JP2015153189A - Statistical user behavior analyzer, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To utilize distribution of measurement result made by an imbalanced user as an outcome corresponding to a behavior pattern of a general user.SOLUTION: The present invention acquires measurement data for designated items in an evaluation object area; normalizes the measurement data; acquires density data for the same items as those of the measurement data from density distribution data storage means for storing in advance density distribution data of user behavior; and corrects bias of the measurement data by weighting the normalized measurement data with the density data.

Description

  The present invention relates to a statistical user behavior analysis apparatus, method, and program, and more particularly to a statistical user behavior analysis apparatus and method for analyzing user behavior from data measured in a wide area in diverse mobile communications, and Regarding the program.

  In recent years, as the use of mobile communication networks such as 3G / LTE (Third Generation / Long Term Evolution) increases, measurement, evaluation, and grasping of the communication quality have become issues.

  Because of the diversity of mobile communications (changes due to time, space, etc.), it is difficult to measure a small number of samples and represent the actual quality with average values.

  Therefore, measurement data called measurement results for each place and time is required. Naturally, it is difficult to measure at any place and time by a specialized person or the like, which is actually difficult from the viewpoint of cost.

  Therefore, there is an approach to measure the diversity of mobile communication by measuring the measurement behavior by the measurement application installed on the terminal on the user's side, and analyzing the logs aggregated on the server side to cover various user behaviors. (For example, see Non-Patent Document 1).

RBBTODAY, “Analysis with Big Data! The au that won the LTE game with iPhone 5s / 5c” [online], 1 October 2013, 19:00 http://www.rbbtoday.com/article/2013/10 /01/112343.html

  However, although the method of Non-Patent Document 1 is not a problem from the viewpoint of individual users or terminals, it is dependent on the behavior of the user who has installed the application from the viewpoint of measuring the entire network. There is a problem that there is a bias.

  Furthermore, there are the following problems.

a. Install user population attribute bias:
Compared to general users, they tend to be heavy users and enthusiasts who prefer certain special applications.

b. Bias of different behavior patterns (date / time / location) of installation users:
Since there are users who measure time for killing time and entertainment, the distribution is different from the date and place of communication use of general users.

  A. b. Therefore, it is not suitable to measure and use the measurement result of the mobile communication network by the application of the user terminal corresponding to the use quality of general users.

  In addition, it is possible to normalize the sample deviation of the measurement results, but it is also possible to perform simple normalization and treat points with a large number of users and points with a small number of users equally with equal weight. It is not suitable for measurement corresponding to.

  The present invention has been made in view of the above points, and a statistical user behavior analysis apparatus, method, and program capable of using a distribution of measurement results by biased users as a result corresponding to a general user's behavior pattern The purpose is to provide.

According to one aspect, a statistical user behavior analysis device that weights measurement data to obtain data measured in a wide area as behavior data of a general user,
Measurement data extraction means for acquiring measurement data including measurement position information, measurement values, time, all items of attribute, or any item, and normalizing;
Density distribution data storage means for storing user action density distribution data in advance;
Density data extraction means for acquiring density data of the same items as the measurement data acquired by the measurement data extraction means from the density distribution data storage means;
There is provided a statistical user behavior analysis device comprising weight calculation means for weighting measurement data normalized by the measurement data extraction means using the density data.

  According to one aspect, the measurement data is once normalized in time and space, and the weight is obtained using density distribution data such as user behavior, thereby correcting the bias using the weight. It becomes possible. As a result, even if “measurement by all users' terminals” cannot be realized directly, the quality distribution in the scene according to the usage behaviors of all users is measured systematically in correspondence with the behaviors of all users. Even without using the very high-cost data that has been implemented, it is possible to appropriately estimate from the total data of the self-measurement by the user obtained at a low cost.

It is an example of composition of a statistical user action analysis device in one embodiment of the present invention. It is a flowchart of the process outline | summary in one embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows a configuration example of a statistical user behavior analysis apparatus according to an embodiment of the present invention.

  The statistical user behavior analysis apparatus 100 shown in the figure includes a density distribution data storage unit 110, a measurement data extraction unit 120, a density data extraction unit 130, a weighted average calculation unit 140, and an output unit 150. The measurement data extraction unit 120 Are connected to an external measurement data DB 10.

  In addition to the above configuration, the statistical user behavior analysis apparatus 100 includes storage means such as a memory that temporarily holds calculation results and the like.

