JP2014134895A - Program, terminal device, and data processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To show to a user what kind of communication is performed in a communication system such as an SNS.SOLUTION: A program of the present invention that visualizes the feature of data comprises the steps of: obtaining data labeled on the basis of at least one variable; generating a feature vector by using the variable of the obtained data as a parameter; mapping the generated feature vector in a multi-dimensional space; and displaying the mapped feature vector on a screen.

Description

  The present invention relates to a technique for visualizing data characteristics, and more particularly to a technique for visualizing data characteristics transmitted or received in communication on a network.

  Conventionally, techniques for visualizing communication have been proposed. For example, in the technique described in Patent Document 1, in order to improve productivity within the organization, the face-to-face communication is actually observed, and the communication style of the personal organization belonging to the organization, the communication style of the association organization included in the organization communication style organization Visualized. Specifically, based on interaction data collected from sensors, visualization of communication is realized by plotting individual communication styles on a two-dimensional map.

  Moreover, in patent document 2, the difference of values with a communication partner and the affirmation degree for the conversation topic are estimated by using the strength of agreement or disagreement of a phrase used for replying a conversation, Visualization is realized by converting to display attributes.

  Also, in Patent Document 3, communication performed by a plurality of means such as e-mail logs and conference recordings within or between organizations is recorded, and integrated into a common index called information capture time to realize communication visualization Techniques to do this are disclosed.

  Non-Patent Document 1 discloses a technique for analyzing communication by a mobile phone from a transmission / reception history and visualizing human relationships as a network.

JP 2012-128882 A JP 2006-209332 A JP 2006-127142 A

Eagle N, Pentland A, “Reality Mining: Sensing Complex Social Systems”, J of Personal and Ubiquitous Computing, July 2005 (Eagle, N., and Pentland, A ., "Reality Mining: Sensing Complex Social Systems", J. of Personal and Ubiquitous Computing, July 2005)

  However, since the technology described in Patent Document 1 is based on the premise that communication is actually observable, it is not easy to adapt it to the purpose of visualizing communication features generated on SNS and communication tools. .

  Moreover, since the technique described in Patent Document 2 specializes in evaluating differences in values in communication interaction between users, it is not suitable for visualizing communication characteristics such as SNS.

  Moreover, in the technique described in Patent Document 3, the transition of organizational communication is displayed and diagnosed based on the temporal transition of communication, but it is applicable to analysis including detailed parameters such as communication quality and status. I can't. Furthermore, features from multiple viewpoints cannot be handled simultaneously. This also applies to Non-Patent Document 1.

  Conventionally, since a system for visualizing what functions are used in SNS has not been proposed, it has not been easy for a user to guess in advance an SNS suitable for him.

  The present invention has been made in view of such circumstances, and provides a program, a terminal device, and a data processing method that can indicate to a user what communication is being performed in a communication system such as SNS. The purpose is to do.

  (1) In order to achieve the above object, the present invention takes the following measures. That is, the program of the present invention is a program for visualizing data characteristics, and generates a feature vector using a process for acquiring labeled data based on at least one variable and using the acquired data variable as a parameter. The computer is caused to execute a series of processes including a process, a process of mapping the generated feature vector in a multidimensional space, and a process of displaying the mapped feature vector on a screen.

  In this way, a feature vector is generated using a variable of data labeled based on at least one variable as a parameter, the generated feature vector is mapped to a multidimensional space, and the mapped feature vector is displayed on the screen. It is possible to visualize what characteristics the data has. Thereby, the user can visually recognize what characteristics data is handled in the SNS and the communication service.

  (2) In the program of the present invention, the variable indicates what function on the system the data is handled in, and the feature is based on a system function feature vector that characterizes the function on the system. A vector is generated.

