CN104113869A - Signaling data-based prediction method and system for potential complaint user - Google Patents

Abstract

The invention, which belongs to the technical field of service support in a mobile communication field, discloses a signaling data-based prediction method and system for a potential complaint user. Network-wide user feature vectors containing complaint user feature vectors and non-complaint user feature vectors are established based on A interface signaling data; a serve similarity degree of the non-complaint users and the complaint users is calculated according to the complaint user feature vectors and the non-complaint user feature vectors; and a potential complaint user in the non-complaint users is determined according to the service similarity degree, wherein the higher the service similarity degree is, the higher the possibility of the potential complaint user. With the method and the system, feature vector modeling is carried out on massive user data and potential complaint user are dug out according to the complaint users. Therefore, before the real complaint user occurs or the user carries out network transferring, early warning can be realized and thus the complaint or transferring event can be eliminated in the bud, thereby providing the basis for user perception improvement and customer maintenance cost reduction for the operator.

Description

A kind of potential report user's Forecasting Methodology and system based on signaling data

Technical field

The present invention relates to the business support technology in moving communicating field, be specifically related to a kind of potential report user's Forecasting Methodology and system based on signaling data.

Background technology

Along with the development of the communications industry, the arrival in 4G epoch, various new communication network problems continue to bring out, and customer complaint amount also grows with each passing day.Simultaneously, the present age, client was no longer such simple mode of making a phone call, send short messages for the service providing of operator, more focus on personalized, good client perception, only be improved user satisfaction, reduce customer complaint, and contingent complaint advanced processing, the development of guarantee Operator Specific Service.Operator has the user data of magnanimity at present, no matter be magnanimity signaling data, or a large amount of customer data data, all there is great value.But most of operator is for the handling and prevention of customer complaint, only to go to consider from the angle of network index, do not make full use of the advantage of mass data resource, by the method for data mining, formulate for user's personalization and complain and solve and complain prediction scheme.

At moving communicating field, prior art scheme, is mainly to process for the complaint having occurred, and based on remedying of accomplished fact, cannot to the user that may occur to complain, predict in advance, prevents trouble before it happens.Complaint handling flow process is generally at present: client produces discontented for the service of operator, then dial customer service phone or complain by the network platform, client service center receives after customer complaint, the single new business centre of support of giving of group, work order, according to complaining particular content to analyze, locate, solve user's complaint problem, is then replied in centre of support.Therefore, the flow process of " customer service--monitoring--customer service " has been experienced in this type of complaint handling, and treatment effeciency is low, and the wasting of resources is large, and the process limited be difficult to guarantee, customer satisfaction is poor.Existing complaint Forecasting Methodology is only also to calculate some network element indexs, as weak covering, anomalous event etc., sets up warning system, with this, judges user's communication quality, thereby the user who complains may occur in prediction.

On the whole, existing complaint Forecasting Methodology mainly contains following several:

One class is Qualitative Forecast Methods, is the history and current situation of things development is made explanations, analyzed and judges, thus one or more possibilities of the comprehensive future trend of pointing out things development, as market survey method, Delphi method etc.

Another kind of is quantitative forecast method, is mainly computing network index, and user's inventory that output network index is poor, as complaining prediction inventory.

There is following problem in existing complaint prediction scheme:

1. the method for market survey can only, for limited sample and customer group, cannot be grasped a large amount of whole customer complaint features.

2. existing to calculate the complaint Forecasting Methodology of network element index, some the rigid indexs of network of all take are judgment condition, are difficult to reflection mobile subscriber subjective perception situation, and carry out personalization analysis according to user self differentiation feature.

3. existing complaint handling is all after complaint has occurred, the mode of remedying afterwards and explaining, contact staff is only confined to complaining above the pacifying, process of client, or to complaining content to do the plain analysis of causes, there is no profound customer complaint analysis.

