may be combined with each other arbitrarily without conflict.

Claims

1. An RFM-Session user modeling method for behavioral analysis, comprising:

step S1: collecting behavior events of a user at an OTT end in an event mode, and storing the behavior events in a database to form a behavior event set;

step S3: modeling the screen flow according to the definition of the OTT-Session model, and calculating an index value;

2. The RFM-Session user modeling method for behavior analysis according to claim 1, wherein in the modeling according to the OTT-Session definition, any one or more of a Session start event, a Session end event, a Session duration, a Session depth, and a Session achievement are defined, and a modeled data structure is formed.

3. The RFM-Session user modeling method for behavior analysis as defined in claim 1, wherein the six model indicators include an RFM indicator and a Session indicator; wherein R in the RFM index is the time of the last session from the expiration date, F is the number of sessions in a period of time, and M is the time length of the user using the OTT equipment; the duration of the session index is the average session duration of the user within a period of time, the depth is the average session depth of the user within a period of time, and the achievement is the session achievement rate of the user within a period of time.

4. The RFM-Session user modeling method for behavioral analysis according to claim 1, wherein said classifying users into K classes is 5 classes, respectively high value users, maintenance users, development users, stay users, and attrition users.

5. An RFM-Session user modeling system for behavioral analysis, comprising:

module S3: modeling the screen flow according to the definition of the OTT-Session model, and calculating an index value;

6. The RFM-Session user modeling system for behavior analysis as defined in claim 5, wherein in the modeling according to the OTT-Session definition, any one or more of a Session start event, a Session end event, a Session duration, a Session depth, and a Session achievement are defined, and a modeled data structure is formed.

7. The RFM-Session user modeling system for behavioral analysis according to claim 5, wherein the six model indicators include an RFM indicator and a Session indicator; wherein R in the RFM index is the time of the last session from the expiration date, F is the number of sessions in a period of time, and M is the time length of the user using the OTT equipment; the duration of the session index is the average session duration of the user within a period of time, the depth is the average session depth of the user within a period of time, and the achievement is the session achievement rate of the user within a period of time.

8. The RFM-Session user modeling system for behavioral analysis according to claim 1, wherein said classifying users into K classes is 5 classes, respectively high value users, maintenance users, development users, stay users, and attrition users.

9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method of any one of claims 1 to 4.