  In this way, the variable is used in communication because it indicates what function on the system the data was handled and generates a feature vector based on the system function feature vector that characterizes the function on the system. It is possible to display on what system function the data was created. For example, when a certain communication service is used as an analysis unit, if a user has posted many comments on a “SYNCHRONOUS” communication tool such as chat, “SYNCHRONUS” on the corresponding service. It is possible to grasp that the service is frequently used. Based on such information, a system function vector for each user can be generated and visualized according to the frequency of variables of each system function feature. At this time, in order to reduce the difference in the number of analysis data, normalization may be performed by the number of all analysis units.

  (3) In the program of the present invention, the variable indicates in what system situation the data is handled, and the feature vector is based on the situation feature vector characterized by the situation on the system. Is generated.

  In this way, the variable is used in communication because it indicates what kind of system situation the data was handled and generates a feature vector based on the situation feature vector that is characterized by the situation on the system. It is possible to display in what situation the data was made. For example, the analysis unit is one communication service. When a group of users has posted many comments on a communication function that cannot be viewed by a third party, such as a private chat, it can be understood that private chat is important as a function that characterizes the communication service. Based on such information, a situation feature vector for each user can be generated and visualized according to the frequency of the labeling results of the variables of each situation feature. At this time, in order to reduce the difference in the number of analysis data, normalization may be performed by the number of all analysis units.

  (4) In the program of the present invention, the variable indicates what kind of user's attitude the data is handled, and generates a feature vector based on a user attitude vector characterized by the user's attitude. It is characterized by that.

  In this way, the variable indicates what kind of user's attitude the data was handled, and the feature vector is generated based on the user attitude vector characterized by the user's attitude. It is possible to display whether such an attitude is suggested. For example, when a user has posted many comments on “THANK” or “GREET”, it can be understood that social actions are being performed on the corresponding communication service (analysis unit). Based on such information, a user attitude vector can be generated and visualized according to the frequency of each user attitude skill vector variable. At this time, in order to reduce the difference in the number of analysis data, normalization may be performed by the number of all analysis units.

  (5) The program according to the present invention is characterized in that an important word included in data is selected and the feature vector is generated based on an important word vector having a word having a high importance as an element.

  In this way, important words included in the data are selected, and feature vectors are generated based on important word vectors whose elements are words of high importance, so the features of data handled in communication are highlighted. It becomes possible to do.

  (6) The program of the present invention further includes a process of inputting data and a process of labeling the input data based on at least one variable.

  As described above, since it further includes a process of inputting data and a process of labeling the input data based on at least one variable, it is possible to automatically label the input data. It becomes.

  (7) Moreover, the terminal device of the present invention is a terminal device that visualizes data characteristics, and includes a data acquisition unit that acquires data that is labeled based on at least one variable, and a variable of the acquired data. A feature vector generation unit that generates a feature vector as a parameter, a mapping unit that maps the generated feature vector to a multidimensional space, and a display unit that displays the mapped feature vector on a screen, To do.

  In this way, a feature vector is generated using a variable of data labeled based on at least one variable as a parameter, the generated feature vector is mapped to a multidimensional space, and the mapped feature vector is displayed on the screen. It is possible to visualize what characteristics the data has. Thereby, the user can visually recognize what characteristics data is handled in the SNS and the communication service.

  (8) Moreover, the terminal device of this invention is characterized by further including the discriminator which labels the input data based on at least 1 variable.

  As described above, since the input data further includes a discriminator that performs labeling based on at least one variable, the input data can be automatically labeled.

  (9) Further, the data processing method of the present invention is a data processing method for visualizing data characteristics, the step of acquiring labeled data based on at least one variable, and the variable of the acquired data. The method includes at least a step of generating a feature vector as a parameter, a step of mapping the generated feature vector in a multidimensional space, and a step of displaying the mapped feature vector on a screen.

  In this way, a feature vector is generated using a variable of data labeled based on at least one variable as a parameter, the generated feature vector is mapped to a multidimensional space, and the mapped feature vector is displayed on the screen. It is possible to visualize what characteristics the data has. Thereby, the user can visually recognize what characteristics data is handled in the SNS and the communication service.

