JP2018152023A - Text mining support method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently perform processing for guiding findings from a graph showing a result of correspondence analysis.SOLUTION: When a scatter diagram showing a result of correspondence analysis is displayed, a support screen in which the scatter diagram and a hint showing the way of viewing the scatter diagram are included is displayed. When a scatter diagram related to words and variables is displayed, a screen that a user instructs is displayed from among a basic screen in which no hint is included, a first support screen in which a determination method of the words in the vicinity of an original point is included as a hint, a second support screen in which a determination method of a degree of association of the words characterizing the variables is included as a hint, a third support screen in which a determination method of a degree of association of the words is included as a hint, and a fourth support screen in which a determination method of a degree of association of the variables is included as a hint.SELECTED DRAWING: Figure 7

Description

  The present invention relates to a data mining technique, and more particularly to a text mining support method and apparatus for supporting execution of text mining.

  In recent years, data mining technology that applies data analysis techniques such as statistics and pattern recognition to a large amount of data to derive knowledge (such as rules appearing in the data) from a large amount of data has attracted attention. Data mining for text data is called text mining. Hereinafter, a case where correspondence analysis (correspondence analysis), which is a kind of data analysis technology, is performed on text data will be considered.

  In the correspondence analysis, a process of rearranging each item so that the correlation between the head item and the front item is maximized with respect to the cross tabulation table is performed. The result of the correspondence analysis is generally expressed using a scatter diagram (two-dimensional graph). For example, when correspondence analysis is performed on the cross tabulation table shown in FIG. 2, a scatter diagram shown in FIG. 3 is obtained.

  In relation to the present invention, Patent Document 1 describes a text mining system that presents a user with an analysis procedure when using a plurality of analysis tools. If the system described in Patent Document 1 is used, even a user who has little knowledge and experience regarding text mining can perform analysis using a plurality of analysis tools in a suitable order.

JP-A-2005-44087

  In correspondence analysis, it is more important to consider the derived scatterplot and derive knowledge than to obtain the scatterplot. However, users who have little knowledge and experience regarding text mining do not know how to view the scatter diagram, and therefore do not know what to do first by looking at the scatter diagram. For this reason, a user with little knowledge and experience cannot efficiently perform the process of deriving knowledge from the scatter diagram.

  The system described in Patent Document 1 presents an analysis procedure to the user, but does not support the process of deriving knowledge from the analysis result. For this reason, even if it uses the system described in patent document 1, said subject cannot be solved.

  Therefore, an object of the present invention is to provide a text mining support method and apparatus for efficiently performing processing for deriving knowledge from a graph showing the result of correspondence analysis.

A first aspect of the present invention is a text mining support method for displaying an analysis result by correspondence analysis,
Inputting the analysis result;
Inputting instructions from the user;
Generating screen data of a screen including a graph indicating the analysis result;
And displaying a screen based on the screen data,
The step of generating the screen data generates screen data of a support screen including the graph and a hint indicating how to read the graph in response to the instruction.

According to a second aspect of the present invention, in the first aspect of the present invention,
The step of generating the screen data includes generating screen data of a screen selected according to the instruction from a plurality of support screens and a basic screen including the graph and not including the hint. To do.

According to a third aspect of the present invention, in the second aspect of the present invention,
The step of inputting the analysis result is a result of associating the first item and the second item as the analysis result, wherein the first component and the second component of the first item, and the second item A result including a first component and a second component is input;
The step of generating the screen data creates, as the graph, a scatter diagram in which the first item and the second item are plotted in a plane with the first component as the horizontal axis and the second component as the vertical axis. It is characterized by doing.

According to a fourth aspect of the present invention, in the third aspect of the present invention,
The plurality of support screens include a first support screen including the hint that the first item near the origin in the scatter diagram does not have a remarkable feature.

According to a fifth aspect of the present invention, in the fourth aspect of the present invention,
In the scatter diagram included in the first support screen, a range near the origin is illustrated.

According to a sixth aspect of the present invention, in the third aspect of the present invention,
The plurality of support screens include a second support screen including, as the hint, the first item in a direction away from the origin toward the second item in the scatter diagram characterizes the second item. To do.

A seventh aspect of the present invention is the sixth aspect of the present invention,
In the scatter diagram included in the second support screen, a range in a direction away from the origin toward the second item selected is illustrated.

According to an eighth aspect of the present invention, in the third aspect of the present invention,
The plurality of support screens include a third support screen including, as the hint, that the first items that are close to each other in the scatter diagram have high similarity.

A ninth aspect of the present invention is the eighth aspect of the present invention,
In the scatter diagram included in the third support screen, a range in the vicinity of the selected first item is illustrated.

