JP6987003B2 - Text mining methods, text mining programs, and text mining equipment - Google Patents

Description

The present invention relates to text mining, and more particularly to a text mining method for displaying a screen including a co-occurrence network of words, a text mining program, and a text mining device.

In recent years, text mining, which analyzes freely described text data and seeks useful information from the analysis results, has attracted attention. In text mining, for example, words are extracted from the text data to be analyzed, and information is obtained by analyzing the appearance frequency and appearance tendency of the words.

When analyzing freely written text data, the analyst needs to get an overall picture of the text data rather than subjectively selecting the target at the initial stage. For this reason, the analyst may use a co-occurrence network of words contained in the text data.

FIG. 19 is a diagram showing an example of a co-occurrence network. The co-occurrence network extracts a pair of words that are often included in the same sentence from text data, and expresses the result in an undirected graph. When the word Wa and the word Wb are often included in the same sentence in the text data to be analyzed, the co-occurrence network has a node corresponding to the word Wa, a node corresponding to the word Wb, and an edge connecting the two. included. The co-occurrence network shown in FIG. 19 includes a node corresponding to "staff", a node corresponding to "correspondence", and an edge connecting both. Looking at the co-occurrence network shown in FIG. 19, it can be seen that in the text data to be analyzed, "staff" and "correspondence" are often included in the same sentence.

Generally, co-occurrence networks are generated based on the entire specified text data. Hereinafter, such a co-occurrence network is referred to as a "whole co-occurrence network". The analyst selects a plurality of notable words (hereinafter referred to as notable words) from the whole co-occurrence network according to the hypothesis and the analysis purpose set by the analyst, and performs the subsequent analysis in consideration of the notable words.

When an analyst selects a attention word, how the attention word is used in a sentence containing the attention word in order to determine whether or not the selected attention word is suitable for an analysis purpose or the like. Consider. For this reason, the analyst may use a co-occurrence network based on text data (hereinafter referred to as limited text data) consisting of sentences including a word of interest in the designated text data. The "sentence including the attention word" here means not only a single sentence including the attention word but also a plurality of sentences divided into block units such as a paragraph containing the sentence including the attention word. May mean (a set of sentences). Hereinafter, such a co-occurrence network is referred to as a "limited co-occurrence network". The analyst can grasp the content of the limited text data by using the limited co-occurrence network. The analyst repeatedly references the whole co-occurrence network and the limited co-occurrence network until all the words of interest are selected.

Hereinafter, a text mining device that generates a co-occurrence network of words included in text data and displays a screen including the generated co-occurrence network will be considered. Patent Document 1 describes a document database display device that generates a total co-occurrence network for each of a plurality of documents and displays a screen including the generated total co-occurrence network. This display device searches for a word input by the user from a plurality of co-occurrence networks, and highlights the searched word on the screen.

Japanese Unexamined Patent Publication No. 8-314980

A conventional text mining device creates a co-occurrence network based on the entire specified text data. Therefore, according to the conventional text mining device, the screen including the whole co-occurrence network can be easily displayed.

On the other hand, when displaying a screen including a limited co-occurrence network using a conventional text mining device, the analyst needs to perform complicated operations. Specifically, each time the analyst selects one noteworthy word from the whole co-occurrence network, the analyst generates limited text data based on the specified text data, and the generated limited text data is used as a text mining device. Need to be given to. Also, the analyst refers to both the global co-occurrence network and the limited co-occurrence network when selecting the term of interest. Therefore, the text mining device needs to store both the image data of the whole co-occurrence network and the image data of the limited co-occurrence network. However, when many co-occurrence networks are generated, it becomes difficult to store and manage image data.

Therefore, it is an object of the present invention to provide a text mining method, a text mining program, and a text mining device capable of displaying a screen including a co-occurrence network when a word of interest is specified by a simple operation.

The first aspect of the present invention is a text mining method for displaying a screen including an analysis result of text data.
Steps to extract words from text data,
Steps to generate a co-occurrence matrix for the word,
Steps to generate a co-occurrence network based on the co-occurrence matrix,
A step of displaying a screen including the co-occurrence network is provided.
When an instruction to specify a word of interest is input in the first screen including the first co-occurrence network based on the entire specified text data, the step of extracting the word is described in the specified text data. In the step of extracting the word from the limited text data including the part including the word of interest and generating the co-occurrence matrix, a second co-occurrence matrix is generated for the word using the limited text data, and the co-occurrence network is generated. The generation step is characterized in that a second co-occurrence network is generated based on the second co-occurrence matrix, and the step of displaying the screen is characterized in displaying a second screen including the second co-occurrence network.

The second aspect of the present invention is the first aspect of the present invention.
By selecting one or more nodes included in the first co-occurrence network in the first screen and selecting start analysis, an instruction to specify the word corresponding to the node as the attention word is input. It is characterized by that.

The third aspect of the present invention is the first aspect of the present invention.
By continuously selecting one node included in the first co-occurrence network in the first screen, an instruction for designating the word corresponding to the node as the attention word is input. ..

The fourth aspect of the present invention is the first aspect of the present invention.
By continuously selecting one edge included in the first co-occurrence network in the first screen, an instruction to specify a word corresponding to two nodes connected to the edge as the attention word is given. It is characterized by being input.

The fifth aspect of the present invention is the first aspect of the present invention.
By selecting one or more edges included in the first co-occurrence network in the first screen and selecting start analysis, the word corresponding to the plurality of nodes connected to the edge is referred to as the attention word. It is characterized in that the instruction specified as is input.