  According to the present invention, it is possible to visualize the characteristics of data. Thereby, the user can visually recognize what characteristics data is handled in the SNS and the communication service.

It is a figure which shows schematic structure of the data processing system which concerns on this embodiment. It is a figure which shows a GUI image. It is a flowchart which shows the operation | movement of coding (labeling). It is a flowchart which shows operation | movement of the feature vector generation module. It is a figure showing the frequency of each system function characteristic variable. It is a flowchart which shows a system function vector extraction process. It is a figure showing the frequency of the labeling result of each situation characteristic variable. It is a flowchart which shows a situation feature vector extraction process. It is a figure showing the frequency of each user attitude skill vector variable. It is a flowchart which shows a user attitude | position vector extraction process. It is a figure which shows a mode that a feature vector is grouped by clustering. It is a figure which shows a mode that the feature vector was visualized hierarchically. It is a flowchart which shows operation | movement of a service feature extraction and presentation module.

  Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing a schematic configuration of a data processing system according to the present embodiment. The data processing system includes a communication data collection module 5, a database 7, a labeling module 9, a feature vector generation module 11, a service feature extraction / presentation module 21, and a display module 23. The communication data collection module 5 collects data from SNS (Social Networking Service) 1 and e-mail / call data 3. For example, data can be collected by crawling using an API (Application Programming Interface). In this case, an API of an Internet radio station or an API of Twitter (registered trademark) can be used.

  For the input, for example, post text information for a certain period of SNS, comment data of multimedia service, electronic mail interaction data, chat data, and the like can be used. Further, when a chat function, a blog function, or the like exists in the same SNS, it can be handled as the same data, or can be divided for each function. The data collected in this way is stored in the database 7.

  The labeling module 9 performs coding (labeling) on the data collected by the communication data collection module 5. This coding provides a GUI (Graphical User Interface) on the WEB and can be performed manually by an operator and stored in the DB. For example, the coding standard can use the communication classification scheme described in Related Document 1.

[Related Literature 1]
Susan C. Herring (2007), A Faceted Classification Scheme for Computer-Mediated Discourse.Language@Internet.http: //www.languageatinternet.org/articles/2007/761
In the present invention, coding is performed based on the following variables.

[System feature variables]
M1 (Synchronicity), M2 (Message transmission), M3 (Persistence of transcript), M4 (Size of message buffer), M5 (Channels of communication), M6 (Anonymous messaging), M7 (Private messaging), M8 (Filtering), M9 (Quoting), M10 (Message format)
In these system characteristic variables, a value can be set for each variable. For example, in the case of M1, a label such as “Synchronous” at the time of 1 and “Asynchronous” at the time of 2 can be given.

[Situation feature variable]
S1 (Participation Structure), S2 (Participant characteristics), S3 (Purpose), S4 (Topic or Theme), S5 (Tone), S6 (Activity), S7 (Norms), S8 (Code)
These situation feature variables can be free descriptions or can be given options in advance.

  FIG. 2 is a diagram illustrating a GUI image. For example, as shown in FIG. 2, communication data is displayed on the left side of the screen, and coding results can be input on the right side. The input method can be input with a check box or selected with a touch panel. The coding itself can be performed by multiple people. In that case, it is possible to compare a plurality of results, and to re-present to the coder the result coincidence rate and the different results.

  In addition to the above variables, scene variables, user attitude variables, and the like can be freely set as variables. For example, a season variable, a season, a time zone, a weather, etc. can be utilized. For example, as the user attitude variable, a speaker's attitude (presentation / approval, rejection) or the like can be used. For example, it can be set as follows using the technique of Related Document 2 below.