According to a tenth aspect of the present invention, in the third aspect of the present invention,
The plurality of support screens include a fourth support screen including, as the hint, that the second items that are close to each other in the scatter diagram have high similarity.

An eleventh aspect of the present invention is the tenth aspect of the present invention,
In the scatter diagram included in the fourth support screen, a mark indicating the second item closest to the selected second item is shown.

A twelfth aspect of the present invention is the third aspect of the present invention,
In the step of inputting the analysis result, the analysis result is a cross tabulation table in which the word is the first item, the sentence part is the second item, and the appearance frequency of each word in each part of the sentence is in-table data. The result of the corresponding analysis is input.

A thirteenth aspect of the present invention is a text mining support device for displaying an analysis result by correspondence analysis,
An analysis result input unit for inputting the analysis result;
An instruction input unit for inputting instructions from the user;
A screen generator that generates screen data of a screen including a graph indicating the analysis result;
An analysis result display unit for displaying a screen based on the screen data;
The screen generation unit generates screen data of a support screen including the graph and a hint indicating how to read the graph in response to the instruction.

A fourteenth aspect of the present invention is the thirteenth aspect of the present invention,
The screen generation unit generates screen data of a screen selected according to the instruction from a plurality of support screens and a basic screen including the graph and not including the hint.

A fifteenth aspect of the present invention is the fourteenth aspect of the present invention,
The analysis result input unit is a result of associating the first item and the second item as the analysis result, and includes the first component and the second component of the first item, and the second item of the second item. A result including a first component and a second component is input;
The screen generation unit creates, as the graph, a scatter diagram in which the first item and the second item are plotted in a plane having the first component as a horizontal axis and the second component as a vertical axis. Features.

A sixteenth aspect of the present invention is the fifteenth aspect of the present invention,
In the analysis result input unit, as the analysis result, a word is the first item, a sentence part is the second item, and a frequency of each word in each part of the sentence is an in-table data. The result of the correspondence analysis is input.

  According to the first or thirteenth aspect, the user derives knowledge from the graph indicating the result of the correspondence analysis using the support screen including the graph indicating the result of the correspondence analysis and the hint indicating how to read the graph. Processing can be performed efficiently.

  According to the second or fourteenth aspect, by selectively displaying the support screen including the hint and the basic screen not including the hint, a screen according to the level of the user can be displayed. In addition, by selectively displaying a plurality of support screens, the user can be presented with a plurality of ways of viewing the graph.

  According to the third or fifteenth aspect, the user can efficiently perform the process of deriving knowledge from the scatter diagram showing the result of the correspondence analysis regarding the first item and the second item.

  According to the fourth aspect, the user efficiently uses the knowledge that the first item in the vicinity of the origin in the scatter diagram does not have a prominent feature to efficiently perform the process of deriving knowledge from the graph indicating the result of the correspondence analysis. Can be done automatically.

  According to the fifth aspect, the user can easily know the first item having no remarkable features by looking at the illustrated range.

  According to the sixth aspect, the user uses the knowledge that the first item in the direction away from the origin toward the second item in the scatter diagram characterizes the second item, and results of the correspondence analysis. Thus, it is possible to efficiently perform the process of deriving knowledge from the graph indicating the above.

  According to the seventh aspect, the user can easily know the first item characterizing the selected second item by looking at the illustrated range.

  According to the eighth aspect, the user efficiently uses the knowledge that the first items that are close to each other in the scatter diagram have a high degree of similarity to perform the process of deriving knowledge from the graph indicating the result of the correspondence analysis. Can be done.

  According to the ninth aspect, the user can easily know the first item having a high degree of similarity to the selected first item by looking at the illustrated range.

  According to the tenth aspect, the user efficiently uses the knowledge that the second items that are close to each other in the scatter diagram have a high degree of similarity to perform the process of deriving knowledge from the graph indicating the result of the correspondence analysis. Can be done.

  According to the eleventh aspect, the user can easily know the second item having the highest degree of similarity with the selected second item by looking at the illustrated mark.

  According to the twelfth or eighteenth aspect, the user can efficiently perform the process of deriving knowledge from the scatter diagram showing the result of the correspondence analysis regarding the word and sentence portions.

It is a block diagram which shows the structure of the text mining assistance apparatus which concerns on embodiment of this invention. It is a figure which shows the cross tabulation table | surface used as the object of correspondence analysis. It is a figure which shows the scatter diagram produced with the text mining assistance apparatus shown in FIG. It is a block diagram which shows the structure of the computer which functions as a text mining assistance apparatus shown in FIG. It is a flowchart which shows operation | movement of the text mining assistance apparatus shown in FIG. It is a figure which shows the basic screen of the text mining assistance apparatus shown in FIG. It is a figure which shows the 1st assistance screen of the text mining assistance apparatus shown in FIG. It is a figure which shows the 2nd assistance screen of the text mining assistance apparatus shown in FIG. It is a figure which shows the 3rd assistance screen of the text mining assistance apparatus shown in FIG. It is a figure which shows the 4th assistance screen of the text mining assistance apparatus shown in FIG.