4. existing method is work order based on occurring to complain, and a lot of user produces severe perception for service, and they can't go to complain, but directly turn net, and so existing complaint handling system just cannot be shown loving care in advance and keep this class user.According to marketing principle, develop a required cost of new user and be and maintain 6 times of old user, thereby look-ahead to go out the user that may complain particularly important to the income of operator and profit.

Summary of the invention

Needs for the defect existing in prior art and practical application, the object of the present invention is to provide a kind of potential report user's Forecasting Methodology and system based on signaling data, by the method and system, can go out potential report user by look-ahead, give warning in advance, improve user's perception, reduce maintenance cost.

For achieving the above object, the technical solution used in the present invention is as follows:

Potential report user's Forecasting Methodology based on signaling data, comprises the following steps:

(1) take A interface signaling data as Foundation the whole network user characteristics vector; Described the whole network service feature vector comprises report user's characteristic vector and report user's characteristic vector not; Parameter in user characteristics vector is the service parameter of corresponding business;

(2) according to report user's characteristic vector and not report user's characteristic vector calculate not report user's and report user business similarity;

(3) according to the potential report user in the definite not report user of business similarity, the possibility that the higher user of business similarity is potential report user is larger.

Further, a kind of potential report user's Forecasting Methodology based on signaling data as above, between step (1) and step (2), also comprises:

(1-2) the whole network user characteristics vector is carried out to standardization.

Further, a kind of potential report user's Forecasting Methodology based on signaling data as above, in step (1), the single business of take is set up the whole network user characteristics vector as granularity; Described business comprises call, note, no-response paging, position renewal and switching on and shutting down.

Further, a kind of potential report user's Forecasting Methodology based on signaling data as above, chosen distance report user is less than setting business duration, corresponding with complaint business the time of complaining and sets up the whole network user characteristics vector.

Further, a kind of potential report user's Forecasting Methodology based on signaling data as above, in step (1-2), describedly carries out standardization by the whole network user characteristics vector and refers to the parameter values of characteristic vector is normalized into [1,1].

Further, a kind of potential report user's Forecasting Methodology based on signaling data as above, the parameter values of characteristic vector is carried out to standardized formula be:

Wherein, t represents the numerical value of certain parameter of certain user in the whole network user characteristics vector, and min represents the minimum value in certain parameter values described in the whole network user, and max represents the maximum in certain parameter values described in the whole network user;

When max=min, the parameter values of corresponding parameter in the whole network characteristic vector is unified to assignment.

Further again, a kind of potential report user's Forecasting Methodology based on signaling data as above, in its step (2), by calculating report user's characteristic vector and the Euclidean distance of corresponding parameter in report user's characteristic vector not, calculate not report user's and report user business similarity, the less similarity of Euclidean distance is higher; The computing formula of described Euclidean distance is:

$d(x,y)=\sqrt{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-y_k)}^2}$

Wherein, d (x, y) represents not report user's and report user business similarity, and x represents report user's characteristic vector, and y represents not report user's characteristic vector, and n represents the number of service parameter in the whole network user characteristics vector, x _kthe numerical value that represents k service parameter in report user's characteristic vector, y _kthe numerical value that represents k service parameter in report user's characteristic vector not.

Further, a kind of potential report user's Forecasting Methodology based on signaling data as above, in step (3), when report user and report user's similarity does not meet the decision threshold of setting, this user is defined as potential report user.

Potential report user's prognoses system based on signaling data, comprising:

Characteristic vector is set up module: for take A interface signaling data as Foundation the whole network user characteristics vector; Described the whole network service feature vector comprises report user's characteristic vector and report user's characteristic vector not; Parameter in user characteristics vector is the service parameter of corresponding business;

Similarity calculation module: for according to report user's characteristic vector and not report user's characteristic vector calculate not report user's and report user business similarity;

Potential report user's determination module: according to the potential report user in the definite not report user of business similarity, the possibility that the higher user of business similarity is potential report user is larger.