[Related Literature 2]
Herring, SC, Das, A., & Penumarthy, S. (2005). CMC act taxonomy. Http://www.slis.indiana.edu/faculty/herring/cmc.acts.html
A1 (Inquire), A2 (Request), A3 (Invite), A4 (Desire), A5 (React), A6 (Manage), A7 (Direct), A8 (Accept), A9 (Apologize), A10 (Repair), A11 (Reject), A12 (Elaborate), A13 (Thank), A14 (Inform), A15 (Claim), A16 (Greet)
The labeling module 9 can be provided with a function as a discriminator such as SVM (Support vector machine) so that coding can be performed automatically. When coding is automatically performed, prior information registered in the system in advance can be used, or learning data can be collected and coded in advance, and automatic labeling can be performed by a discriminator. For example, when the information obtained by SNS (A) is fixed from M1 to M10, similar information can be automatically given to other data obtained by SNS (A).

  In addition, labeling can be automatically applied to S1-S8 and the like. For example, when a coding result for a large amount of communication data can be accumulated as learning data, a label can be automatically given by a discriminator. For example, the communication data to which the coding result S1 is given is converted into a feature vector by TFIDF, and a label can be automatically given by determining whether S1 is correct or not by SVM.

  FIG. 3 is a flowchart showing an operation of coding (labeling). First, data is acquired via the communication data collection module 5 (step S1), and it is determined whether there is a discriminator (step S2). If there is a discriminator, the process proceeds to step S7. On the other hand, if there is no discriminator in step S2, it is determined whether there is label data (step S3). If there is label data, the process proceeds to step S6. If there is no label data, the label data is acquired (step S4), displayed on the GUI (step S5), and a discriminator is generated based on the data labeled by the operator (step S6).

  Here, for example, in the case of discriminating whether or not the label A is using the “Support Vector Machine”, the discriminator extracts an important word from the text data group to which the label A is given, and determines the frequency of the important word. Based on the feature vector (for example, Bag of Words based on TF / IDF). Whether or not it is label A by using the learning data group to which the label A is assigned and its feature vector group as positive data, the learning data to which the label data A is not assigned and its feature vector group as negative data for learning Discriminators can be generated. Then, a label is given (step S7), and the process ends.

  In FIG. 1, the feature vector generation module 11 includes a situation feature skill extraction function 13, a system function skill extraction function 15, a user attitude skill extraction function 17, and an important word extraction function 19. Based on the data, the communication data is featured. For example, the input coding result can be expressed as a multidimensional vector using the input value of each variable as a parameter. For example, system function features, situation features, and user attitude features can be extracted as vectors. Also, these vectors can be combined and extracted as a feature vector. Further, communication data is selected by selecting an important word from the accumulated communication data based on the TFIDF method, and calculating the frequency of the word as each element in the data with a word having high importance as a vector element. Can be vectorized.

  FIG. 4 is a flowchart showing the operation of the feature vector generation module. First, a coding result and communication data are input (step T1). Next, a system function feature vector is extracted (step T2). Next, a situation feature vector is extracted (step T3). Next, a user attitude vector is extracted (step T4). Next, an important word vector is extracted (step T5). Then, the extracted vectors are integrated (step T6), and the process ends. Next, processing for extracting a vector as described above will be described.

[System function vector extraction processing]
Based on the label data assigned to the text obtained from each category, what kind of system function the communication data is performed on is extracted as a feature. For example, when a certain communication service is used as an analysis unit, if a user has posted many comments on a “SYNCHRONOUS” communication tool (such as chat), “SYNCHRONUS” on the corresponding service. It can be understood that the service is often used. Based on this information, a user system function vector is generated.

  FIG. 5 is a diagram showing the frequency of each system function characteristic variable. When visualizing in this way, in order to reduce the difference in the number of analysis data, normalization can be performed for all the number of analysis units. Further, it is not always necessary to execute processing in service units, and macro communication data generated in a certain service may be used as one analysis unit.