  Hereinafter, a text mining support method, a text mining support device, and a text mining support program according to an embodiment of the present invention will be described with reference to the drawings. The text mining support method according to the present embodiment is typically executed using a computer. The text mining support device according to the present embodiment is typically configured using a computer. The text mining support program according to the present embodiment is a program for implementing the text mining support method using a computer. A computer that executes the text mining support program functions as a text mining support device.

  FIG. 1 is a block diagram showing a configuration of a text mining support apparatus according to an embodiment of the present invention. A text mining support apparatus 10 shown in FIG. 1 includes an analysis result input unit 11, an instruction input unit 12, a screen generation unit 13, and an analysis result display unit 14. The text mining support device 10 receives the result of the correspondence analysis performed on the text data. The text mining support device 10 displays a scatter diagram showing the input analysis result on the screen.

  In FIG. 1, a text analysis device 5 is provided in front of the text mining support device 10. Text data 1 is input to the text analysis device 5. In the following description, it is assumed that the text data 1 is text data having a plurality of parts (hereinafter referred to as “chapter”). Also, in the scene where correspondence analysis is performed, “chapter” is also referred to as “variable”. The text analysis device 5 extracts words included in the text data 1 and creates a cross tabulation table with the words as front-side items, chapters as head items, and the appearance frequency of each word in each chapter as in-table data. The text analysis device 5 performs correspondence analysis on the created cross tabulation table and outputs an analysis result 2. When the correspondence analysis is performed, two or more components representing the characteristics of the processing target data are obtained. The analysis result 2 includes at least the first and second components of each word, the first and second components of each variable, the contribution rate of the first component, and the contribution rate of the second component.

  FIG. 2 is a diagram illustrating a cross tabulation table to be subjected to correspondence analysis. The cross tabulation table shown in FIG. 2 is created by inputting the text data of the novel “personal disqualification” as text data 1 into the text analysis device 5. This novel has five chapters: “Foreword”, “First Note”, “Second Note”, “Third Note”, and “Afterword”, “Self”, “Human”, “Flounder”, “ It includes words such as “feeling”. The cross tabulation table shown in FIG. 2 includes words such as “self”, “human”, “flounder”, and “feeling” as front side items, and “Foreword”, “first note”, and “second note” as front items. It includes five variables (chapters), “Third note” and “Afterword”. In the “first note”, the word “human” appears 38 times. Correspondingly, in the cross tabulation table shown in FIG. 2, “38” is written in the column (hatched portion) of the front item “Human” and the front item “First Manual”. In order to suitably perform the correspondence analysis, the cross tabulation table shown in FIG. 2 includes only words having an appearance frequency equal to or higher than a predetermined value.

  FIG. 3 is a diagram illustrating a scatter diagram created by the text mining support apparatus 10. As described above, the analysis result 2 input to the text mining support device 10 includes at least the first and second components of each word, the first and second components of each variable, the contribution ratio of the first component, and , The contribution ratio of the second component is included. The screen generation unit 13 creates a scatter plot by plotting words and variables in a plane with the first component as the horizontal axis and the second component as the vertical axis. For example, the scatter diagram shown in FIG. 3 is created based on the analysis result 2 for the cross tabulation table shown in FIG. The analysis result display unit 14 displays a screen including the created scatter diagram.

In FIG. 3, black circles are written at word positions, white squares are written at variable positions, words are written in standard font, and variables are written in italic font. FIG. 3 shows the contribution ratio of the first component and the contribution ratio of the second component. In general, the contribution ratio of the first component is larger than the contribution ratio of the second component. Considering this point, the distance d between the two points P (p ₁ , p ₂ ) and Q (q ₁ , q ₂ ) in the scatter diagram is the contribution ratio k ₁ of the first component and the contribution of the second component. It is defined as the following equation (1) using the rate k ₂ .
d = √ [{k ₁ (p ₁ −q ₁ )} ² + {k ₂ (p ₂ −q ₂ )} ² ] (1)
The distance in the following description refers to the distance in the scatter diagram defined by the equation (1). The circle described in the scatter diagram looks like an ellipse whose length in the first component direction is shorter than the length in the second component direction.