The sixth aspect of the present invention is the first aspect of the present invention.
When a merge instruction is input in a second screen including a plurality of second co-occurrence networks, the step of displaying the screen is characterized in that the plurality of second co-occurrence networks are displayed in a tab format. ..

The seventh aspect of the present invention is the sixth aspect of the present invention.
It is characterized in that the merge instruction is input by grasping one second co-occurrence network in the second screen and separating it in another second co-occurrence network.

The eighth aspect of the present invention is the first aspect of the present invention.
The limited text data is characterized by consisting of a sentence including the word of interest in the designated text data.

The ninth aspect of the present invention is the eighth aspect of the present invention.
The limited text data when a plurality of attention words are designated is characterized by comprising a sentence including all of the plurality of attention words among the designated text data.

The tenth aspect of the present invention is the eighth aspect of the present invention.
The limited text data when a plurality of attention words are designated is characterized by comprising a sentence including any one of the plurality of attention words among the designated text data.

The eleventh aspect of the present invention is the first aspect of the present invention.
The step of generating the co-occurrence matrix is characterized in that a co-occurrence matrix having a Jaccard coefficient as an element is generated.

The twelfth aspect of the present invention is a text mining program for displaying a screen including an analysis result of text data.
Steps to extract words from text data,
Steps to generate a co-occurrence matrix for the word,
Steps to generate a co-occurrence network based on the co-occurrence matrix,
The CPU causes the computer to execute the step of displaying the screen including the co-occurrence network by using the memory.
When an instruction to specify a word of interest is input in the first screen including the first co-occurrence network based on the entire specified text data, the step of extracting the word is described in the specified text data. In the step of extracting the word from the limited text data including the part including the word of interest and generating the co-occurrence matrix, a second co-occurrence matrix is generated for the word using the limited text data, and the co-occurrence network is generated. The generation step is characterized in that a second co-occurrence network is generated based on the second co-occurrence matrix, and the step of displaying the screen is characterized in displaying a second screen including the second co-occurrence network.

The thirteenth aspect of the present invention is the twelfth aspect of the present invention.
By selecting one or more nodes included in the first co-occurrence network in the first screen and selecting start analysis, an instruction to specify the word corresponding to the node as the attention word is input. It is characterized by that.

The fourteenth aspect of the present invention is the twelfth aspect of the present invention.
By continuously selecting one node included in the first co-occurrence network in the first screen, an instruction for designating the word corresponding to the node as the attention word is input. ..

The fifteenth aspect of the present invention is the twelfth aspect of the present invention.
By continuously selecting one edge included in the first co-occurrence network in the first screen, an instruction to specify a word corresponding to two nodes connected to the edge as the attention word is given. It is characterized by being input.

The sixteenth aspect of the present invention is the twelfth aspect of the present invention.
By selecting one or more edges included in the first co-occurrence network in the first screen and selecting start analysis, the word corresponding to the plurality of nodes connected to the edge is referred to as the attention word. It is characterized in that the instruction specified as is input.

The seventeenth aspect of the present invention is the twelfth aspect of the present invention.
When a merge instruction is input in a second screen including a plurality of second co-occurrence networks, the step of displaying the screen is characterized in that the plurality of second co-occurrence networks are displayed in a tab format. ..

The eighteenth aspect of the present invention is the seventeenth aspect of the present invention.
It is characterized in that the merge instruction is input by grasping one second co-occurrence network in the second screen and separating it in another second co-occurrence network.

A nineteenth aspect of the present invention is a text mining device that displays a screen containing analysis results of text data.
A word extractor that extracts words from text data,
A co-occurrence matrix generator that generates a co-occurrence matrix for the word,
A co-occurrence network generation unit that generates a co-occurrence network based on the co-occurrence matrix,
It is provided with a screen display unit that displays a screen including the co-occurrence network.
When an instruction to specify a attention word is input in the first screen including the first co-occurrence network based on the entire specified text data, the word extraction unit uses the attention word in the specified text data. The word is extracted from the limited text data including the portion including, the co-occurrence matrix generation unit generates a second co-occurrence matrix using the limited text data for the word, and the co-occurrence network generation unit generates the first. A second co-occurrence network is generated based on the two co-occurrence matrix, and the screen display unit displays a second screen including the second co-occurrence network.

The twentieth aspect of the present invention is the nineteenth aspect of the present invention.
When a merge instruction is input in a second screen including a plurality of second co-occurrence networks, the screen display unit is characterized in that the plurality of second co-occurrence networks are displayed in a tab format.

According to the first, twelfth or nineteenth aspect described above, when an instruction to specify a word of interest is input in the first screen including the first co-occurrence network based on the entire specified text data, the designation is made. A second screen including a second co-occurrence network based on the portion of the text data that contains the word of interest is displayed. Therefore, the screen including the co-occurrence network when the attention word is specified can be displayed by a simple operation.

According to the second or thirteenth aspect, the instruction to specify one or more attention words is input by a simple operation by selecting one or more nodes and the analysis start in the first screen. It is possible to display a screen including a co-occurrence network when one or more attention words are specified.

According to the third or fourteenth aspect, by continuously selecting one node in the first screen, an instruction to specify one attention word is input by a simple operation, and one attention is given. A screen containing the co-occurrence network when a word is specified can be displayed.