  The measurement data DB 10 holds, as measurement data M, measurement values, latitudes, longitudes, altitudes, times, ages, genders, and the like as shown in Table 1. Note that the measurement data is not limited to the example in Table 1 and can be arbitrarily set.

表１の各計測値、緯度・経度等の位置情報、属性等は単に時系列に記録されており、属性毎のように分類されているものではない。また、計測値はスループット値、遅延値、損失率等なんでもよい。実際の計測データでは、複数の計測値が含まれるが、ここでは説明の簡単化のため、１種類の計測値のみを扱うものとする。

Each measured value in Table 1, position information such as latitude and longitude, attributes, and the like are simply recorded in time series, and are not classified as per attribute. The measured value may be anything such as a throughput value, a delay value, a loss rate, and the like. Although actual measurement data includes a plurality of measurement values, only one type of measurement value is handled here for simplicity of explanation.

  The density distribution data storage unit 110 is generated in advance and holds, as user density data D, density, latitude, longitude, altitude, time zone, age, sex, etc. as shown in Table 2.

表２において、緯度、経度、標高は、予め決められた値（後述の評価対象エリア座標にある「座標位置」に対応する値）とする。座標、年代、性別共に繰り返し現れるが、全ての属性が共通なレコードは含まれない（同じ座標で、年代が異なるもの、性別が異なるもの等は含まれる）。

In Table 2, latitude, longitude, and altitude are values determined in advance (values corresponding to “coordinate positions” in evaluation target area coordinates described later). Coordinates, age, and gender appear repeatedly, but records that share all attributes are not included (those that have the same coordinates but have different ages, different genders, etc.).

For example, “0” is male and “1” is female. For example, in tz ₁ , the time zone indicates a time interval from tz _1s (start time) to tz _1e (end time). Further, not all the coordinate data of the evaluation target area coordinate set K are included, and there may be missing data. For example, if there is a 10 × 10 100-cell mesh, not all cells have data, and there may be cells that do not contain any data. Similarly, not all ages and genders are covered.

The ages yz ₁ , yz ₂ , etc. in the measurement data DB 10 indicate labels according to the ages, for example,
Z1: Less than Z ₁
Z2: Z ₁ or more, less than Z2
Z3: ...
It is defined as Here, Z ₁ , Z ₂ , Z ₃ ,... Are age values indicating the breaks of each age, such as Z ₁ = 20, Z ₂ = 35, and Z ₃ = 50.

Also, the identification target area coordinate set of the measurement data M in the measurement data DB 10 is
K = {(Lat ₁ , Lon ₁ , Alt ₁ ), (Lat ₂ , Lon ₂ , Alt ₂ ),…, (Lat _k , Lon _k , Alt _k )}
And K above is the mesh of the evaluation area, and the i (i = 1,2, ... K) th latitude (Lat), longitude (Lon), altitude (Alt) of each divided unit (cell) Indicates the position of the southwest bottom. The size of each cell is dl _at , d _lon , and d _alt in the latitude, longitude, and elevation directions, respectively. That is, the range of the first cell is as follows.

Latitude: Lat ₁ or more and less than Lat ₁ + d _lat Longitude: Lon ₁ or more and less than Lon ₁ + d _lon Altitude: Alt ₁ or more and less than Alt ₁ + d _alt Actually, two-dimensional data of only latitude and longitude may be used. In that case, the altitude data is all dummy data such as zero.

  The weighted average calculation unit 140 uses an average as the sample processing function f. As an example, the processing function (average processing function f) in the case of the average of k data is as follows.

f (x1, x2, x3,…, xk) = (x1 + x2 + x3 +… + xk) / k
In the present embodiment, under the above assumption,
(1) Weighting process focusing on location only;
(2) Weighting process by gender and age group;
(3) Perform weighting processing by location, age, and gender.

  FIG. 2 is a flowchart of an outline of processing in one embodiment of the present invention.

  The procedure of the outline of processing shown in the figure is the same for the above (1), (2), and (3).

  Step 101) When the user determines which of the evaluation target area and the above (1), (2), and (3) is to be performed, the user inputs via the user interface unit 101. Step 102) i = 1.

  Step 103) The density data extraction unit 130 extracts the density data D of the i-th cell corresponding to the processing target type of the evaluation target area from the density distribution data storage unit 110, and obtains the sum thereof.

  Step 104) If i> K, go to Step 106, and if i ≦ K, go to Step 105.