  FIG. 4 is a block diagram illustrating a configuration of a computer that functions as the text mining support device 10. A computer 20 shown in FIG. 4 includes a CPU 21, a main memory 22, a storage unit 23, an input unit 24, a display unit 25, a communication unit 26, and a recording medium reading unit 27. For example, a DRAM is used as the main memory 22. For the storage unit 23, for example, a hard disk or a solid state drive is used. The input unit 24 includes a keyboard 28 and a mouse 29, for example. For example, a liquid crystal display is used for the display unit 25. The communication unit 26 is an interface circuit for wired communication or wireless communication. The recording medium reading unit 27 is an interface circuit of the recording medium 30 that stores programs and the like. As the recording medium 30, for example, a non-transitory recording medium such as a CD-ROM or a DVD-ROM is used.

  When the computer 20 executes the text mining support program 31, the storage unit 23 stores the text mining support program 31 and the analysis result 2. The text mining support program 31 and the analysis result 2 may be received from the server or another computer using the communication unit 26, or read from the recording medium 30 using the recording medium reading unit 27, for example.

  When the text mining support program 31 is executed, the text mining support program 31 and the analysis result 2 are copied and transferred to the main memory 22. The CPU 21 processes the analysis result 2 stored in the main memory 22 by executing the text mining support program 31 stored in the main memory 22 by using the main memory 22 as a working memory. At this time, the computer 20 functions as the text mining support device 10. The configuration of the computer 20 described above is merely an example, and the text mining support device 10 can be configured using an arbitrary computer.

A user who has knowledge and experience regarding text mining has the following knowledge about a scatter diagram showing the result of correspondence analysis. Users with knowledge and experience can use this knowledge to derive knowledge from the scatter plot.
First knowledge “words near the origin do not have significant features.”
Second knowledge “words in a direction away from the origin toward the variable have a high degree of association with the variable and characterize the variable.”
Third knowledge “words close to each other have high similarity”
Fourth knowledge “variables with close distances have high similarity.”

  On the other hand, a user who has little knowledge and experience regarding text mining does not have such knowledge. For this reason, a user with little knowledge and experience cannot efficiently perform the process of deriving knowledge from the scatter diagram. In order to solve this problem, the text mining support device 10 not only displays a screen including a scatter diagram as a basic screen, but also a hint indicating how to read the scatter diagram and the scatter diagram according to an instruction from the user. Display a screen that contains as a support screen.

  With reference to FIG. 1, operation | movement of each part of the text mining assistance apparatus 10 is demonstrated. The analysis result input unit 11 receives the analysis result 2 output from an external device (for example, the text analysis device 5). The instruction input unit 12 receives an instruction from the user. The screen generation unit 13 creates a scatter diagram showing the analysis result 2 and generates screen data of a screen including the scatter diagram. In response to an instruction from the user input using the instruction input unit 12, the screen generation unit 13 includes screen data of a support screen including a scatter diagram and a hint, and a screen of a basic screen including a scatter diagram and not including a hint Selectively create data. The analysis result display unit 14 displays a screen based on the screen data generated by the screen generation unit 13. Hereinafter, there are four types of support screens displayed on the text mining support device 10, and the four types of support screens are referred to as first to fourth support screens.

  FIG. 5 is a flowchart showing the operation of the text mining support apparatus 10. First, the CPU 21 transfers the analysis result 2 output from the text analysis device 5 to the main memory 22. Thereby, the analysis result 2 is input to the text mining support device 10 (step S101). Next, the CPU 21 creates a scatter diagram based on the analysis result 2 (step S102). A scatter diagram is created by plotting words and variables in a plane with the first component on the horizontal axis and the second component on the vertical axis. Next, the CPU 21 creates screen data of a basic screen including the scatter diagram created in step S102 (step S103). Next, the CPU 21 displays a basic screen on the display unit 25 based on the screen data created in step S103 (step S104).

  FIG. 6 is a diagram showing a basic screen. A basic screen 100 shown in FIG. 6 includes a screen selection window 101 and a scatter diagram window 102. The scatter diagram shown in FIG. 3 is written in the scatter diagram window 102. The screen selection window 101 has six radio buttons 103. Hereinafter, the six radio buttons 103 are referred to as first to sixth radio buttons. The first to sixth radio buttons are associated with the basic screen, the first to fourth support screens, and the end, respectively. When the basic screen 100 is displayed, the user operates the keyboard 28 or the mouse 29 to press one of the first to sixth radio buttons. Thereby, an instruction from the user is input.