According to the fourth or fifteenth aspect, by continuously selecting one edge in the first screen, an instruction to specify two attention words is input by a simple operation, and two attentions are given. A screen containing the co-occurrence network when a word is specified can be displayed.

According to the fifth or sixteenth aspect, by selecting one or more edges and the start of analysis in the first screen, instructions for designating a plurality of attention words can be input by a simple operation, and a plurality of edges can be input. It is possible to display a screen including the co-occurrence network when the attention word is specified.

According to the sixth, 17th, or twentieth aspect, the plurality of second co-occurrence networks are displayed compactly by displaying the plurality of second co-occurrence networks in a tab format when the merge instruction is input. can do.

According to the seventh or eighteenth aspect, by grasping and releasing the second co-occurrence network in the second screen, a merge instruction can be input with a simple operation, and a plurality of second co-occurrence networks can be displayed compactly. can do.

According to the eighth aspect, when the instruction to specify the attention word is input, the specified text data is divided into sentence units to obtain the limited text data, and the second co-occurrence based on the obtained limited text data. A screen including the network can be displayed.

According to the ninth or tenth aspect, it is possible to display a screen including the second co-occurrence network when AND processing or OR processing is performed on a plurality of notable words.

According to the eleventh aspect, the co-occurrence of words contained in the text data can be suitably analyzed by generating a co-occurrence matrix having a Jaccard coefficient as an element.

It is a block diagram which shows the structure of the text mining apparatus which concerns on embodiment of this invention. It is a block diagram which shows the structure of the computer which functions as the text mining apparatus shown in FIG. It is a flowchart which shows the operation of the text mining apparatus shown in FIG. It is a figure which shows the example of the co-occurrence matrix generated by the text mining apparatus shown in FIG. It is a figure which shows the example of the window including the whole co-occurrence network displayed by the text mining apparatus shown in FIG. 1. It is a figure which shows the 1st operation which specifies the attention word in the window shown in FIG. It is a figure which shows the 2nd operation which specifies the attention word in the window shown in FIG. It is a figure which shows the 3rd operation which specifies the attention word in the window shown in FIG. It is a figure which shows the 4th operation which specifies the attention word in the window shown in FIG. It is a figure which shows the 5th operation which specifies the attention word in the window shown in FIG. It is a figure which shows the 6th operation which specifies the attention word in the window shown in FIG. It is a figure which shows the example of the window including the limited co-occurrence network displayed by the text mining apparatus shown in FIG. 1. It is a figure which shows the example of the window including the limited co-occurrence network displayed by the text mining apparatus shown in FIG. 1. It is a figure which shows the example of the display screen of the text mining apparatus shown in FIG. It is a figure which shows the example of the display screen of the text mining apparatus shown in FIG. It is a figure which shows the example of the display screen of the text mining apparatus shown in FIG. It is a figure which shows the operation of merging windows in the text mining apparatus shown in FIG. It is a figure which shows the display screen after performing the operation shown in FIG. It is a figure which shows the example of the co-occurrence network.

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

FIG. 1 is a block diagram showing a configuration of a text mining device according to an embodiment of the present invention. The text mining device 10 shown in FIG. 1 includes an instruction input unit 11, a text data storage unit 12, a word extraction unit 13, a co-occurrence matrix generation unit 14, a co-occurrence network generation unit 15, and a screen display unit 16. .. The text mining device 10 generates a co-occurrence network as an analysis result of the text data based on the text data stored in the text data storage unit 12, and displays a screen including the generated co-occurrence network.

The outline of the operation of the text mining device 10 is as follows. Instructions from the user (analyzer of text data) are input to the instruction input unit 11. The text data storage unit 12 stores one or more freely described text data. The word extraction unit 13 reads out the text data designated from the text data storage unit 12, and performs morphological analysis on the read text data to extract words from the text data. The co-occurrence matrix generation unit 14 generates a co-occurrence matrix for the words extracted by the word extraction unit 13. The co-occurrence network generation unit 15 generates a co-occurrence network based on the co-occurrence matrix generated by the co-occurrence matrix generation unit 14. The screen display unit 16 displays a screen including the co-occurrence network generated by the co-occurrence network generation unit 15.

The user inputs an instruction for designating the text data to be analyzed, an instruction for designating a word of interest, and the like by using the instruction input unit 11. The word extraction unit 13, the co-occurrence network generation unit 15, and the screen display unit 16 perform an operation for displaying a screen including the co-occurrence network according to an instruction from the user. When the instruction to specify the text data is input, the whole co-occurrence network based on the whole of the specified text data is generated, and the screen including the whole co-occurrence network is displayed. When an instruction to specify a noteworthy word is input in the screen including the whole co-occurrence network, a limited co-occurrence network based on the sentence including the noteworthy word in the specified text data is generated, and the screen including the limited co-occurrence network is generated. Is displayed.

FIG. 2 is a block diagram showing a configuration of a computer functioning as a text mining device 10. The computer 20 shown in FIG. 2 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 the main memory 22, for example, DRAM is used. For example, a hard disk or a solid state drive is used for the storage unit 23. The input unit 24 includes, for example, a keyboard 28 and a mouse 29. For the display unit 25, for example, a liquid crystal display is used. 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 a program or the like. As the recording medium 30, for example, a non-transient recording medium such as a CD-ROM, a DVD-ROM, or a USB memory is used.