  Step 105) Set i = i + 1 and return to Step 103.

  Step 106) i = 1.

  Step 107) The measurement data extraction unit 120 extracts the measurement data M of the i-th cell corresponding to the processing type of the evaluation target area from the measurement data DB 10 in the same manner as Step 103, and performs statistical processing.

  Step 108) If i> K, go to Step 110, and if i ≦ K, go to Step 109.

  Step 109) i = i + 1 and return to Step 107.

  Step 110) The weight calculation unit 140 obtains the sum of the density data D for each cell obtained in Step 103 and the measurement data for each cell on which the statistical processing obtained in Step 107 is performed. Calculate the weighted average of the cells.

  Step 111) The output unit 150 outputs the obtained weighted average as a measurement result.

  Below, each process of said (1), (2), (3) is demonstrated in detail.

(1) Weighting process focusing on location only:
This process performs weighting with the user density of only the place.

Here, the processing is performed in units of cells in the evaluation target area, and the density for each processing cell is calculated. The total number of data is k. The coordinate set of measurement data M to be evaluated is
K = {(Lat ₁ , Lon ₁ , Alt ₁ ), (Lat ₂ , Lon ₂ , Alt ₂ ),…, (Lat _k , Lon _k , Alt _k )}
And

  There are cases where the evaluation target area is limited and all areas are targeted. The following example shows a case where the target area is composed of cell numbers 2, 3, and 5.

<Density data extraction for i-th cell>
The processing in step 103 described above is as follows.

The density data extraction unit 130 receives from the density distribution data storage unit 110.
Longitude: "Lat _i or higher && Lat _i + d _{lat lower} "
And,
Longitude: “Lon _i or higher && less than Lon _i + d _lon ”
And,
Elevation: "Alt _i or higher && Alt _i + d _alt less"
The density data D is extracted and the sum is obtained.

The sum of the densities of the i-th cell in the search result (p _x , p _y , ...) is
Di = Σ (p _x , p _y ,…)
It becomes. If the detection result is 0, D _i = 0.

Repeat the above process for all cells, and the density vector
Dv = (D ₁ , D ₂ ,… D _k )
Is generated and stored in a memory (not shown).

<Measurement data extraction of i-th cell>
The processing in step 107 is as follows.

The measurement data extraction unit 120 obtains the longitude: “Lat _i or more && Lat _i + d _lat ” or less from the measurement data DB 10.
And,
Longitude: “Lon _i or higher && less than Lon _i + d _lon ”
And,
Elevation: "Alt _i or higher && Alt _i + d _alt less"
The measurement data M is extracted and its statistical processing (the average processing function f is applied) is performed.

The average value of the i-th cell in the search result (r _i , r _j ,…) is
rmean _i = f (r _i , r _j ,…)
It becomes. If the detection result is 0, rmean _i = 0.

Repeat the above process for all cells and measure
Rv = (rmean ₁ , rmean ₂ ,…, rmean _k )
Is generated and stored in a memory (not shown).

<Calculate the weighted average of the area you want to calculate>
In the following, it is assumed that the evaluation target area is composed of cell numbers 2, 3, and 5 as an example.

The weight calculation unit 140 uses the density vector (Dv) and the measurement vector (Rv) to calculate the average = (D ₂ * rmean ₂ + D ₃ * rmean ₃ + D ₅ * rmean ₅ ) / (D ₂ + D ₃ + D ₅ )
Ask for.

If the area you want to calculate is all areas, the weighted average of all areas (Σ (Dv * Rv) / ΣDv)
Σ (Dv * Rv) / ΣDv = (D ₁ * rmen ₁ + D ₂ * rmean ₂ +… + D _k * rmean _k ) / (D ₁ + D ₂ +… + D _k )
Ask for.

  In step 111, the output unit 150 outputs the weighted average as a measurement result.

(2) User density weighting process in place and time zone:
Next, an example of weight processing for each time zone in addition to position information will be described.

In step 101 of FIG. 2, the place and time zone to be extracted by the user are selected. Here, tz _{3 is} assumed as an example of the time zone. The location is the same as (1) above.

<Extract density data for i-th cell>
The processing in step 103 in FIG. 2 is as follows.

The density data extraction unit 130 receives the longitude: “Lat _i or more && Lat _i + d _lat ” from the density distribution data storage unit 110.
And,
Longitude: “Lon _i or higher && less than Lon _i + d _lon ”
And,
Elevation: "Alt _i or higher && Alt _i + d _alt less"
And,
Time zone: tz ₃
Search and extract the data of, and find the sum.

