CN106127663A - A kind of communication user consumption feature attribute reduction matching extracting method and data system - Google Patents

Abstract

The present invention proposes a kind of communication user consumption feature attribute reduction matching extracting method and data system, including: step 1, gather data: gather user according to common carrier real-world operation situation and use set meal consumption data, one user of communication consumption data genaration of the continuous m of user according to common carrier collection month uses the table of set meal consumption, comprises A in table₁, A₂... A_mAttribute, these attributes are m the numerical attribute that generating principle is identical, A_mRepresent the communication consumption attribute of user's m-th moon；The data of collection are carried out discretization and yojan between attribute by step 2 so that it is dimensionality reduction, with reflection telex network consumption situation of change that can be simple and clear, supports Information System configuration, it is achieved customer churn prediction, set meal recommendation etc. are applied.

Description

A kind of communication user consumption feature attribute reduction matching extracting method and data system

Technical field

The present invention relates to data mining technology field, particularly relate to a kind of communication user consumption feature attribute reduction matching and carry Access method and data system.

Background technology

Along with communication system continuous development of the reform, communication service industry competition is more and more fierce, how to safeguard keep use Family and accurate Products Show have become as each operator precision marketing and the most important thing maintaining work.Common carrier is in operation During have accumulated substantial amounts of business datum, important decision information that wherein may be the most valuable, data mining technology Can Extracting Knowledge from data effectively.As in off-network analyses and prediction, find that off-network user is possessed by data mining Feature, it was predicted that future customer at net state, it is possible to effectively for operator provide early warning, help operator open up targeted specifically Open user to keep and improvement business, possess the highest commercial application value.

The most perfect along with the development of infrastructure network, the appearance of especially 4G network, the demand of user is more and more diversified, The needs of service level and business tine are had higher requirement by user.The data that user uses various set meal to produce are various Various kinds, data type is complicated, and existing classification type data have again seriality data, such as continuously monthly flow, the telephone expenses etc. of half a year How information, carry out pretreatment to continuous data, embody the characteristic trend of these attributes, to digging user in the way of more succinct Loss model has important meaning.

For attributes such as the cost of the phone call of user, datas on flows, common carrier monthly generates a table, and record user the The cost of the phone call of i month and use data on flows.If individually considering the cost of the phone call of user every month and using data on flows, Have only to the operation of simple branch mailbox, by these seriality attribute divide into several classes's class labels, but do so, use can only be drawn Certain month in the past set meal in family uses distribution situation, can not describe the trend trend of user's future usage set meal, to customer loss, It is more weak that the models such as accurate Products Show excavate support.

Summary of the invention

Goal of the invention: the technical problem to be solved is for the deficiencies in the prior art, it is provided that a kind of communication is used Family consumption feature attribute reduction matching extracting method and data system, described method comprises the steps: step 1, collection data: Gather user according to common carrier real-world operation situation and use set meal consumption data, the use gathered according to common carrier One user of communication consumption data genaration of the continuous m in family month uses the table of set meal consumption, comprises A in table₁, A₂, ......A_mAttribute, these attributes are m the numerical attribute that generating principle is identical, A_mRepresent that the communication consumption of user's m-th moon belongs to Property；

The data of collection are carried out discretization between attribute by step 2, extract eigenvalue between attribute, make attribute dimensionality reduction, with energy letter Single reflection telex network consumption situation of change understood, supports the application such as customer churn prediction, accurate recommendation.

In step 2, extract eigenvalue between attribute and comprise the steps:

Step 2-1, it is assumed that X_uiRepresenting the user u communication consumption data in the i-th moon, i span is 1～m, in conjunction with using The communication consumption data of the continuous m in family month, use equation below to obtain the fluctuation situation of telex network consumption, i.e. variance

$S_{X_{um}}^2=\frac{1}{m}\[{(X_{u1}-\stackrel{\‾}{X_u})}^2+{(X_{u2}-\stackrel{\‾}{X_u})}^2+...+{(X_{um}-\stackrel{\‾}{X_u})}^2\],$

WhereinRepresent user u continuous m the middle of the month communication consumption meansigma methods；

Step 2-2: variance is normalized:

$\stackrel{\‾}{S_{X_{um}}^2}=\frac{1}{m}{\mathrm{\Σ}}_{i=1}^m{\left(\frac{X_{ui}-\stackrel{\‾}{X_u}}{\stackrel{\‾}{X_u}}\right)}^2,$

