JP2008084151A - Information display device and information display method - Google Patents

Abstract

It is not easy to obtain necessary information from document data.
An information display device classifies a document set by first and second classification methods set by a user. The matrix 50 displayed on the information display device includes a column classification item column 52 and a row classification item column 54 for displaying the classification items when classified by the first and second classification methods, and the classified row and column classification. It includes a graphic display field 56 that represents the numerical value related to the product set of the document set by the color of dots on the two-dimensional matrix. If the set classification method is clustering, representative words and phrases selected according to a predetermined criterion from words and phrases extracted from each document when clustering is performed are displayed as classification items.
[Selection] Figure 2

Description

  The present invention relates to an information display technique, and more particularly to an information display technique for visualizing and displaying information obtained from accumulated document data and an information display method applied thereto.

  The technological environment in the information processing field such as computers and networks has made dramatic progress in recent years. Along with this, it has become easy to store a large amount of various data in a storage device or obtain it via a network or a recording medium. Therefore, the emphasis of the issues in the information society is shifting from the point of how to obtain information to the point of efficiently selecting necessary information from a huge amount of information.

In response to such a problem, various techniques for narrowing down necessary data from various databases have been developed. For example, in publications of patent applications, etc., a technique has been proposed in which data is narrowed down by performing aggregation based on keywords, patent classifications, etc., and displaying a matrix map for publications that hit an input search expression (for example, Patent Document 1). Further, in the knowledge management system, a technique for narrowing down the number of posted articles to be acquired by two-dimensionally displaying the number of posts or the degree of evaluation value for each community or specialized field has been proposed (for example, Patent Document 2).
JP 2005-165858 A JP 2005-85017 A

  Since the technique as described above is targeted for data to which classification and attributes are added in advance on the premise of search, it can be realized only by constructing a system according to such a data structure. However, in such a case, the system cannot be used for data having a different data structure or for another purpose, and in the case of the system in the above technique, it is limited to a purpose of use such as searching for a public gazette or searching for a posted article. Is done.

  The present invention has been made in view of such circumstances, and an object of the present invention is to provide a technique capable of easily and intuitively acquiring a variety of information desired by a user.

  One embodiment of the present invention relates to an information display device. This information display device forms a two-line document set group by classifying a plurality of documents stored in a storage unit and a plurality of documents stored in the storage unit by a first classification method and a second classification method. And the correlation between the classification results obtained by the first classification method and the second classification method performed by the classification processing unit, the two series of document set groups are expanded into rows and columns, and the product of the document set groups is obtained. And a matrix display unit that displays numerical information relating to the set as a two-dimensional matrix expressed in a predetermined figure.

  Here, the “first classification method” and the “second classification method” may be the same classification method. Therefore, the “two-line document set group” may be the same document set group. Further, “expressed as a graphic” is a change in a geometrical shape such as a circle, polygon, line, etc., depending on any one of color, pattern, size, or a combination thereof. Alternatively, the geometric shape itself may be changed or the geometric shapes may be combined.

  Another aspect of the present invention also relates to an information display device. The information display device includes a storage unit that stores a plurality of documents, a classification processing unit that forms a plurality of document sets by classifying the plurality of documents stored in the storage unit by a predetermined classification method, and a classification processing unit. Based on the number of occurrences of the phrase extracted from each classified document set, the numerical information related to the phrase is calculated, and the correlation between the result of classification performed by the classification processing unit and the numerical information related to the phrase is calculated based on the numerical value related to the phrase. And a matrix display unit that displays information as a two-dimensional matrix expressed in a predetermined figure.

  Another aspect of the present invention relates to an information display method. The information display method includes a step of receiving a selection input of a first classification method and a second classification method for classifying a plurality of documents from a user, and a plurality of selected first classification method and second classification method. The step of classifying a document to form a two-series document set group and the correlation between the classification results obtained by the first classification method and the second classification method are expanded into two rows and columns, and the document is expanded. Displaying numerical information related to the intersection set of the set groups as a two-dimensional matrix expressed in a predetermined figure.

  It should be noted that any combination of the above-described constituent elements and a representation of the present invention converted between a method, an apparatus, a system, etc. are also effective as an aspect of the present invention.

  According to the present invention, a user can easily and intuitively obtain desired information from document data.

  FIG. 1 shows the configuration of the information display device in this embodiment. The information display device 10 includes an input unit 20 for a user to input instructions for display, a storage unit 12 that stores document data, a classification processing unit 14 that classifies document data using a predetermined classification method, and two series of classification items. Is included in a row and a column, and a matrix display unit 22 for displaying numerical values constituting the respective elements in a two-dimensional matrix. The matrix display unit 22 includes a matrix generation unit 16 that acquires numerical data based on the classified document data and generates display data of the matrix, and a display unit 18 that displays the matrix. The above components are connected via a bus 24, and send and receive data to and from each other.