Further, a kind of potential report user's prognoses system based on signaling data as above, this system also comprises:

Vector standardized module: for the whole network user characteristics vector is carried out to standardization.

Beneficial effect of the present invention is: method and system of the present invention, according to the data characteristics that report user occurs, find the potential report user that complaint may occur, and improve the whole user's of the whole network satisfaction, can effectively reduce user and turn net rate.The complaint modeling of the method based on report user, thereby Auto-matching goes out the user that may also have same problem in network, give warning in advance, by complaint or turn net situation and eliminate in bud, thereby realize and drive the whole network user according to report user, realize the focusing orientation problem of network problem, can effectively reduce the network operation cost of operator, and the method is compared with traditional Forecasting Methodology, can effectively improve the accuracy of prediction.

Accompanying drawing explanation

Fig. 1 is the structured flowchart of a kind of potential report user's prognoses system based on signaling data in the specific embodiment of the invention;

Fig. 2 is the flow chart of a kind of potential report user's Forecasting Methodology based on signaling data in the specific embodiment of the invention.

Embodiment

Below in conjunction with Figure of description and embodiment, the present invention is described in further detail.

Fig. 1 shows the structured flowchart of a kind of potential report user's prognoses system based on signaling data in embodiment, this system mainly comprises that characteristic vector sets up module 11, vectorial standardized module 12, similarity calculation module 13 and potential report user's determination module 14, wherein:

Characteristic vector is set up module 11 for take A interface signaling data as Foundation the whole network user characteristics vector; Described the whole network service feature vector comprises report user's characteristic vector and report user's characteristic vector not; Parameter in user characteristics vector is the service parameter of corresponding business;

Vector standardized module 12 is for carrying out standardization by the whole network user characteristics vector;

Similarity calculation module 13 for according to report user's characteristic vector and not report user's characteristic vector calculate not report user's and report user business similarity;

Potential report user's determination module 14 is according to the potential report user in the definite not report user of business similarity, and the possibility that the higher user of business similarity is potential report user is larger.

Fig. 2 shows the flow chart of a kind of potential report user's Forecasting Methodology based on signaling data based on system shown in Fig. 1 in this embodiment, and the method comprises the following steps:

Step S21: take A interface signaling data as Foundation the whole network user characteristics vector;

Forecasting Methodology of the present invention is based on A interface signaling data being carried out to modeling and the analysis of characteristic vector, according to the data characteristics that report user occurs, finds the potential report user that complaint may occur.Be that this programme is that to complain the data that report user occurred in work order be Sample Establishing model, selecting the signaling data relevant with complaint is modeling object.

First, take A interface signaling data as Foundation the whole network user characteristics vector, described the whole network user characteristics vector comprises report user's characteristic vector and report user's characteristic vector not, parameter in user characteristics vector is the service parameter of corresponding business, that is to say that the user characteristics vector in present embodiment is business (communication service) characteristic vector, report user's characteristic vector is that report user's related service parameter is carried out to modeling, and report user's vector is not that the corresponding relevant parameter of the user to also not having to complain carries out modeling.

A interface is the communication interface between network subsystem (NSS) and base station sub-system (BSS), the common technology that the service parameter data by A interface signaling data acquisition different communication business are this area.

When setting up the whole network user characteristics vector, the single business of take is set up user characteristics vector as granularity, and wherein, described business comprises call, note, no-response paging, position renewal, switching on and shutting down etc.When which time business of concrete selection is set up user characteristics vector, in present embodiment, preferably with chosen distance report user, be less than setting business duration, corresponding with complaint business the time of complaining and set up the whole network user characteristics vector.For example, a report user complains talk business certain day 10, can select so the whole network user's of 10 left and right call supplemental characteristic to set up characteristic vector.

Different business, the parameter of its characteristic vector is different, user can select the service parameter needing according to actual needs.For example, for this business of call, the detailed record sheet of business diagnosis that can utilize system is that BDR table is set up talk business characteristic vector, and in present embodiment, talk business characteristic vector field is as shown in table 1 below.