  FIG. 6 is a flowchart showing system function vector extraction processing. First, label data is acquired (step P1), and the number of elements is counted (step P2). Next, it is determined whether all system function variables have been counted (step P3). If all system function variables have not been counted, the process proceeds to step P2. When all system function variables are counted, normalization is performed for all the number of analysis units (step P4), and a vector is generated (step P5).

[Situation feature vector extraction processing]
Based on the label data assigned to the text obtained from each category, the situation in which the communication is performed is extracted as a feature. For example, consider the case where the analysis unit is one communication service. When a group of users has posted many comments on communication functions (private chat etc.) that cannot be viewed by third parties, it can be understood that private chat is important as a function that characterizes communication services. . Based on this information, a situation feature vector of the user is generated.

  FIG. 7 is a diagram showing the frequency of the labeling result of each situation feature variable. When visualizing in this way, in order to reduce the difference in the number of analysis data, normalization can be performed for all the number of analysis units. For example, for items such as S2 for which elements cannot be set in advance, the age and gender distribution of the SNS / community used by the user can be extracted in advance and registered as a pattern. For example, a case where the community at SNS1 is a woman and a community only in their 30s is registered as a pattern PAT1. : 3: 4 can be registered as pattern 2 to calculate the frequency.

  FIG. 8 is a flowchart showing the situation feature vector extraction process. First, label data is acquired (step Q1), and a pattern is registered (step Q2). Next, the number of elements is counted (step Q3), and it is determined whether all the status variables have been counted (step Q4). If all the status variables are not counted, the process proceeds to step Q3. On the other hand, if all the system function variables are counted, normalization is performed with all the number of analysis units (step Q5), and the vector is changed. Generate (step Q6).

[User attitude vector extraction processing]
Based on the label data given to the text obtained from each category, what kind of attitude the text posted by the user suggests is extracted as a feature. For example, when a user has posted many comments on “THANK” or “GREET”, it can be understood that social actions are being performed on the corresponding communication service (analysis unit). Based on this information, a user attitude vector is generated.

  FIG. 9 is a diagram showing the frequency of each user attitude skill vector variable. When visualizing in this way, in order to reduce the difference in the number of analysis data, normalization can be performed for all the number of analysis units.

  FIG. 10 is a flowchart showing the user attitude vector extraction process. First, label data is acquired (step R1), and the number of elements is counted (step R2). Next, it is determined whether all user attitude variables have been counted (step R3). If all user attitude variables have not been counted, the process proceeds to step R2. When all the user attitude variables are counted, normalization is performed with all the analysis unit numbers (step R4), and a vector is generated (step R5).

  In FIG. 1, the service feature extraction / presentation module 21 performs mapping on a multidimensional space based on the feature vector created by the feature vector generation module 11. For example, it can be expressed as a plot on a two-dimensional plane by applying principal component analysis and using the first principal component and the second principal component as expression axes. In addition, three-dimensional display is possible by adding the third main component. In addition, visualization techniques such as SOM (Self-Organizing Map) can be used.

  FIG. 11 is a diagram illustrating a state in which feature vectors are grouped by clustering. A plurality of groups 101 are formed within the range of the maximum grouping frame 100. Each group 101 includes at least one feature vector 102. Further, like the feature vector 102, feature vectors having different features belong to different groups depending on the classification criteria. As shown in FIG. 11, the communication data expressed by the feature vector can be grouped and displayed by further applying a clustering method. The clustering method can be classified by, for example, the K-means method. By clicking the plot on the screen, the classification result of the communication data on each SNS can be viewed. Note that this processing can also be performed before the principal component analysis is applied.

  FIG. 12 is a diagram illustrating a state in which feature vectors are visualized hierarchically. In FIG. 12, a plurality of groups 206, 207, and 220 are formed in the highest hierarchy 201 within the range of the maximum grouping frame 200. The uppermost groups 206, 207, and 220 have second-level groups 202, 203, and 204, respectively. The second layer group 202 includes feature vectors 208 and 209, and the feature vectors 208 and 209 have third layer groups 210 and 211. The same applies to the groups 203 and 204 in the second hierarchy. That is, the second layer groups 214 and 215 have third layer groups 212 and 213, respectively. Further, the second layer groups 218 and 219 have third layer groups 216 and 217, respectively.