When the computer 20 executes the text mining program 31, the storage unit 23 stores the text mining program 31 and the text data 32. The text mining program 31 and the text data 32 may be, for example, those received from a server or another computer using the communication unit 26, or those read from the recording medium 30 using the recording medium reading unit 27.

When the text mining program 31 is executed, the text mining program 31 and the text data 32 are copied and transferred to the main memory 22. The CPU 21 uses the main memory 22 as a working memory, executes a text mining program 31 stored in the main memory 22, processes to extract words from the text data 32, and creates a co-occurrence matrix for the extracted words. The process of generating, the process of generating a co-occurrence network based on the generated co-occurrence matrix, the process of displaying the screen including the generated co-occurrence network, and the like are performed. At this time, the computer 20 functions as a text mining device 10. The configuration of the computer 20 described above is only an example, and the text mining device 10 can be configured by using any computer.

FIG. 3 is a flowchart showing the operation of the text mining device 10. Before performing the operation shown in FIG. 3, the text data storage unit 12 stores one or more freely described text data. Each text data contains a plurality of sentences. The text mining device 10 processes the text data designated by the user among the text data stored in the text data storage unit 12.

In FIG. 3, the instruction input unit 11 first receives an instruction to specify text data from the user (step S101). At this time, in addition to the instruction for designating the text data, the instruction input unit 11 includes an instruction for setting a reference value of the co-occurrence matrix (details will be described later), an instruction for switching between AND processing and OR processing (details will be described later), and co-occurrence. You may receive an instruction to set the details of the display mode of the origin network. The received instruction is output to each part of the text mining device 10.

Next, the word extraction unit 13 reads out the designated text data from the text data storage unit 12 (step S102). Next, the word extraction unit 13 extracts words from the read text data by performing morphological analysis on the text data read in step S102 (step S103). At this time, the word extraction unit 13 extracts only the words necessary for the later analysis from the read text data. Next, the co-occurrence matrix generation unit 14 generates a co-occurrence matrix for the words extracted in step S103 using the text data read in step S102 (step S104).

FIG. 4 is a diagram showing an example of a co-occurrence matrix generated by the co-occurrence matrix generation unit 14. The element of the co-occurrence matrix is the Jaccard index obtained for a pair of words. For the text data to be analyzed, let A be a set of sentences including the word Wa, and let B be a set of sentences including the word Wb. The Jaccard index K (Wa, Wb) for a word pair (Wa, Wb) is given by the following equation (1).
K (Wa, Wb) = | A∩B | / | A∪B |… (1)
However, in the equation (1), the symbol ∩ represents an operation for obtaining an intersection, the symbol ∪ represents an operation for obtaining a union, and | S | represents the number of elements included in the set S.

In step S104, the co-occurrence matrix generation unit 14 obtains a Jaccard index for all the word pairs extracted from the entire text data read in step S102, and obtains a co-occurrence matrix having the obtained Jaccard coefficient as an element. Generate. The rows and columns of the co-occurrence matrix correspond to the types of words extracted from the entire text data read in step S102. When n kinds of words are extracted from the whole read text data, the co-occurrence matrix generated in step S104 is an n-row, n-column symmetric matrix having all 1 diagonal elements.

The co-occurrence matrix generation unit 14 may divide the text data into units other than sentences to obtain the Jaccard index. For example, the co-occurrence matrix generation unit 14 may obtain the Jaccard index according to the equation (1), where A is a set of paragraphs containing the word Wa and B is a set of paragraphs containing the word Wb. When the sentence included in the text data has a date, the co-occurrence matrix generation unit 14 divides the text data into a plurality of parts consisting of sentences having the same date, and sets a set of parts including the word Wa as A. The Jaccard index may be obtained according to the equation (1), where B is a set of parts including the word Wb. Further, the co-occurrence matrix generation unit 14 may generate a co-occurrence matrix including other values indicating the co-occurrence of words (for example, Simpson coefficient, cosine distance, etc.) as elements.

Next, the co-occurrence network generation unit 15 generates an entire co-occurrence network based on the co-occurrence matrix generated in step S104 (step S105). Next, the screen display unit 16 displays a screen including the entire co-occurrence network generated in step S105 (step S106). FIG. 5 is a diagram showing an example of a window including a whole co-occurrence network displayed in step S106. The window 41 shown in FIG. 5 includes a whole co-occurrence network 51 and an analysis button 61. The analysis button 61 is provided to instruct the start of analysis.

The co-occurrence network generation unit 15 has a reference value (hereinafter referred to as V) of the co-occurrence matrix. The reference value V may be a predetermined value or a value set by the user using the instruction input unit 11. In the co-occurrence matrix generated in step S104, when the maximum value of the Jaccard coefficient K (Wa, *) included in the row corresponding to the word Wa is equal to or larger than the reference value V, the co-occurrence network generation unit 15 determines the word Wa. Include the corresponding node (the node described as the word Wa) in the global co-occurrence network. Further, in the co-occurrence matrix generated in step S104, when the Jaccard coefficient K (Wa, Wb) related to the word pair (Wa, Wb) is equal to or more than the reference value V, the co-occurrence network generation unit 15 is assigned to the word Wa. The edge connecting the corresponding node and the node corresponding to the word Wb is included in the whole co-occurrence network.