  In FIG. 1, each element described as a functional block for performing various processes can be configured by a CPU, a memory, and other LSIs in terms of hardware, and a program having a language processing function in terms of software. Etc. Therefore, it is understood by those skilled in the art that these functional blocks can be realized in various forms by hardware only, software only, or a combination thereof, and is not limited to any one.

  The input unit 20 may be any one of commonly used input devices such as a keyboard, a mouse, a trackball, a trackpad, or a combination thereof. The input unit 20 allows the user to specify a document set to be processed from the document data stored in the storage unit 12, and on the matrix display screen displayed on the display unit 18, the classification items constituting the matrix rows and columns are displayed. The type and type of data to be displayed are selected, or a desired area on the displayed matrix is selected.

  The storage unit 12 may be any one of a reading device for a recording medium such as a hard disk, a memory, a DVD (Digital Versatile Disk), a CD (Compact Disk), or the like, or a combination thereof. Select from hardware as appropriate. Therefore, the number is not limited. A part of the storage unit 12 may be connected to the bus 24 via a network (not shown). In this case, other functional blocks may transmit / receive data to / from the storage unit 12 via a server (not shown) connected to the network.

  The document data stored in the storage unit 12 may be, for example, data in which text data that is a document body is associated with attributes that characterize the document, such as a creation date, a creator, and a classification code. Only certain text data may be used. Here, the document may be composed of a long sentence such as a newspaper article or a patent publication, or may be composed of a single sentence or a single word. In addition to document data, the storage unit 12 also includes data necessary for matrix display, such as template data for images to be displayed, data related to graphics displayed on the matrix, and parameters for calculating numerical values represented by the graphics. Remembered.

  The classification processing unit 14 reads out the data of the document set designated by the user through the input unit 20 from the storage unit 12, and classifies the data according to the classification method corresponding to the type of classification items constituting the rows and columns of the matrix selected by the user. I do. As a classification method, a method for classifying each attribute stored in association with each document in advance, a method for assigning each document according to a predetermined criterion to one of a set of words and phrases prepared in advance, and a specific classification item are not provided. Any method such as a method of classifying documents based on similarity between documents (hereinafter referred to as clustering) may be used. The classified document set is stored in the memory of the storage unit 12 in association with each classification item based on the identification information given for each document.

  In the present embodiment, the classified results are displayed in a matrix format, but the classification method performed by the classification processing unit 14 is not necessarily two. That is, the result of classification by one classification method may be displayed in the same way in rows and columns. By selecting the type of classification item, the user can efficiently obtain a desired item from various information contained in the document set. Specific examples of classification items, corresponding classification methods, and information obtained by matrix display will be described later.

  The matrix generation unit 16 of the matrix display unit 22 arranges the results classified by the classification processing unit 14 in rows or columns according to the classification item selected by the user in the input unit 20 and generates display data of a two-dimensional matrix. . For example, let us consider a case in which it is desired to display the number of documents for each classification item of the processing target document set read from the storage unit 12. At this time, the matrix generation unit 16 first generates a matrix having as an element a product set of combinations of two series of document set groups, which is a result of classification into the classification items selected as rows and columns. That is, the document set group forming a row is {M1, M2,..., Mm} (m is the number of items), and the document set group forming a column is {N1, N2,. ), A matrix like Equation 1 is generated. Then, the identification information of documents belonging to each product set and the corresponding combination of two classification items of rows and columns are stored in a memory or the like in association with each other.

  Next, the matrix generation unit 16 performs an operation on the product set, which is each element of the matrix, according to the type of data to be displayed selected by the user using the input unit 20, and digitizes the matrix. In the above example, the final numerical matrix is obtained by counting the number of documents belonging to the product set.

  Further, the matrix generation unit 16 visualizes each element of the numerical matrix as a graphic. Thereby, even if there are many classification items, the user can grasp a numerical distribution and a tendency more intuitively. As will be described later, the matrix generation unit 16 rearranges the rows and columns of the matrix and narrows down the classification items to be displayed, either automatically or according to a user instruction. Therefore, the user can efficiently narrow down data for acquiring local information in addition to grasping the overall trend.