IfValue between [0,0.5], then judge that this user communication consumption in the middle of this m month is plateau, Terminate to extract eigenvalue between attribute；Otherwise, it is determined that this telex network Consumption Fluctuation is relatively big, proceed to step 2-3；

Step 2-3, three time periods before, during and after the communication consumption data of continuous for user m month being divided into, obtains each The meansigma methods of time period telex network consumption:

Wherein,It is used for determining the front time period,

Wherein,It is used for determining the time period of centre,

$X_3=\frac{X_{uk+1}+X_{uk+2}+...+X_{um}}{m-k},$

Wherein, X₁The communication consumption data of time period, X before expression user₂Represent the communication consumption number of time period in user According to, X₃The communication consumption data of time period after expression user,

Step 2-4, before obtaining respectively, meansigma methods X of middle time period communication consumption data₄In with, the rear time period communicates and disappears Take meansigma methods X of data₅, it may be assumed that

$X_4=\frac{X_1+X_2}{2}+1,$

$X_5=\frac{X_2+X_3}{2}+1;$

Step 2-5, if loss discriminant coefficient ε=0.5, this coefficient is experience value, according to before user, the middle time period communication Meansigma methods X of consumption data₄With in user, meansigma methods X of rear time period communication consumption data₅Ratio, judge as follows:

IfJudge that telex network consumption is downward trend, and have loss possible；

IfJudge that telex network consumption is plateau；

IfJudge that telex network consumption is ascendant trend；

Step 2-6, separately sets field A and moves towards situation for the trend identifying telex network consumption, replace original leading to The value that letter consumption attribute is gone up in every month, replaces m the communication consumption attribute of m month by field A, and A takes 2,0 and 1 Three values, representing telex network consumption respectively is downward trend, moderate tone and ascendant trend.

The present invention can draw user's this set meal of future usage on the basis of the continuous some months of user uses the data of set meal Trend moves towards value, namely extracts the eigenvalue of these connection attributes, the most fully continuous for user some months was used set Meal situation all utilizes, and uses set meal situation to link together these months again, than the number of the most directly these months of use According to more efficient, the rule excavated has more cogency, and whether prediction user will be continuing with future this set meal the most more has Reference significance.

The invention also discloses a kind of computer marketing data system, system is automatically by note or the side of tone information Formula is recommended user and is met customer flow use requirement, the simultaneously set meal of network minimal.

System is recommended user automatically by the mode that Virtual network operator APP pushes and is met customer flow use requirement, with Time network minimal set meal.

System is recommended user automatically by the mode of note or tone information and is met user's communication duration and use requirement, The set meal of network minimal simultaneously.

System automatically by the mode that Virtual network operator APP pushes recommend user meet user's communication duration use want Ask, simultaneously the set meal of network minimal.

Beneficial effect: multiple serial number attribute matchings are converted into single category attribute by the inventive method, for logical Consume (the telephone expenses amount of money, data on flows etc.) data credit household's continuous moon and carry out feature extraction, utilize the method, can by user even Multiple numerical attributes corresponding to the consumption information of continuous multiple months are converted into a category attribute, effectively reduce consumer record data set Dimension, embody more intuitively user's propensity to consume of continuous many months.Contribute to user's Future Consumption trend trend have Effect mining analysis.

Accompanying drawing explanation

Being the present invention with detailed description of the invention below in conjunction with the accompanying drawings and further illustrate, the present invention's is above-mentioned And/or otherwise advantage will become apparent.

Fig. 1 is that between attribute, discretization extracts eigenvalue process schematic.

Detailed description of the invention

The technical problem that the present invention intends to solve:

Traditional method to seriality attribute data discretization is confined to single attribute (discretization in corresponding attribute)；This Invention is paid close attention to multiple serial number attributes (the monthly flows etc. of the most continuous multiple months) with same nature, and extraction can Embody these property trends eigenvalues, discretization between corresponding attribute.

Technical scheme:

For the homogeneous data continuous value on multiple attributes, it is proposed that extract the attribute matching of the continuous feature of many attributes Method.

The core content of the present invention is within existing continuous multiple months, to use set meal consumption data according to user, carries out between attribute Discretization, extracts eigenvalue.Specific embodiments:

Technical method for making the present invention realize is apparent to understand, below in conjunction with detailed description of the invention, is expanded on further The present invention.

According to common carrier real-world operation situation, common carrier generates a user and use set meal to consume feelings every month The table of condition, comprises A in table₁, A₂... A_dAttribute, these attributes are d the numerical attribute that generating principle is identical, with continuous d As a example by the flow consumption of individual month, represent first month flow consumption attribute A respectively₁, second month flow consumption attribute A₂..., the flow consumption attribute A of d month_d。

(1) extract eigenvalue flow process between attribute and see Fig. 1, specific as follows:

Assume X_uiRepresent that the user u flow in the i-th moon uses consumption data.