When the extraction result is (p _x , p _y ,…), the density sum of the i-th cell is
D _i = Σ (p _x , p _y ,…)
It is. If the extraction result is 0, Di = 0.

Repeat the above process for all cells, and the density vector
Dv = (D ₁ , D ₂ ,…, D _k )
Is generated and stored in a memory (not shown).

<Extract measurement data of i-th cell>
The processing in step 107 in FIG. 2 is as follows.

The measurement data extraction unit 120 reads from the measurement data DB 10
Longitude: "Lat _i or higher && Lat _i + d _{lat lower} "
And,
Longitude: “Lon _i or higher && less than Lon _i + d _lon ”
And,
Elevation: "Alt _i or higher && Alt _i + d _alt less"
And,
Time information: (tz _3s (tz ₃ start time) or more && tz _3e (tz ₃ end time)
Search and extract the measurement data M.

  The statistical processing (the average processing function f is applied in this processing).

When the extraction result is (r _i , r _j ,…), the average value of the i-th cell is
rmean _i = f (r _i , r _j ,…)
It is. If the extraction result is 0, rmean _i = 0.

Repeat the above process for all cells and measure
Rv = (rmean ₁ , rmean ₂ ,…, rmean _k )
Is generated and stored in a memory (not shown).

<Calculate the weighted average of the area you want to calculate>
The processing of step 110 in FIG. 2 is as follows.

  Here, as an example, it is assumed that the evaluation target area includes cell numbers 2, 3, and 5.

The weight calculation unit 140 calculates the average of 2, ₃ , and 5 placed in the time zone tz ₃
Average = (D ₂ * rmean ₂ + D ₃ * rmean ₃ + D ₅ * rmean ₅ ) / (D ₂ + D ₃ + D ₅ )
Ask for.

Also, if the evaluation target area to be calculated and all areas are the same, that is, the weighted average Σ (Dv * Rv) / ΣDv of the previous area in the time zone tz3 is
Σ (Dv * Rv) / ΣDv = (D ₁ * rmen ₁ + D ₂ * rmean ₂ +… + D _k * rmean _k ) / D ₁ + D ₂ +… + D _k )
Is required.

  In step 111, the output unit 150 outputs the weighted average as a measurement result.

(3) User density weighting process by location, age, and gender:
This processing shows a case where age and gender are taken into consideration in weighting in addition to the location unit.

In step 101 of FIG. 2, the location, age, and sex to be processed are determined. Here, the extraction target age is assumed to be Zj. That is, the age of the measurement subject is Z _j−1 or more and less than Z _j . The gender is “0” (male). The location is the same as (1) above.

<Density data extraction for i-th cell>
In step 103, the density data extraction unit 130 reads from the density distribution data storage unit 110.
Longitude: "Lat _i or higher && Lat _i + d _{lat lower} "
And,
Longitude: “Lon _i or higher && less than Lon _i + d _lon ”
And,
Elevation: "Alt _i or higher && Alt _i + d _alt less"
And Age: Zj
And gender: 0
The density data D is searched and extracted, and the sum is obtained.

When the extraction result is (p _x , p _y ,…), the density sum of the i-th cell is
D _i = Σ (p _x , p _y ,…)
It is. If the extraction result is 0, Di = 0.

Repeat the above process for all cells, and the density vector
Dv = (D ₁ , D ₂ ,…, D _k )
Is generated and stored in a memory (not shown).

<Measurement data extraction of i-th cell>
The processing in step 107 in FIG. 2 is as follows.

The measurement data extraction unit 120 reads from the measurement data DB 10
Longitude: "Lat _i or higher && Lat _i + d _{lat lower} "
And,
Longitude: “Lon _i or higher && less than Lon _i + d _lon ”
And,
Elevation: "Alt _i or higher && Alt _i + d _alt less"
And,
Age: Zj
And gender: 0
Search and extract the measurement data M.

  The statistical processing (the average processing function f is applied in this processing).

When the extraction result is (r _i , r _j ,…), the average value of the i-th cell is
rmean _i = f (r _i , r _j ,…)
It is. If the extraction result is 0, rmean _i = 0.

Repeat the above process for all cells and measure
Rv = (rmean ₁ , rmean ₂ ,…, rmean _k )
Is generated and stored in a memory (not shown).