  Depending on the type of data to be displayed, the matrix generation unit 16 determines the number of appearances of phrases including impression expression words extracted from the document set classified by the classification processing unit 14 and phrases such as noun phrases, adjective phrases, and verb phrases. Count for each document set. If necessary, the degree of effect is also calculated based on the number of appearances. The phrase extraction process itself may be performed by an external processing device. In this case, data in which the extracted phrase is associated with each document is stored in the storage unit 12. The matrix generation unit 16 counts the number of appearances of words by referring to the data. Specific types of extracted words and types of display data will be described in detail later.

  The display unit 18 displays display data of the generated matrix on the screen under the control of the matrix generation unit 16, and the user designates a document set to be processed in the input unit 20 and sets the type of classification item. Displays a reception screen when making selections. The matrix display screen and the reception screen may be displayed together on the same screen. The display unit 18 also displays two vertical and horizontal guide lines that move on the displayed matrix. The user can select an area on the matrix by moving the intersection of the guide lines using the input unit 20. When a region on the matrix is selected, the matrix generation unit 16 causes the display unit 18 to display numerical information represented by the graphic existing in the region and information related to the classification item to which the graphic belongs.

  The display unit 18 further displays the text of the document set corresponding to the graphic existing in the area on the matrix selected by the user. The text of the document set may be displayed on another window superimposed on the matrix display window, or may be displayed by switching to the matrix display area.

  FIG. 2 shows an example of a matrix displayed on the display unit 18. The matrix 50 includes a column classification item field 52, a row classification item field 54, and a graphic display field 56. In the example of FIG. 2, all of the column classification item column 52 and the row classification item column 54 are “camera”, “case”, “size”, “shutter”, “lens”, “battery”,. , “Battery” is displayed. For example, correlation of classification results or classification technique when a set of documents related to a camera-related article is classified into these classification items set in advance by two classification methods or two classifiers according to the subject matter. For example, when the tendency of the classifier is grasped, the display as shown in FIG. 2 is useful.

  In the graphic display field 56, dots are displayed as graphics representing the numerical values of the elements of the matrix. In the following description, the numerical values are represented by dots, but the same applies to figures having other shapes. In FIG. 2, the dots are all the same size and have different colors according to the numerical values. However, here, for convenience of display, the color change is represented by a pattern change. For example, the denser the pattern, the higher the value. An element in which no dot is displayed is a case where a numerical value is 0 or a numerical value is not calculated.

  In FIG. 2, it can be seen that the numerical value is high in the combination of the same classification items such as the elements constituting the diagonal of the matrix, that is, (“camera”, “camera”), (“case”, “case”). Furthermore, in this example, the column of the classification item “lens” has a larger number of dots than the other columns. If this result is applied to the case where the two classification methods are used to classify the same classification item and the number of documents belonging to the respective classification items is displayed, the classification tendency is approximately even in the two classification methods. It can be understood that there is no difference, and that documents classified as “lens” by the classification method that constitutes the columns are dispersed by the classification method that constitutes the rows.

  FIG. 3 shows another example of the matrix displayed on the display unit 18. In the matrix 50 in this example, the category items “2004”, “2005”, and “2006” are displayed in the column category item column 52, and the categories “camera” and “mobile phone” are listed in the row category item column 54. Items are displayed. In the example of FIG. 2, the dots displayed in the graphic display field 56 have the same size, but in FIG. 3, the size of the dots is different. Furthermore, although the colors of the dots are different, the patterns are expressed with different patterns for convenience as in FIG. That is, in the example of FIG. 3, two numerical values are represented by the size and color of one dot.

  FIG. 3 shows, for example, a matrix in the case where the number of documents related to “camera” or “mobile phone” is represented by the size of a dot with respect to the creation year and the rate of change in the number of documents from 2004 is represented by a dot color. It is. At this time, the number of documents related to “camera” is smaller over the entire period than the number of documents related to “mobile phone”, but it can be understood that the increase rate is higher for documents related to “camera”.

  Next, the operation of the information display apparatus 10 having the above configuration will be described. FIG. 4 is a flowchart showing a matrix display processing procedure by the information display apparatus 10. First, the user designates a document set to be processed through the input unit 20 (S10). For example, a reception screen for input on the display unit 18, that is, a screen for selecting from various document sets stored in the storage unit 12, is displayed, and the user selects and inputs the screen. Alternatively, a recording medium such as a CD-ROM storing the document set may be read into the storage unit 12 or a download instruction may be given from the server to the storage unit 12 via a network.