In the general co-occurrence network 51 shown in FIG. 5, the nodes corresponding to the words having a high frequency of appearance are displayed in large size. When displaying a screen including a co-occurrence network, when the Jaccard index K (Wa, Wb) is large, the edge connecting the node corresponding to the word Wa and the node corresponding to the word Wb may be displayed thickly. .. Further, the color of the edge may be switched according to the Jaccard index, or both the thickness and the color of the edge may be switched. The co-occurrence network is divided into multiple parts that can be reached via the edge. When displaying the screen including the co-occurrence network, a plurality of nodes included in each part may be displayed in the color assigned to each part. The positions of the nodes and edges included in the co-occurrence network are meaningless.

Next, the instruction input unit 11 receives an instruction to specify a word of interest from the user (step S111). When step S111 is executed, a screen including the whole co-occurrence network is displayed. The user inputs an instruction to specify a word of interest by operating the mouse 29 and selecting an element of the whole co-occurrence network. When inputting an instruction, the user may use the keyboard 28 instead of the mouse 29, or may perform an operation such as directly touching the display screen. Hereinafter, it is assumed that the screen including the window 41 shown in FIG. 5 is displayed when the step S111 is executed.

6 to 11 are diagrams showing the first to sixth operations for designating a word of interest in the window 41, respectively. In FIGS. 6 to 11, the balloon indicates the operation procedure, and the white arrow indicates the movement of the mouse cursor 62. Callouts and arrows do not appear on the actual screen. Hereinafter, clicking (double-clicking) the button of the mouse 29 when the mouse cursor 62 is on a certain element in the display screen is referred to as "clicking (double-clicking) the element".

As shown in FIG. 6, the user first clicks the node corresponding to the word (here, “open-air bath”) specified as the word of interest in the window 41 (first click), and then clicks the analysis button 61. Click (second click). By this operation, the word corresponding to the node clicked for the first time is designated as the word of interest. In this way, by selecting one node included in the whole co-occurrence network in the screen including the whole co-occurrence network and selecting the start of analysis, an instruction to specify one attention word is input.

As shown in FIG. 7, the user double-clicks the node corresponding to the word designated as the attention word (here, “open-air bath”) in the window 41. By this operation, the word corresponding to the double-clicked node is designated as the attention word. In this way, by continuously selecting one node included in the whole co-occurrence network in the screen including the whole co-occurrence network, an instruction to specify one attention word is input.

As shown in FIG. 8, the user first clicks the node corresponding to the word designated as the attention word (here, “open-air bath”) in the window 41 (first click), and then designates it as the attention word. Click the node corresponding to another word (here, "price") (second click), and finally click the analysis button 61 (last click). By this operation, the two words corresponding to the nodes clicked the first time and the second time are designated as the words of interest. The user may click p nodes (p is an integer of 3 or more) in order in the window 41, and finally click the analysis button 61. By this operation, p words corresponding to p nodes are designated as attention words. In this way, by selecting a plurality of nodes included in the global co-occurrence network in the screen including the global co-occurrence network and selecting start analysis, instructions for designating a plurality of attention words are input.

As shown in FIG. 9, the user double-clicks the edge connecting the two nodes corresponding to the two words (here, "open-air bath" and "stairs") specified as the words of interest in the window 41. do. As a result, the two words corresponding to the two nodes connected to the double-clicked edge are designated as the words of interest. In this way, by continuously selecting one edge included in the whole co-occurrence network in the screen including the whole co-occurrence network, an instruction to specify two attention words is input.

As shown in FIG. 10, the user first clicks on the edge connecting the two nodes corresponding to the two words (here, "open-air bath" and "stairs") designated as the attention words in the window 41. Then click the analysis button 61 (second click). As a result, the two words corresponding to the two nodes connected to the edge clicked the first time are designated as the words of interest. In this way, by selecting one edge included in the whole co-occurrence network in the screen including the whole co-occurrence network and selecting the start of analysis, an instruction to specify two attention words is input.

As shown in FIG. 11, the user first clicks on the edge connecting the two nodes corresponding to the two words (here, "open-air bath" and "stairs") designated as the attention words in the window 41. (1st click), then click the edge connecting the 2 nodes corresponding to the other 2 words (here "price" and "think") that you specify as the word of interest (2nd click) Click), and finally click the analysis button 61 (last click). By this operation, four words corresponding to the four nodes connected to the two edges clicked the first time and the second time are designated as the words of interest. The user may click q edges (q is an integer of 3 or more) in order in the window 41, and finally click the analysis button 61. By this operation, 2q words corresponding to 2q nodes connected to q edges are designated as attention words. In this way, by selecting a plurality of edges included in the whole co-occurrence network in the screen including the whole co-occurrence network and selecting start analysis, instructions for designating a plurality of attention words are input.

In step S111, the instruction input unit 11 sets, in addition to the instruction to specify the word of interest, an instruction to set a reference value of the co-occurrence matrix, an instruction to switch between AND processing and OR processing, and details of the display mode of the co-occurrence network. You may receive instructions to do so. The received instruction is output to each part of the text mining device 10.

Next, the word extraction unit 13 obtains limited text data consisting of sentences including the attention word by extracting a sentence including the attention word specified in step S111 from the text data read in step S102 (step S112). ..

The word extraction unit 13 has a flag indicating which of the AND process and the OR process is to be performed when a plurality of noteworthy words are specified. The value of the flag may be a predetermined value or a value set by the user using the instruction input unit 11. When the flag indicates AND processing, the word extraction unit 13 obtains limited text data by extracting a sentence including all of a plurality of designated words of interest from the read text data. When the flag indicates OR processing, the word extraction unit 13 obtains limited text data by extracting a sentence including any of the designated words of interest from the read text data.