  The CPU 21 receives an instruction from the user input using the screen selection window 101 (step S105). Next, the CPU 21 proceeds to one of the following steps in accordance with an instruction from the user (step S106). When the instruction from the user is “basic screen” (when the first radio button is pressed), the CPU 21 proceeds to step S107. In this case, the CPU 21 generates screen data of the basic screen as in step S103 (step S107). When the instruction from the user is the “first support screen” (when the second radio button is pressed), the CPU 21 proceeds to step S108. In this case, the CPU 21 generates screen data of the first support screen (step S108). When the instruction from the user is the “second support screen” (when the third radio button is pressed), the CPU 21 proceeds to step S109. In this case, the CPU 21 generates screen data for the second support screen (step S109). When the instruction from the user is the “third support screen” (when the fourth radio button is pressed), the CPU 21 proceeds to step S110. In this case, the CPU 21 generates screen data for the third support screen (step S110). When the instruction from the user is the “fourth support screen” (when the fifth radio button is pressed), the CPU 21 proceeds to step S111. In this case, the CPU 21 generates screen data for the fourth support screen (step S111). When the instruction from the user is “end” (when the sixth radio button is pressed), the CPU 21 ends the process.

  After executing any of steps S107 to S111, the CPU 21 proceeds to step S112. Next, the CPU 21 displays a screen on the display unit 25 based on the screen data created in any of steps S107 to S111 (step S112). Next, the CPU 21 proceeds to step S105. As described above, the text mining support device 10 displays a screen selected from the basic screen and the first to fourth support screens in accordance with an instruction from the user.

  The components of the computer 20 shown in FIG. 4 and the steps shown in FIG. 5 correspond to the components of the text mining support apparatus 10 shown in FIG. 1 as follows. The CPU 21 that executes step S <b> 101 functions as the analysis result input unit 11. The CPU 21 that executes the input unit 24 and step S105 functions as the instruction input unit 12. The CPU 21 that executes steps S102 to S103 and S106 to S111 functions as the screen generation unit 13. The display unit 25 and the CPU 21 that executes steps S104 and S112 function as the analysis result display unit 14.

  FIG. 7 is a diagram illustrating the first support screen. The first support screen 110 shown in FIG. 7 includes a screen selection window 101, a scatter diagram window 112, a word list window 113, and a hint window 114. The first support screen 110 relates to the first knowledge “words near the origin do not have a prominent feature”. The user can efficiently perform the process of deriving knowledge from the scatter diagram using the first knowledge by looking at the first support screen 110.

  Before the first support screen 110 is displayed, the user operates the keyboard 28 or the mouse 29 to designate a range determined to be near the origin. The initial value of the range determined to be near the origin may be determined in advance. In the scatter diagram window 112, the scatter diagram shown in FIG. 3 is described. In the scatter diagram in the scatter diagram window 112, a circle 115 (appearance is an ellipse) indicating the vicinity of the origin is described. The circle 115 is preferably described in a different color (for example, red) from the scatter diagram. Thus, in the scatter diagram included in the first support screen 110, the range near the origin is illustrated using the circle 115. Therefore, the user can easily know a word that does not have a remarkable feature by looking at the illustrated range.

  In the word list window 113, a word list in which words near the origin (words in the circle 115) and distances from the origin of the words are arranged in order of increasing distance. The upward triangles in the word list window 113 indicate that they are arranged in order of increasing distance. In the hint window 114, the first knowledge is described with the title "point of analysis". The hint window 114 is arranged at a position overlapping the scatter diagram window 112.

  The size of the circle 115 is determined by an arbitrary method. For example, the size of the circle 115 may be determined by designating the number of words (for example, 10) included in the circle 115 by the user. Alternatively, the size of the circle 115 may be determined by the user specifying the ratio of words included in the circle 115 (for example, 10% of the whole). Alternatively, the user may determine the size of the circle 115 by specifying the distance from the origin using the mouse 29 in the first support screen 110.

  On the first support screen 110 shown in FIG. 7, words near the origin (words in the circle 115) are displayed in the same manner as other words. Instead, on the first support screen, a word near the origin may be displayed in a different manner from other words (for example, in a light color), or a word near the origin may not be displayed. In the second to fourth support screens, the words displayed in a different mode from the other words on the first support screen may be displayed in a mode different from the other words, and the words not displayed on the first support screen are displayed. You don't have to.

  FIG. 8 is a diagram illustrating the second support screen. The second support screen 120 illustrated in FIG. 8 includes a screen selection window 101, a scatter diagram window 122, a word list window 123, and a hint window 124. The second support screen 120 relates to the second knowledge “words in a direction away from the origin toward the variable have a high degree of association with the variable and characterize the variable”. The user can efficiently perform the process of deriving knowledge from the scatter diagram using the second knowledge by looking at the second support screen 120.