  Next, the user selects the type of classification items constituting the rows and columns of the matrix and the type of data to be displayed as the matrix (S14). Also in this step, the user may select and input the classification items that can be selected for the document set selected in S10 on the display unit 18. At this time, when classifying according to the attribute of the document, the type of attribute such as “gender” and “creation date” is displayed as a selection item. On the other hand, when it is desired to display the result of clustering, since the classification item is unknown, a selection item such as “clustering” is displayed. Similarly, when it is desired to express a numerical value for each extracted phrase, the type of phrase is displayed as a selection item because the extracted phrase is unknown. For the type of data, for example, a name of a numerical value that can be calculated such as “number of documents” or “ratio of the number of documents” is displayed as a selection item. You may display the screen which receives the detailed setting about each selection item as needed.

  The classification processing unit 14 reads out the document set designated in S10 from the storage unit 12, and executes classification based on the type of the classification item selected in S14 (S16). If the read document set is already classified in the storage unit 12, and the user has selected only the classification method, the process of S16 is skipped.

  As described above, the matrix generation unit 16 of the matrix display unit 22 forms a product set of two series of document sets constituting rows and columns for each combination of items, and sets the type of display data selected in S14. Based on this, a numerical value for each product set is calculated. Alternatively, the phrase corresponding to the type of phrase selected in S14 is extracted for each document set classified by the classification processing unit 14, and the number of appearances or a numerical value based on the number of appearances is calculated. Then, the color and size of the dots displayed on the matrix are determined based on the numerical values, and matrix data including the dots and the classification items of the rows and columns is generated (S18). The display unit 18 outputs the matrix data on the screen (S20).

  At this time, the display unit 18 always displays a reception screen for selecting and inputting the type of classification item and the type of display data in addition to the matrix. If the user inputs to select a new category item type while displaying the matrix (Y in S22), the classification processing unit 14 performs a new classification based on the category item type (S16). ), The matrix generation unit 16 regenerates the matrix data (S18). When the user inputs to select a new display data type (Y in S24), the matrix generation unit 16 calculates a new numerical value based on the display data type and regenerates the matrix data (S18). ). Through these processes, the display unit 18 displays a new matrix according to the type of classification item selected by the user or the type of display data (S20).

  The above processing is performed until the user inputs an instruction to end the matrix display (N in S26), and the matrix display is ended by inputting the end instruction (Y in S26).

  FIG. 5 shows a configuration example of a screen displayed on the display unit 18. The matrix display screen 60 includes a document set designation area 66, a matrix display area 51, a classification item selection area 62, a legend display area 64, a sort instruction button 67, and a narrowing instruction button 68. In S10 of FIG. 4, the user designates the document set to be processed by inputting the storage location and the document set name in the storage unit 12 to the document set designation area 66. As in a general document specifying method, the tree structure of the storage unit 12 may be referred to and selected from there.

  The matrix display area 51 includes a column classification item display area 53, a row classification item display area 55, and a graphic display area 57. In the category item selection area 62, row category items and column category item types that can be displayed as a matrix, and display data type candidates are displayed. In this figure, “Cluster name”, “Region”, “Occupation” are displayed as the types of classification items in the row or column, and “Quantity”, “Percentage”, “Text” are displayed as the types of data to be displayed. Yes. Here, “body” is selected as a function of the present embodiment when the user designates a certain document set represented by dots and the body of the document belonging to the document set is displayed as text data. .

  In S14 of FIG. 4, the user selects the type of classification item to be displayed in the matrix row and column from the classification item selection area 62, and the item is displayed in the column classification item display area 53 using the pointing device or the like as the input unit 20. The type of the classification item is determined by performing a drag-and-drop operation to the classification item display area 55 of the line. Similarly, the type of data is selected from the classification item selection area 62, and the type of display data is determined by performing a drag and drop operation on the graphic display area 57. When generating matrix data in S <b> 18 of FIG. 4, the matrix generation unit 16 determines the color and size of the dots according to the calculated numerical value range of each element, and displays the legend in the legend display area 64.

  The sort instruction button 67 and the narrow-down instruction button 68 are selected by the user when the matrix rows or columns displayed in the matrix display area 51 are exchanged or the number of displays is reduced according to predetermined criteria. These functions will be described later.

  With the screen configuration described above, the user can set conditions visually and intuitively, and can easily obtain desired information even if there are various forms of documents and various classification methods. it can. Note that the screen shown in FIG. 5 is an example, and the present embodiment is not limited to this. For example, depending on the type of classification item and the type of display data, more detailed settings may be required. At this time, the setting is performed by displaying another window in an overlapping manner or adding an option to the classification item selection area 62 as necessary. In any case, a screen configuration that can be selected intuitively using a pointing device is desirable.