Next, the word extraction unit 13 extracts words from the limited text data by performing morphological analysis on the limited text data obtained in step S112 (step S113). Next, the co-occurrence matrix generation unit 14 generates a co-occurrence matrix for the words extracted in step S113 using the limited text data obtained in step S112 (step S114). Next, the co-occurrence network generation unit 15 generates a limited co-occurrence network based on the co-occurrence matrix generated in step S114 (step S115). The processing targets are different between steps S103 to S105 and steps S113 to S115, but the processing contents are the same.

Generally, the types of words extracted from the limited text data obtained in step S112 are smaller than the types of words extracted from the text data read in step S102. The co-occurrence matrix generated in step S114 is different from the co-occurrence matrix generated in step S104. The limited co-occurrence network generated in step S115 is different from the whole co-occurrence network generated in step S105.

Next, the screen display unit 16 displays a screen including the limited co-occurrence network generated in step S115 (step S116). 12 and 13 are diagrams showing an example of a window containing a limited co-occurrence network displayed in step S116. The window 42 shown in FIG. 12 includes a limited co-occurrence network 52 when one noteworthy word (here, “open-air bath”) is specified. The window 43 shown in FIG. 13 includes a limited co-occurrence network 53 when two notable words (here, “open-air bath” and “bathhouse”) are specified.

14 and 15 are diagrams showing an example of a display screen of the text mining device 10. The screen display unit 16 may display the window including the whole co-occurrence network and the window including the limited co-occurrence network side by side without overlapping, or may display both of them in an overlapping manner. In the screen 71 shown in FIG. 14, the window 41 including the whole co-occurrence network 51 and the window 42 including the limited co-occurrence network 52 are displayed side by side without overlapping. The user can see the whole co-occurrence network 51 and the limited co-occurrence network 52 at the same time on the screen 71. In the screen 72 shown in FIG. 15, the window 42 including the limited co-occurrence network 52 is displayed superimposed on the window 41 including the whole co-occurrence network 51. The user can switch between the whole co-occurrence network 51 and the limited co-occurrence network 52 on the screen 72.

Next, the instruction input unit 11 receives an instruction from the user (step S121). Next, the text mining device 10 determines whether or not the instruction received in step S121 is an instruction for designating a word of interest (step S122). If Yes in step S122, the control of the text mining device 10 proceeds to step S112. In this case, steps S112 to S116 are executed for the attention word specified in step S121, and a screen including a limited co-occurrence network based on limited text data consisting of sentences including the attention word specified in step S121 is displayed.

FIG. 16 is a diagram showing an example of a display screen of the text mining device 10. In the screen 73 shown in FIG. 16, the window 41 including the whole co-occurrence network 51 and the window 42 including the limited co-occurrence network 52 are overlapped with each other, and the limited co-occurrence network 54 when "bath" is designated as a noteworthy word is included. The window 44 is displayed. The screen 73 is displayed when "open-air bath" is designated as a noteworthy word in step S111 and "bathhouse" is designated as a noteworthy word in step S121. The user can switch between the whole co-occurrence network 51 and the limited co-occurrence networks 52 and 54 on the screen 73.

If No in step S122, the control of the text mining device 10 proceeds to step S123. In this case, the instructions received in step S121 are, for example, an instruction to move the window, an instruction to hide the window, an instruction to close the window, an instruction to merge the windows, and the like. The user inputs these instructions by operating the instruction input unit 11 when the screen including the whole co-occurrence network and the limited co-occurrence network is displayed. The screen display unit 16 displays the updated screen according to the instruction received in step S121 (step S123). After that, the control of the text mining device 10 proceeds to step S121.

FIG. 17 is a diagram showing an operation of merging windows. The screen 74 shown in FIG. 17 includes a window 42 including a limited co-occurrence network 52 when "open-air bath" is designated as a noteworthy word, and a limited co-occurrence network 54 when "bathhouse" is designated as a noteworthy word. Window 44 is displayed. The user can simultaneously view the two limited co-occurrence networks 52 and 54 on the screen 74.

The hatched arrow shown in FIG. 17 indicates that the mouse cursor 62 has moved while the button of the mouse 29 is pressed. This arrow does not appear on the actual screen. The user performs an operation (drop operation) of grasping the limited co-occurrence network 52 in the screen 74 and releasing it in the limited co-occurrence network 54. More specifically, the user presses the button of the mouse 29 when the mouse cursor 62 is in the window 42, moves the mouse cursor 62 into the window 44 while holding down the button of the mouse 29, and causes the mouse cursor 62. Release the mouse 29 button when is in the window 44. By this operation, an instruction to merge windows is input.

FIG. 18 is a diagram showing a display screen after performing the operation shown in FIG. On the screen 75 shown in FIG. 18, a window 45 for displaying a plurality of limited co-occurrence networks in a tab format is displayed. In FIG. 18, the tab 64 described as “open-air bath” is selected, and the window 45 displays the limited co-occurrence network 52 when “open-air bath” is designated as a noteworthy word. When the tab 63 described as "bathhouse" is selected, the window 45 displays the limited co-occurrence network 54 shown in FIG.

When the user clicks the close button (x mark) in the window 45, the window 45 closes. When the user clicks the close button in the tab 63, the tab 63 disappears. When the user clicks the close button in the tab 64, the tab 64 is not displayed and the limited co-occurrence network 54 is displayed in the window 45.