  Before the second support screen 120 is displayed, the user operates the keyboard 28 or the mouse 29 to select one variable (chapter). Here, a case where the variable “Foreword” is selected will be described. In the scatter diagram window 122, the scatter diagram shown in FIG. 3 is described. The scatter diagram in the scatter diagram window 122 forms an angle of a predetermined angle (for example, 10 °) between the arrow 125 that starts from the origin and passes through the selected variable, and the arrow 125 that starts from the origin. Two half lines 126, 127 are described. Within the region sandwiched between the half lines 126 and 127, there are words in a direction away from the origin toward the selected variable. Thus, in the scatter diagram included in the second support screen 120, the range in the direction away from the origin toward the selected variable is illustrated using the half lines 126 and 127. Therefore, the user can easily know the word characterizing the selected variable by looking at the range shown.

  In the word list window 123, a distance between a word in a direction away from the origin toward the selected variable (a word in an area between the half lines 126 and 127) and the distance from the origin of the word is long. An ordered word list is listed. The downward triangles in the word list window 123 indicate that they are arranged in order of increasing distance. In the word list window 123, in relation to the second knowledge, “It can be determined that the degree of relevance is higher when the distance from the origin is farther”. In the hint window 124, the second knowledge is described with the title “point of analysis”. The hint window 124 is arranged at a position overlapping the scatter diagram window 122.

  The angle between the arrow 125 and the half lines 126 and 127 can be determined by an arbitrary method as long as the arrow 125 and the half lines 126 and 127 are included in the same quadrant. When the half lines 126 and 127 are described by giving an angle to the arrow 125, if the half lines 126 and 127 are included in a different quadrant from the arrow 125, the half lines 126 and 127 are described on the first or second component axis. Is done. The arrow 125 is preferably written in a color (for example, red) different from the scatter diagram. The half lines 126 and 127 are preferably described in a color (for example, blue) different from the scatter diagram and the arrow 125.

  FIG. 9 is a diagram showing a third support screen. The third support screen 130 shown in FIG. 9 includes a screen selection window 101, a scatter diagram window 132, a word list window 133, and a hint window 134. The third support screen 130 relates to the third knowledge “words that are close to each other have high similarity”. The user can efficiently perform the process of deriving knowledge from the scatter diagram using the third knowledge by looking at the third support screen 130.

  Before the third support screen 130 is displayed, the user operates the keyboard 28 or the mouse 29 to select one word, and designates a range determined to be near the selected word. Here, a case where the word “eye” is selected will be described. In the scatter diagram window 132, the scatter diagram shown in FIG. 3 is described. In the scatter diagram in the scatter diagram window 132, a circle 135 (appearance is an ellipse) indicating the vicinity of the selected word is described. The circle 135 is preferably described in a color different from the scatter diagram (for example, red). As described above, in the scatter diagram included in the third support screen 130, the range around the selected word is illustrated using the circle 135. Therefore, the user can easily know a word having a high similarity to the selected word by looking at the range shown.

  In the word list window 133, a word list in which words in the vicinity of the selected word (words in the circle 135) and the distance from the designated word of the word are arranged in order of increasing distance is described. In the word list window 133, as third knowledge, “the distance from the word is closer, it can be determined that the similarity is higher” is described. In this example, the variable closest to the selected word “eye” is “face”. Therefore, the word having the highest degree of similarity with the selected variable “eye” is “face”. The hint window 134 is labeled “Analysis Point” with a message to that effect. The hint window 134 is arranged at a position overlapping the scatter diagram window 132.

  Similar to the size of the circle 115 in the first support screen 110, the size of the circle 135 is determined by an arbitrary method. For example, the user can specify the size of the circle 135 by a method of specifying the number of words included in the circle 135, a method of specifying a ratio of words included in the circle 135, a method of specifying a distance from the selected word, and the like. To decide.

  FIG. 10 is a diagram illustrating the fourth support screen. The fourth support screen 140 shown in FIG. 10 includes a screen selection window 101, a scatter diagram window 142, a variable list window 143, and a hint window 144. The fourth support screen 140 relates to the fourth knowledge “variables with close distances have high similarity”. The user can efficiently perform the process of deriving knowledge from the scatter diagram using the fourth knowledge by looking at the fourth support screen 140.

  Before the fourth support screen 140 is displayed, the user operates the keyboard 28 or the mouse 29 to select one variable. Here, a case where the variable “Foreword” is selected will be described. In the scatter diagram window 142, the scatter diagram shown in FIG. 3 is described. In the scatter diagram in the scatter diagram window 142, there is described an arrow 145 starting from the selected variable and ending with the variable closest to the selected variable. The arrow 145 is preferably described in a color (for example, red) different from the scatter diagram. As described above, the scatter diagram included in the fourth support screen 140 shows the arrow 145 indicating the variable that is the closest to the selected variable. Therefore, the user can easily know the variable having the highest similarity with the selected variable by looking at the arrow 145 shown in the figure.