In table 1, field one row are the parameter of talk business characteristic vector.

Table 1

Step S22: the whole network user characteristics vector is carried out to standardization;

In present embodiment, the whole network user characteristics vector aims of standardization are that the parameter values of the whole network user characteristics vector of setting up in step S21 is normalized between-1 to 1, can certainly select other standard, this is because of some parameter index numerical value is very large but importance is not maximum, by standardization, making all index importance identical, is that the parameter of characteristic vector has identical value scope.The formula of present embodiment Plays is as follows:

Wherein, t represents the numerical value of certain parameter of certain user in the whole network user characteristics vector, min represents the whole network user (all users in the whole network user characteristics vector, comprise report user and report user not) described in minimum value in certain parameter values, max represents the maximum in certain parameter values described in the whole network user;

When max=min, be that the maximum of the numerical value of a certain parameter in the whole network user is when identical with minimum value, can not adopt above-mentioned formula to carry out standardization, at this moment the parameter values of corresponding parameter in characteristic vector can be unified to assignment, if assignment is that mean value or assignment are 0.

For example, have 5 users in the whole network user, the 1st user is the user that actual generation is complained, and other 4 users are the not report users in existing network, and may comprise potential report user in these 4 users.Talk business for these 5 users, in present embodiment, select five parameter indexs in the talk business characteristic vector shown in table 1: rise and to exhale the time, rise and exhale longitude, rise and exhale latitude, on-hook longitude, on-hook latitude, the whole network user characteristics vector that comprises these five parameters of foundation is as shown in table 2.

In table 2, every a line represents a user's characteristic vector, and wherein report user's a line representative is report user's characteristic vector.

?	StartTime	Rise and exhale longitude	Rise and exhale latitude	On-hook longitude	On-hook latitude
						Report user	16	111.65865	40.81055	111.65865	40.81055
User 1	20	111.412865	40.028756	111.412865	40.028756
						User 2	8	111.52961	40.704372	111.52961	40.704372
User 3	11	111.67084	40.82067	111.67084	40.82067
						User 4	10	111.644173	40.82008	111.64417	40.82008

Table 2

In above-mentioned table 2, the maximum of each row and minimum value are that maximum and the minimum value of each service parameter is as shown in table 3:

?	StartTime	Rise and exhale longitude	Rise and exhale latitude	On-hook longitude	On-hook latitude
						max	20	111.67084	40.82067	111.67084	40.82067
min	8	111.412865	40.028756	111.412865	40.028756
						max-min	12	0.257975	0.791914	0.257975	0.791914

Table 3

Adopt the talk business characteristic vector in standardization formula his-and-hers watches 2 to carry out standardization, report user's the StartTime field of take is example, and this field value is 16, and its standardized calculation formula is as follows:

$(2\×\frac{16-8}{20-8})-1=0.33$

Data in table 2 after all users' characteristic vector standard parameter are as shown in table 4:

?	StartTime	Rise and exhale longitude	Rise and exhale latitude	On-hook longitude	On-hook latitude
						Report user	0.33	0.91	0.97	0.91	0.97
User 1	1	-1	-1	-1	-1
						User 2	-1	-0.09	0.71	-0.09	0.71
User 3	-0.5	1	1	1	1
						User 4	-0.67	0.79	1.00	0.79	1.00

Table 4

Step S23: according to report user's characteristic vector and not report user's characteristic vector calculate not report user's and report user business similarity;

After setting up the whole network user characteristics vector, need to the method by comparing calculate in the whole network user not report user's and report user business similarity, thereby in report user, do not finding the immediate potential report user with report user.Lookup method concrete in present embodiment is, by calculating report user's characteristic vector and Euclidean distance calculating the whole network user of corresponding parameter in report user's characteristic vector not and report user's similarity, Euclidean distance is less, and similarity is higher, is that potential report user's possibility is just larger.In present embodiment, the computing formula of Euclidean distance is as follows:

$d(x,y)=\sqrt{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-y_k)}^2}$

Wherein, d (x, y) represents not report user's and report user business similarity, and x represents report user's characteristic vector, and y represents not report user's characteristic vector, and n represents the number of service parameter in the whole network user characteristics vector, x _kthe numerical value that represents k service parameter in report user's characteristic vector, y _kthe numerical value that represents k service parameter in report user's characteristic vector not.