As described above, in the text mining method according to the present embodiment, a step of extracting words from text data (steps S102, S103, S112, S113) and a step of generating a co-occurrence matrix for the extracted words (step S104). , S114), a step of generating a co-occurrence network based on the generated co-occurrence matrix (steps S105 and S115), and a step of displaying a screen including the co-occurrence network (steps S106 and S116). When an instruction to specify a word of interest is input in the first screen (screen including the window 41) including the first co-occurrence network (total co-occurrence network 51) based on the entire specified text data, the word is input. The extraction step (steps S112 and S113) is a step (step S114) of extracting a word from the limited text data consisting of a portion including the attention word (sentence including the attention word) in the designated text data and generating a co-occurrence matrix. ) Generates a second co-occurrence matrix using limited text data for the extracted word, and the step of generating a co-occurrence network (step S115) is a second co-occurrence network (limited co-occurrence network) based on the second co-occurrence matrix. The step of generating 52 to 54) and displaying the screen (step S116) displays the second screen (screen including windows 42 to 45) including the second co-occurrence network. As described above, in the text mining method according to the present embodiment, the text mining method is designated when an instruction to specify a word of interest is input in the first screen including the first co-occurrence network based on the entire designated text data. The second screen including the second co-occurrence network based on the part of the text data including the word of interest is displayed. Therefore, the screen including the co-occurrence network when the attention word is specified can be displayed by a simple operation.

In addition, by selecting one or more nodes included in the first co-occurrence network in the first screen and selecting start analysis, an instruction to specify the word corresponding to the node as the word of interest is input (). 6 and 8). By selecting one or more nodes and start analysis in the first screen in this way, you can easily enter instructions to specify one or more attention words, and one or more attention words can be selected. A screen including the co-occurrence network at the specified time can be displayed. Further, by continuously selecting one node included in the first co-occurrence network in the first screen, an instruction to specify the word corresponding to the node as the word of interest is input (FIG. 7). By selecting in this way continues to one of the nodes in the first screen, co-occurrence network when inputs an instruction to specify one noteworthy by a simple operation, specifying one noteworthy Can display screens that include.

In addition, by continuously selecting one edge included in the first co-occurrence network in the first screen, an instruction to specify the word corresponding to the two nodes connected to the edge as the word of interest is input. (Fig. 9). By continuously selecting one edge in the first screen in this way, an instruction to specify two attention words can be input with a simple operation, and a co-occurrence network when two attention words are specified. Can display screens that include. Also, by selecting one or more edges included in the first co-occurrence network in the first screen and selecting start analysis, the words corresponding to the multiple nodes connected to the edges are specified as the words of interest. Instructions to be input are input (FIGS. 10 and 11). By selecting one or more edges and the start of analysis in the first screen in this way, an instruction to specify multiple attention words can be input with a simple operation, and co-occurrence when multiple attention words are specified. You can display a screen that includes the network.

Further, when the merge instruction is input in the second screen (screen 74) including the plurality of second co-occurrence networks (limited co-occurrence networks 52, 54) (FIG. 17), there are a plurality of steps for displaying the screen. The second co-occurrence network of No. 1 is displayed in a tab format (FIG. 18). As a result, a plurality of second co-occurrence networks can be displayed compactly. Further, by grasping one second co-occurrence network (limited co-occurrence network 52) in the second screen and separating it in another second co-occurrence network (limited co-occurrence network 54), a merge instruction is input. .. Therefore, the merge instruction can be input by a simple operation, and a plurality of second co-occurrence networks can be displayed compactly.

The limited text data may be composed of sentences including the word of interest in the designated text data. In this case, when an instruction to specify a word of interest is input, the specified text data is divided into sentence units to obtain limited text data, and a screen including the second co-occurrence network based on the obtained limited text data is displayed. can do. The limited text data when a plurality of attention words are specified may be composed of a sentence including all of the plurality of attention words in the specified text data. In this case, it is possible to display a screen including the second co-occurrence network when AND processing is performed on a plurality of attention words. The limited text data when a plurality of attention words are specified may be composed of a sentence including any one of the plurality of attention words among the specified text data. In this case, it is possible to display a screen including the second co-occurrence network when OR processing is performed on a plurality of attention words. Further, the step of generating a co-occurrence matrix generates a co-occurrence matrix having a Jaccard coefficient as an element. Therefore, the co-occurrence of words contained in the text data can be suitably analyzed.

The text mining device 10 and the text mining program 31 according to the present embodiment have the same characteristics as the above-mentioned text mining method, and have the same effects. According to the text mining method, the text mining device 10, and the text mining program 31 according to the present embodiment, it is possible to display a screen including a co-occurrence network when a word of interest is specified by a simple operation.