  The variable list window 143 describes a variable list in which variables that are relatively close to the selected variable and the distances are arranged in order of increasing distance. In the variable list window 143, the fourth knowledge is described as "It can be determined that the closer the distance from the variable is, the higher the similarity is." In this example, the variable closest to the selected variable “Foreword” is “Afterword”. Therefore, the variable with the highest similarity to the selected variable “Foreword” is “Afterword”. The hint window 144 is labeled with the title “Analysis Point”. The hint window 144 is arranged at a position overlapping the scatter diagram window 142.

  The text mining support device 10 may display a support screen other than the support screen described above. The support screen may include any content as long as it includes a scatter diagram and a hint indicating how to view the scatter diagram. The hint may explicitly indicate how to view the scatter diagram or may suggest how to view the scatter diagram. The hint may be included in any part of the support screen. The hint may be described in a window that overlaps with the scatter diagram window, may be described in a window that does not overlap with the scatter diagram window, or may be described in a message box having a fixed position.

  As described above, the text mining support method according to the present embodiment includes a screen including a step of inputting the analysis result 2, a step of inputting an instruction from the user, and a graph (scatter diagram) showing the analysis result 2 Generating screen data and displaying a screen based on the screen data. The step of generating screen data generates screen data of a support screen including a graph and a hint indicating how to read the graph according to the instruction. Therefore, the user can efficiently perform the process of deriving knowledge from the graph indicating the result of the correspondence analysis using the support screen including the graph indicating the result of the correspondence analysis and the hint indicating how to read the graph.

  The step of generating screen data is selected according to an instruction from a plurality of support screens (first to fourth support screens 110, 120, 130, and 140) and a basic screen 100 that includes a graph and does not include a hint. Screen data of the generated screen is generated. Thus, by selectively displaying the support screen including the hint and the basic screen not including the hint, it is possible to display a screen according to the level of the user. In addition, by selectively displaying a plurality of support screens, the user can be presented with a plurality of ways of viewing the graph.

  In the step of inputting the analysis result, as the analysis result 2, the first item (word) and the second item (variable) are associated with each other, and the first component and the second component of the first item, The step of generating the screen data by inputting the result including the first component and the second component of the two items is a graph in which the first item is in a plane with the first component as the horizontal axis and the second component as the vertical axis. And a scatter plot in which the second item is plotted. Therefore, the user can efficiently perform the process of deriving knowledge from the scatter diagram showing the result of the correspondence analysis regarding the first item and the second item.

  The plurality of support screens are in a direction away from the origin to the second item in the first support screen 110 including a hint that the first item near the origin in the scatter diagram does not have a remarkable feature. A second support screen 120 including a hint that the first item characterizes the second item as a hint, a third support screen 130 including a hint that the first items that are close to each other in the scatter diagram have high similarity, and The second items that are close to each other in the scatter diagram include a fourth support screen 140 that includes a hint that the degree of similarity is high. Therefore, the user can efficiently perform the process of deriving knowledge from the graph indicating the result of the correspondence analysis using the hint included in each support screen.

  In the scatter diagram included in the first support screen 110, a range near the origin is illustrated using a circle 115. In the scatter diagram included in the second support screen 120, the range in the direction away from the origin toward the second item selected is illustrated using the half lines 126 and 127. In the scatter diagram included in the third support screen 130, a range near the selected first item is illustrated using a circle 135. In the scatter diagram included in the fourth support screen 140, a mark (arrow 145) indicating the second item closest to the selected second item is shown. Therefore, the user looks at the range or mark shown on each support screen, and the first item that does not have a prominent feature, the first item that characterizes the selected second item, the selected first item, The first item having a high similarity and the second item having a high similarity with the selected second item can be easily known.

  In the step of inputting the analysis result, as a result of the analysis, a correspondence analysis is performed on the cross tabulation table in which the word is the first item, the sentence part is the second item, and the appearance frequency of each word in each part of the sentence is the in-table data. The result of performing is input. Therefore, the user can efficiently perform the process of deriving knowledge from the scatter diagram showing the result of the correspondence analysis regarding the word and sentence portions.

  The text mining support apparatus 10 according to the present embodiment and the text mining support program 31 according to the present embodiment have the same features as the text mining support method according to the present embodiment, and have the same effects.