<Calculate the weighted average of the area you want to calculate>
The processing of step 110 in FIG. 2 is as follows.

  Here, as an example, it is assumed that the evaluation target area includes cell numbers 2, 3, and 5.

The weighted average calculation unit 140 calculates the average of the cells 2, ₃ , and 5 in the time zone tz3.
Average = (D ₂ * rmean ₂ + D ₃ * rmean ₃ + D ₅ * rmean ₅ ) / (D ₂ + D ₃ + D ₅ )
Ask for.

Also, if the evaluation target area to be calculated and all areas are the same, that is, the weighted average Σ (Dv * Rv) / ΣDv of the previous area in the time zone tz3 is
Σ (Dv * Rv) / ΣDv = (D ₁ * rmean ₁ + D ₂ * rmean ₂ +… + D _k * rmean _k ) / (D ₁ + D ₂ +… + D _k )
Is required.

  In step 111, the output unit 150 outputs the weighted average as a measurement result.

  The operation of each component of the statistical user behavior analysis apparatus 100 shown in FIG. 1 is constructed as a program and installed in a computer used as the statistical user behavior analysis apparatus, or executed via a network. Can be distributed.

  The present invention is not limited to the above-described embodiments, and various modifications and applications are possible within the scope of the claims.

10 Measurement data DB
100 Statistical User Behavior Analysis Device 101 User Interface Unit 110 Density Distribution Data Storage Unit 120 Measurement Data Extraction Unit 130 Density Data Extraction Unit 140 Weighted Average Calculation Unit 150 Output Unit

Claims (7)

A statistical user behavior analysis device that weights measurement data to obtain data measured in a wide area as behavior data of general users,
A measurement data extraction means for acquiring measurement data of a specified item in a certain evaluation target area and performing normalization;
Density distribution data storage means for storing user action density distribution data in advance;
Density data extraction means for acquiring density data of the same items as the measurement data acquired by the measurement data extraction means from the density distribution data storage means;
For the measurement data normalized by the measurement data extraction means, weighted average calculation means for weighting using the density data;
A statistical user behavior analysis apparatus characterized by comprising: The measurement data extraction means includes
The evaluation target area is divided into meshes, measurement data corresponding to items of the divided units (hereinafter referred to as “cells”) is obtained, normalized to the measurement data using an average processing function, and all cells Including means for obtaining an average value (Rv) of
The density data extracting means includes
Means for obtaining the density data of items corresponding to the cells of the measurement data and calculating a sum of density data of all cells;
The weighted average calculating means includes
The statistical user behavior analysis apparatus according to claim 1, further comprising means for calculating a weighted average of all areas from an average value (Rv) of all cells and a sum (Dv) of the density data of all cells. The measurement data extraction means and the density data extraction means are:
Based on a designated item to be evaluated, includes means for acquiring the measurement data and the density data of only position information, a combination of position information and time zone, or a combination of position information and arbitrary attribute information The statistical user behavior analysis apparatus according to claim 1 or 2. A statistical user behavior analysis method that weights measurement data to obtain data measured in a wide area as behavior data of general users,
A measurement data extraction step for obtaining and normalizing measurement data of a specified item in the evaluation target area;
A density data extraction step for acquiring density data of the same items as the measurement data acquired in the measurement data extraction step from the density distribution data storage means that stores the density distribution data of the user behavior in advance;
For the measurement data normalized in the measurement data extraction step, a weight calculation step for weighting using the density data;
Statistical user behavior analysis method characterized by performing. In the measurement data extraction step,
The evaluation target area is divided into meshes, measurement data corresponding to items of the divided units (hereinafter referred to as “cells”) is obtained, normalized to the measurement data using an average processing function, and all cells The average value (Rv) of
In the density data extraction step,
Obtaining the density data of items corresponding to the cells of the measurement data, calculating the sum of the density data of all cells,
In the weight calculation step,
The statistical user behavior analysis method according to claim 4, wherein a weighted average of all areas is calculated from an average value (Rv) of all cells and a sum (Dv) of the density data of all cells. In the measurement data extraction step and the density data extraction step,
5. The measurement data and the density data of only position information, a combination of position information and time zone, or a combination of position information and arbitrary attribute information are acquired based on a specified evaluation target item. Or the statistical user behavior analysis method according to 5. Computer
The statistical user behavior analysis program for functioning as each means of the statistical user behavior analysis device according to any one of claims 1 to 3.

Images