Step 1: the flow combining the continuous m of user month uses consumption data, obtains user and uses the fluctuation situation of flow, I.e. variance:

WhereinRepresent that user u is even Continuous m the meansigma methods using flow the middle of the month.

Step 2: variance is normalized:IfValue in [0,0.5] Between, then illustrating that this user used flow in the middle of this m month is plateau；Otherwise, user uses flowed fluctuation relatively big, But not can determine that this undulatory property is to rise or decline, say, that not can determine that user uses flow to be into rising Trend, or downward trend.Need further analyzing and processing.

Step 3: three time periods before, during and after user uses traffic conditions be divided in the past.Obtain each time period user The meansigma methods of use data on flows:

$X_1=\frac{X_{u1}+X_{u2}+...+X_{uk}}{k},X_2=\frac{X_{up}+X_{up+1}+...+X_{uq}}{q-p+1},X_3=\frac{X_{uk+1}+X_{uk+2}+...+X_{um}}{m-k},$

Wherein,

Step 4: before obtaining the most respectively, the middle time period use data on flows meansigma methods and in, the rear time period use flow The meansigma methods of data, it may be assumed that

$X_4=\frac{X_1+X_2}{2}+1,X_5=\frac{X_2+X_3}{2}+1;$

Step 5: set ε=0.5, according to before user, the middle time period use in meansigma methods and the user of data on flows, the rear time Section uses the ratio of the meansigma methods of data on flows, is concluded that

If 1.Then user uses traffic conditions to be downward trend；

If 2.Then user uses traffic conditions to be plateau；

If 3.Then user uses traffic conditions to be ascendant trend；

Step 6: in sum, separately sets field A and uses for identifying user the trend of flow to move towards situation, thus generation The value gone up in every month for original flow attribution.M the flow attribution of m month is replaced by field A.Here A takes 2,0,1 3 values.Represent decline, steady, rising respectively.

Attribute approximating method based on continuous many numerical attributes characteristics extraction proposed by the invention, not only makes full use of User uses the data that set meal is consumed in the past, but also the value of m the same alike result that same attribute was produced m the middle of the month Fully contact is together, thus effectively reduces the dimension of consumer record data set, makes mining data source dimension be substantially reduced, this Sample, the most simply comprises user and uses the behavior of set meal in the past, also comprise the trend of user's future usage set meal in data source Move towards feature, so will be greatly improved the correctness of digging efficiency and Result.Illustrate in conjunction with the examples below.

Embodiment 1

Set up computer marketing data system according to above-mentioned calculation procedure, use this system user each to operator to disappear Expense data are scanned, and gather, then are analyzed calculating.As a example by user u, its six collect the middle of the month " use flow Consumption " as shown in Table 1:

Table one

January	February	March	April	May	June
						0.2G	1.3G	2.8G	4.9G	5.2G	16.6G

Following steps calculating is carried out according to the data in table one:

Step 1:

Step 2: due to 6.76 not between interval [0,0.5], so needing further to judge；

Step 3:

Step 4:

Step 5:So the propensity to consume of user's " use flow " is in rising trend.

System storage result of calculation, and contrasting with the package information of operator, can automatically select recommendable can Meeting user and make traffic requirement, simultaneously the set meal of network minimal, be pushed to user with short message mode or other modes, set meal is as follows Shown in table:

50 yuan	80 yuan	100 yuan	150 yuan	200 yuan	300 yuan
						1G	2G	3G	5G	10G	20G

Then system can recommend the flow bag of 20G/300 unit to user with short message mode automatically, had both met the flow of user Using makes again user feel the most economical.

Embodiment 2

As a example by user u, its six " the voice call durations " collected the middle of the month as shown in Table 2:

Table two

January	February	March	April	May	June
						47	52	166	13	49	2

Following steps calculating is carried out according to the data in table two:

Step 1:

Step 2: due to 0.9 not between interval [0,0.5], so needing further to judge.

Step 3:

Step 4:

Step 5:

According to result, system storage result of calculation, judges that this user has off-network trend, can automatically identify and with lettergram mode Being pushed to maintain personnel, prompting is maintained personnel and is paid close attention to this user and take marketing measures.