  In the above description, the text mining support device 10 displays a scatter diagram that two-dimensionally shows the result of the correspondence analysis. The present invention is not limited to this, and the present invention can also be applied to a text mining support method and apparatus for displaying a graph (for example, a three-dimensional graph) showing the result of correspondence analysis in a multidimensional manner. Similarly to the text mining support method and apparatus for displaying a scatter diagram showing the result of the correspondence analysis on the cross tabulation table related to the text data, the graph showing the result of the correspondence analysis on the cross tabulation table on any data other than the text data ( A data mining support method and apparatus for displaying a scatter diagram or a three-dimensional graph can be configured.

  According to the text mining support method and apparatus of the present invention, by displaying a support screen including a graph indicating the result of correspondence analysis and a hint indicating how to read the graph, the user can learn from the graph indicating the result of correspondence analysis. Can be efficiently performed.

2 ... Analysis result 10 ... Text mining support device 11 ... Analysis result input unit 12 ... Instruction input unit 13 ... Screen generation unit 14 ... Analysis result display unit 31 ... Text mining support program 100 ... Basic screen 101 ... Screen selection window 102, 112 , 122, 132, 142 ... Scatter chart window 110, 120, 130, 140 ... Support screen 113, 123, 133 ... Word list window 143 ... Variable list window 114, 124, 134, 144 ... Hint window

Claims (16)

A text mining support method for displaying analysis results by correspondence analysis,
Inputting the analysis result;
Inputting instructions from the user;
Generating screen data of a screen including a graph indicating the analysis result;
And displaying a screen based on the screen data,
The step of generating the screen data generates a screen data of a support screen including the graph and a hint indicating how to read the graph in accordance with the instruction.   The step of generating the screen data includes generating screen data of a screen selected according to the instruction from a plurality of support screens and a basic screen including the graph and not including the hint. The text mining support method according to claim 1. The step of inputting the analysis result is a result of associating the first item and the second item as the analysis result, wherein the first component and the second component of the first item, and the second item A result including a first component and a second component is input;
The step of generating the screen data creates, as the graph, a scatter diagram in which the first item and the second item are plotted in a plane with the first component as the horizontal axis and the second component as the vertical axis. The text mining support method according to claim 2, wherein:   The text mining according to claim 3, wherein the plurality of support screens include a first support screen including the hint that the first item in the vicinity of the origin in the scatter diagram does not have a remarkable feature. Support method.   The text mining support method according to claim 4, wherein a range near the origin is shown in the scatter diagram included in the first support screen.   The plurality of support screens include a second support screen including, as the hint, the first item in a direction away from the origin toward the second item in the scatter diagram characterizes the second item. The text mining support method according to claim 3.   The text mining support method according to claim 6, wherein the scatter diagram included in the second support screen shows a range in a direction away from the origin toward the second item selected.   4. The text mining support according to claim 3, wherein the plurality of support screens include a third support screen including, as the hint, that the first items that are close to each other in the scatter diagram have high similarity. Method.   9. The text mining support method according to claim 8, wherein a range near the selected first item is illustrated in the scatter diagram included in the third support screen.   4. The text mining support according to claim 3, wherein the plurality of support screens include a fourth support screen including, as the hint, that the second items that are close to each other in the scatter diagram have high similarity. Method.   The text mining support according to claim 10, wherein the scatter diagram included in the fourth support screen includes a mark indicating the second item closest to the selected second item. Method.   In the step of inputting the analysis result, the analysis result is a cross tabulation table in which the word is the first item, the sentence part is the second item, and the appearance frequency of each word in each part of the sentence is in-table data. The text mining support method according to claim 3, wherein a result of correspondence analysis is inputted. A text mining support device that displays an analysis result by correspondence analysis,
An analysis result input unit for inputting the analysis result;
An instruction input unit for inputting instructions from the user;
A screen generator that generates screen data of a screen including a graph indicating the analysis result;
An analysis result display unit for displaying a screen based on the screen data;
The screen generation unit generates screen data of a support screen including the graph and a hint indicating how to read the graph in accordance with the instruction.   The screen generation unit generates screen data of a screen selected according to the instruction from a plurality of support screens and a basic screen including the graph and not including the hint. Item 14. The text mining support device according to Item 13. The analysis result input unit is a result of associating the first item and the second item as the analysis result, and includes the first component and the second component of the first item, and the second item of the second item. A result including a first component and a second component is input;
The screen generation unit creates, as the graph, a scatter diagram in which the first item and the second item are plotted in a plane having the first component as a horizontal axis and the second component as a vertical axis. The text mining support device according to claim 14, characterized in that it is characterized by   In the analysis result input unit, as the analysis result, a word is the first item, a sentence part is the second item, and a frequency of each word in each part of the sentence is an in-table data. The text mining support device according to claim 15, wherein the result of the correspondence analysis is input.

Images