  As shown in FIG. 12, the service feature extraction / presentation module 21 can also hierarchically visualize feature vectors based on system feature variables, feature vectors based on situation feature variables, and feature vectors based on key words of communication data. is there. In FIG. 12, the space based on the system feature variables is the first space, but the order is variable. Further, for feature vectors based on important words of communication data, representative important words can be displayed on the space based on the result of clustering.

  FIG. 13 is a flowchart showing the operation of the service feature extraction / presentation module. First, a feature vector is extracted (step V1), and it is determined whether or not to perform dimension compression in the K-means method (step V2). If dimensional compression is not performed, the process proceeds to step V5. On the other hand, if dimensional compression is performed, principal component analysis is performed (step V3), and the Nth principal component is extracted (step V4). Next, clustering is performed (step V5), and display processing is performed to visualize the clustering result (step V6). Next, it is determined whether or not all feature vectors have been processed (step V7). If all feature vectors have not been processed, the process proceeds to step V1. On the other hand, when all the feature vectors have been processed, representative important words are extracted (step V8), and a feature vector based on the extracted important words is displayed on the space (step V9), and the process ends. .

  As described above, according to the present embodiment, the communication status can be visualized even for an SNS in which it is difficult to actually acquire data, so what kind of communication is performed on the SNS. It is possible to provide the user with information indicating whether or not. As a result, the user can easily select an SNS suitable for the user. Further, according to the present embodiment, feature quantities (system variables, situation variables, communication variables, etc.) from a plurality of viewpoints that could not be realized by the conventional technology can be displayed in a unified manner.

5 Communication Data Collection Module 7 Database 9 Labeling Module 11 Feature Vector Generation Module 13 Situation Feature Skill Extraction Function 15 System Function Skill Extraction Function 17 User Attitude Skill Extraction Function 19 Key Word Extraction Function 21 Service Feature Extraction / Presentation Module 23 Display Module

Claims (9)

A program for visualizing the characteristics of data,
Processing to obtain labeled data based on at least one variable;
Processing to generate a feature vector using the acquired data variable as a parameter;
Mapping the generated feature vector into a multidimensional space;
A program for causing a computer to execute a series of processes of displaying the mapped feature vector on a screen.   The variable represents the function on which the data is handled by the system, and the feature vector is generated based on a system function feature vector characterized by the function on the system. 1. The program according to 1.   The variable represents the situation on the system in which data is handled, and the feature vector is generated based on a situation feature vector characterized by the situation on the system. The listed program.   2. The program according to claim 1, wherein the variable indicates a user's attitude in which data is handled, and a feature vector is generated based on a user attitude vector characterized by the user's attitude. .   5. The feature vector according to claim 2, wherein an important word included in the data is selected, and the feature vector is generated based on an important word vector including a word having high importance as an element. program. The process of entering data,
The program according to claim 1, further comprising: a process of labeling the input data based on at least one variable. A terminal device for visualizing data characteristics,
A data acquisition unit for acquiring labeled data based on at least one variable;
A feature vector generation unit that generates a feature vector using the acquired data variable as a parameter;
A mapping unit for mapping the generated feature vector to a multidimensional space;
And a display unit that displays the mapped feature vector on a screen.   8. The terminal device according to claim 7, further comprising a discriminator that performs labeling on the input data based on at least one variable. A data processing method for visualizing data characteristics,
Obtaining labeled data based on at least one variable;
Generating a feature vector using the acquired data variable as a parameter;
Mapping the generated feature vector to a multidimensional space;
Displaying at least one of the mapped feature vectors on a screen.

Images