10 ... Text mining device 11 ... Instruction input unit 12 ... Text data storage unit 13 ... Word extraction unit 14 ... Co-occurrence matrix generation unit 15 ... Co-occurrence network generation unit 16 ... Screen display unit 20 ... Computer 21 ... CPU
22 ... Main memory 29 ... Mouse 30 ... Recording medium 31 ... Text mining program 32 ... Text data 41-45 ... Window 51 ... Overall co-occurrence network 52-54 ... Limited co-occurrence network 61 ... Analysis button 62 ... Mouse cursor 63-64 … Tab 71-75… screen

Claims (20)

It is a text mining method that displays a screen containing the analysis results of text data.
Steps to extract words from text data,
Steps to generate a co-occurrence matrix for the word,
Steps to generate a co-occurrence network based on the co-occurrence matrix,
A step of displaying a screen including the co-occurrence network is provided.
When an instruction to specify a word of interest is input in the first screen including the first co-occurrence network based on the entire specified text data, the step of extracting the word is the step of the specified text data. In the step of extracting the word from the limited text data including the part including the word of interest and generating the co-occurrence matrix, a second co-occurrence matrix is generated for the word using the limited text data, and the co-occurrence network is formed. The generation step generates a second co-occurrence network based on the second co-occurrence matrix, and the step of displaying the screen displays a second screen including the second co-occurrence network. Text mining. Method. By selecting one or more nodes included in the first co-occurrence network in the first screen and selecting start analysis, an instruction to specify the word corresponding to the node as the attention word is input. The text mining method according to claim 1, wherein the text mining method is characterized by the above. By continuously selecting one node included in the first co-occurrence network in the first screen, an instruction for designating the word corresponding to the node as the attention word is input. , The text mining method according to claim 1. By continuously selecting one edge included in the first co-occurrence network in the first screen, an instruction to specify a word corresponding to two nodes connected to the edge as the attention word is given. The text mining method according to claim 1, wherein the text mining method is input. By selecting one or more edges included in the first co-occurrence network in the first screen and selecting start analysis, the word corresponding to the plurality of nodes connected to the edge is referred to as the attention word. The text mining method according to claim 1, wherein an instruction specified as is input. When a merge instruction is input in a second screen including a plurality of second co-occurrence networks, the step of displaying the screen is characterized in that the plurality of second co-occurrence networks are displayed in a tab format. , The text mining method according to claim 1. The text mining according to claim 6, wherein the merge instruction is input by grasping one second co-occurrence network in the second screen and separating it in another second co-occurrence network. Method. The text mining method according to claim 1, wherein the limited text data is composed of a sentence including the attention word among the designated text data. The text mining according to claim 8, wherein the limited text data when a plurality of attention words are designated comprises a sentence including all of the plurality of attention words among the designated text data. Method. The text according to claim 8, wherein the limited text data when a plurality of attention words are designated comprises a sentence including any one of the plurality of attention words among the designated text data. Mining method. The text mining method according to claim 1, wherein the step of generating the co-occurrence matrix is characterized by generating a co-occurrence matrix having a Jaccard coefficient as an element. A text mining program for displaying screens containing analysis results of text data.
Steps to extract words from text data,
Steps to generate a co-occurrence matrix for the word,
Steps to generate a co-occurrence network based on the co-occurrence matrix,
The CPU causes the computer to execute the step of displaying the screen including the co-occurrence network by using the memory.
When an instruction to specify a word of interest is input in the first screen including the first co-occurrence network based on the entire specified text data, the step of extracting the word is the step of the specified text data. In the step of extracting the word from the limited text data including the part including the word of interest and generating the co-occurrence matrix, a second co-occurrence matrix is generated for the word using the limited text data, and the co-occurrence network is formed. The generation step generates a second co-occurrence network based on the second co-occurrence matrix, and the step of displaying the screen displays a second screen including the second co-occurrence network. Text mining. program. By selecting one or more nodes included in the first co-occurrence network in the first screen and selecting start analysis, an instruction to specify the word corresponding to the node as the attention word is input. The text mining program according to claim 12, wherein the text mining program is characterized in that. By continuously selecting one node included in the first co-occurrence network in the first screen, an instruction for designating the word corresponding to the node as the attention word is input. , The text mining program according to claim 12. By continuously selecting one edge included in the first co-occurrence network in the first screen, an instruction to specify a word corresponding to two nodes connected to the edge as the attention word is given. The text mining program according to claim 12, wherein the text mining program is input. By selecting one or more edges included in the first co-occurrence network in the first screen and selecting start analysis, the word corresponding to the plurality of nodes connected to the edge is referred to as the attention word. 12. The text mining program according to claim 12, wherein an instruction is input. When a merge instruction is input in a second screen including a plurality of second co-occurrence networks, the step of displaying the screen is characterized in that the plurality of second co-occurrence networks are displayed in a tab format. , The text mining program according to claim 12. The text mining according to claim 17, wherein the merge instruction is input by grasping one second co-occurrence network in the second screen and separating it in another second co-occurrence network. program. A text mining device that displays a screen containing the analysis results of text data.
A word extractor that extracts words from text data,
A co-occurrence matrix generator that generates a co-occurrence matrix for the word,
A co-occurrence network generation unit that generates a co-occurrence network based on the co-occurrence matrix,
It is provided with a screen display unit that displays a screen including the co-occurrence network.
When an instruction to specify a noteworthy word is input in the first screen including the first co-occurrence network based on the entire designated text data, the word extraction unit uses the designated text data to indicate the noteworthy word. The word is extracted from the limited text data including the portion including, the co-occurrence matrix generation unit generates a second co-occurrence matrix using the limited text data for the word, and the co-occurrence network generation unit generates the first co-occurrence network. A text mining device, characterized in that a second co-occurrence network is generated based on a two co-occurrence matrix, and the screen display unit displays a second screen including the second co-occurrence network. When a merge instruction is input in a second screen including a plurality of second co-occurrence networks, the screen display unit displays the plurality of second co-occurrence networks in a tab format. Item 19. The text mining apparatus according to Item 19.

Images