The ultimate principle of the present invention and principal character and advantages of the present invention have more than been shown and described.The skill of the industry The art personnel simply explanation it should be appreciated that the present invention is not restricted to the described embodiments, described in above-described embodiment and description The principle of the present invention, without departing from the spirit and scope of the present invention, the present invention also has various changes and modifications, these Changes and improvements both fall within scope of the claimed invention.Claimed scope by appending claims and Its equivalent defines.

Claims (6)

1. a communication user consumption feature attribute reduction matching extracting method, it is characterised in that comprise the following steps:

Step 1, gathers data: according to one use of communication consumption data genaration of the continuous m of user month that common carrier gathers Family uses the table of set meal consumption, comprises A in table₁, A₂... A_mAttribute, these attributes are the m number that generating principle is identical Value attribute, A_mRepresent the communication consumption attribute of user's m-th moon；

The data of collection are carried out discretization between attribute by step 2, extract eigenvalue between attribute, make attribute dimensionality reduction and reflect user Communication consumption situation of change.

Method the most according to claim 1, it is characterised in that in step 2, extracts eigenvalue between attribute and includes walking as follows Rapid:

Step 2-1, it is assumed that X_uiRepresenting the user u communication consumption data in the i-th moon, i span is 1～m, in conjunction with user even The communication consumption data of continuous m month, use equation below to obtain user and use the fluctuation situation of communication, i.e. variance

$S_{X_{um}}^2=\frac{1}{m}\[{(X_{u1}-\stackrel{\‾}{X_u})}^2+{(X_{u2}-\stackrel{\‾}{X_u})}^2+...+{(X_{um}-\stackrel{\‾}{X_u})}^2\],$

WhereinRepresent user u continuous m the middle of the month communication consumption meansigma methods；

Step 2-2: variance is normalized:

$\stackrel{\‾}{S_{X_{um}}^2}=\frac{1}{m}{\mathrm{\Σ}}_{i=1}^m{\left(\frac{X_{ui}-\stackrel{\‾}{X_u}}{\stackrel{\‾}{X_u}}\right)}^2,$

IfValue between [0,0.5], then judge that this user used communication consumption to be plateau in the middle of this m month, Terminate to extract eigenvalue between attribute；Otherwise, it is determined that this user uses communication consumption undulatory property relatively big, proceed to step 2-3；

Step 2-3, three time periods before, during and after the communication consumption data of continuous for user m month being divided into, obtains each time The meansigma methods of section telex network consumption:

Wherein,It is used for determining the front time period,

Wherein,It is used for determining the time period of centre,

$X_3=\frac{X_{uk+1}+X_{uk+2}+...+X_{um}}{m-k},$

Wherein, X₁The communication consumption data of time period, X before expression user₂Represent the communication consumption data of time period, X in user₃ The communication consumption data of time period after expression user；

Step 2-4, before obtaining respectively, meansigma methods X of middle time period communication consumption data₄In with, rear time period communication consumption data Meansigma methods X₅, it may be assumed that

$X_4=\frac{X_1+X_2}{2}+1,$

$X_5=\frac{X_2+X_3}{2}+1;$

Step 2-5, if loss discriminant coefficient ε=0.5, according to before user, meansigma methods X of middle time period communication consumption₄With user In, meansigma methods X of rear time period communication consumption data₅Ratio, judge as follows:

IfJudge that user uses communication consumption to be downward trend, have loss possible；

IfJudge that telex network consumption is plateau；

IfJudge that telex network consumption is ascendant trend；

Step 2-6, separately sets field A and moves towards situation for identifying the telex network propensity to consume, replace original communication consumption The value that attribute is gone up in every month, replaces m the communication consumption attribute of m month by field A, and A takes 2,0 and 1 three value, Representing telex network consumption respectively is downward trend, moderate tone and ascendant trend.

3. the computer marketing data system using method described in claim 2, it is characterised in that system is automatically by short The mode of letter or tone information is recommended user and is met customer flow use requirement, the simultaneously set meal of network minimal.

4. the computer marketing data system using method described in claim 2, it is characterised in that system is automatically by net The mode that network operator APP pushes is recommended user and is met customer flow use requirement, the simultaneously set meal of network minimal.

5. the computer marketing data system using method described in claim 2, it is characterised in that system is automatically by short The mode of letter or tone information is recommended user and is met user's communication duration use requirement, the simultaneously set meal of network minimal.

6. the computer marketing data system using method described in claim 2, it is characterised in that system is automatically by net The mode that network operator APP pushes is recommended user and is met user's communication duration use requirement, the simultaneously set meal of network minimal.