In actual applications, a plurality of if the report user in the whole network user characteristics vector has, need to calculate respectively the not business similarity between report user and each report user.

In present embodiment, for the whole network user characteristics vector in above-mentioned table 2, the number n of service parameter is 5, and user 1 is calculated as follows with report user's Euclidean distance:

$\begin{array}{c}{D}_1(x,y)=\sqrt{{(0.33-1)}^2+{(0.91+1)}^2+{(0.97+1)}^2+{(0.91+1)}^2+{(0.97+1)}^2}\\ =3.95\end{array}$

In table 24 the Euclidean distance between report user and report user is not as shown in table 5 below:

Customs Assigned Number	Euclidean distance
		User 1	3.95
User 2	1.98
		User 3	0.86
User 4	1.02

Table 5

Step S24: according to the potential report user in the definite not report user of business similarity, the possibility that the higher user of business similarity is potential report user is larger.

Because the Euclidean distance of characteristic vector just represents the similarity between the whole network user and report user, distance is less, illustrates that both speech quality, client perception situation are more approaching.In table 5, can see, immediate with report user is user 3, although illustrate that user 3 does not complain, but be very likely that user awareness is bad, the user that speech quality or network quality are poor, if processed not in time, directly turns net but may cause user not complain.

In actual applications, can select K and the immediate potential report user of report user, as prediction inventory.The selection of K should be arranged in suitable scope by experience, if K value is too little, can only find potential report user seldom, and scope is too little, if K value is too large, also divides the good user of a lot of perception into potential report user, causes erroneous judgement to increase.Therefore,, by rational judgment threshold is set, when the user in report user not and report user's similarity meet the decision threshold of setting, this user is defined as potential report user.

In addition, potential report user's seek scope, also must be not necessarily the whole network user, because the increase of data volume means amount of calculation and searches the increase of time, also the Hot Spot that can select several problems to take place frequently according to actual conditions is searched potential report user, also can improve the hit rate of prediction.

In order to verify that this programme is for potential report user's prediction effect, by obtaining certain districts and cities, complain work order, in the complaint serious area of network problem that takes place frequently, test, by this programme, exported potential report user's inventory, by the mode of call-on back by phone, determine whether prediction is accurate.Test the scheme of traditional calculations network index simultaneously and complained the accuracy of prediction.

Complain 100 of one week report users of work order, by analysis of history report user data, utilize this programme analysis to export 1000 users of potential report user's inventory, and these 1000 users are carried out to call-on back by phone, result shows that wherein 675 users think that speech quality has problems, there is the problems such as call drop, when call voiceless sound, note send unsuccessfully, illustrate that this programme accuracy rate is about 67.5%.Investigate the complaint work order in next week simultaneously, complain for totally 110 times, wherein 58 people are present in potential report user's inventory of this programme prediction, and the hit rate that this programme is described is 52.7%.

And by calculating the network indexes such as weak covering, anomalous event, export 1000 potential report users, the result of call-on back by phone, has 376 users to reflect and speech quality existing problems illustrates that traditional scheme predictablity rate is about 37.6%.And in 110 report users of reality of second week, only have 8 people to appear in prediction inventory, illustrate that traditional scheme complaint prediction effect is far below this programme.

By the data that existing network is tested and client pays a return visit, show, potential report user's forecasting accuracy of this programme is far above traditional scheme.

Obviously, those skilled in the art can carry out various changes and modification and not depart from the spirit and scope of the present invention the present invention.Like this, if within of the present invention these are revised and modification belongs to the scope of the claims in the present invention and equivalent technology thereof, the present invention is also intended to comprise these changes and modification interior.

Claims (10)

1. the potential report user's Forecasting Methodology based on signaling data, comprises the following steps:

(1) take A interface signaling data as Foundation the whole network user characteristics vector; Described the whole network service feature vector comprises report user's characteristic vector and report user's characteristic vector not; Parameter in user characteristics vector is the service parameter of corresponding business;

(2) according to report user's characteristic vector and not report user's characteristic vector calculate not report user's and report user business similarity;

(3) according to the potential report user in the definite not report user of business similarity, the possibility that the higher user of business similarity is potential report user is larger.

2. a kind of potential report user's Forecasting Methodology based on signaling data as claimed in claim 1, is characterized in that: between step (1) and step (2), also comprise:

(1-2) the whole network user characteristics vector is carried out to standardization.

3. a kind of potential report user's Forecasting Methodology based on signaling data as claimed in claim 1 or 2, is characterized in that: in step (1), the single business of take is set up the whole network user characteristics vector as granularity; Described business comprises call, note, no-response paging, position renewal and switching on and shutting down.

4. a kind of potential report user's Forecasting Methodology based on signaling data as claimed in claim 3, is characterized in that: chosen distance report user is less than setting business duration, corresponding with complaint business the time of complaining and sets up the whole network user characteristics vector.

5. a kind of potential report user's Forecasting Methodology based on signaling data as claimed in claim 2, is characterized in that: in step (1-2), describedly the whole network user characteristics vector is carried out to standardization refer to the parameter values of characteristic vector is normalized into [1,1].

6. a kind of potential report user's Forecasting Methodology based on signaling data as claimed in claim 5, is characterized in that: the parameter values of characteristic vector is carried out to standardized formula is:

Wherein, t represents the numerical value of certain parameter of certain user in the whole network user characteristics vector, and min represents the minimum value in certain parameter values described in the whole network user, and max represents the maximum in certain parameter values described in the whole network user;

When max=min, the parameter values of corresponding parameter in the whole network characteristic vector is unified to assignment.

7. a kind of potential report user's Forecasting Methodology based on signaling data as claimed in claim 1, it is characterized in that: in step (2), by calculating report user's characteristic vector and the Euclidean distance of corresponding parameter in report user's characteristic vector not, calculate not report user's and report user business similarity, the less similarity of Euclidean distance is higher; The computing formula of described Euclidean distance is:

$d(x,y)=\sqrt{\underset{k=1}{\overset{n}{\mathrm{\Σ}}}{(x_k-y_k)}^2}$

Wherein, d (x, y) represents not report user's and report user business similarity, and x represents report user's characteristic vector, and y represents not report user's characteristic vector, and n represents the number of service parameter in the whole network user characteristics vector, x _kthe numerical value that represents k service parameter in report user's characteristic vector, y _kthe numerical value that represents k service parameter in report user's characteristic vector not.

8. a kind of potential report user's Forecasting Methodology based on signaling data as described in claim 1 or 7, it is characterized in that: in step (3), when report user and report user's similarity does not meet the decision threshold of setting, this user is defined as potential report user.

9. the potential report user's prognoses system based on signaling data, comprising:

Characteristic vector is set up module: for take A interface signaling data as Foundation the whole network user characteristics vector; Described the whole network service feature vector comprises report user's characteristic vector and report user's characteristic vector not; Parameter in user characteristics vector is the service parameter of corresponding business;

Similarity calculation module: for according to report user's characteristic vector and not report user's characteristic vector calculate not report user's and report user business similarity;

Potential report user's determination module: according to the potential report user in the definite not report user of business similarity, the possibility that the higher user of business similarity is potential report user is larger.

10. a kind of potential report user's prognoses system based on signaling data as claimed in claim 9, is characterized in that, this system also comprises:

Vector standardized module: for the whole network user characteristics vector is carried out to standardization.