  Here, one mode when selecting the type of the classification item will be described. FIG. 6 shows an example of a matrix displayed in the matrix display area 51 when the row classification items have a hierarchical structure. This example categorizes articles about cameras by their modified noun phrases such as “case”, “size”, “lens”, etc., and further modifies those phrases as “solid”, “heavy”, “ It shows the case of further subdividing and classifying by adjective phrases such as “durable”. That is, the classification item at this time has a hierarchical structure in which the modified noun phrase is an upper layer and the adjective phrase is a lower layer. On the other hand, the column classification item has a single-layer structure that classifies by model, such as “model A”, “model B”, and “model C”.

  In such a situation, candidates such as “camera phrase (upper / lower)”, “camera phrase (upper)”, and “camera phrase (lower)” are displayed in the classification item selection area 62, for example. Here, the “camera word / phrase” is a name assigned to a group of classification items prepared in advance, in this case, a classification item having a hierarchical structure. A display such as “noun phrase / adjective phrase” may be used. When clustering is performed, “clustering (upper / lower)” or the like may be used.

  When a “camera word / phrase (upper / lower)” is dragged and dropped to the line category item display area 55, the matrix display area 51 displays the category items of the upper and lower layers as shown in FIG. Is displayed as a hierarchical structure. Therefore, for example, the numerical value represented by the dot is a numerical value related to a product set of a document set belonging to “Case” and a document set belonging to “Model A”. In this example, it is possible to grasp, for each model, whether there are many articles whose content is “what (modified noun phrase)” is “how (adjective phrase)”.

  On the other hand, when “camera word / phrase (upper order)” is selected, the upper layer classification items are displayed in the matrix display area 51 as shown in FIG. The numerical value represented by the dot at this time is, for example, a numerical value related to a product set of a document set belonging to “case” and a document set belonging to “model A”. As a result, it is possible to grasp “what (the modified noun phrase) has many articles related to each model.” Similarly, when “camera phrase (subordinate)” is selected, FIG. As shown, lower layer classification items are displayed. At this time, the documents belonging to the lower-level classification item “heavy” regardless of the higher-level classification items are aggregated to obtain a document set, and a numerical value relating to the product set of the document set belonging to “model A” is displayed. . As a result, regardless of “what (modified noun phrase)”, it is possible to grasp for each model how many articles are described.

  Although the classification item hierarchy is two layers in the example of FIG. 6, display switching is performed in the same manner for three or more layers. As described above, since display data can be switched only by a pointing device, even if the classification item has a hierarchical structure, it is possible to efficiently perform the process from grasping the overall trend to detailed analysis. In addition, the results due to the difference in hierarchy can be easily compared.

  When the user presses the sort instruction button 67, the matrix generation unit 16 replaces the rows and columns of the matrix displayed in the matrix display area 51, and generates sorted matrix data. Depending on the classification item, a submenu for selecting column sorting, row sorting, column and row sorting may be further displayed. In addition, the criteria for sorting may be selected from the submenu. Sorting criteria include (1) sorting by category item name, (2) sorting by total value, (3) sorting by percentage, (4) sorting by degree of dispersion, depending on the type of classification item and the type of display data. 5) Select from diagonalized sort.

  Sorting by classification item name is performed based on the character code of the character string of the classification item, for example, in the order of AIUEO. Sorting by the total value is performed, for example, in descending order based on the total value of the numerical values of the elements constituting each column or each row. The sorting based on the ratio is performed based on the ratio of the numerical value of each element to the total value of the numerical values of the elements constituting each column or each row. The sorting based on the degree of dispersion is performed based on whether or not the values are distributed to more classification items. For example, a definition such as “the number of classification items in which a value (dot) exists in a certain row (or column) / the number of classification items belonging to that row (or column)” is used.

  In the diagonalization sort, attention is paid to the value of the element on the diagonal line of the matrix, and a case where only one of the row and the column is sorted and a case where both the row and the column are sorted are prepared. FIG. 7 schematically shows how diagonalization sorting is performed. In FIG. 7A, the order of the classification items displayed in the row classification item column 54 is fixed, and the values of the elements in the diagonal area 70 are displayed in the column classification item column 52 so as to be the largest. This is an example in which the display order of the classification items is changed and sorting is performed. By focusing on the diagonal region 70, the trend may be grasped only by comparing the arrangement order of the row and column classification items.

As a method for sorting only one of the rows or columns in the diagonalization sort, the following higher priority one-side diagonal sort algorithm can be considered. Although the case where the row is fixed and the column is sorted is described here, the case where the column is fixed and the row is sorted can be similarly realized by replacing “row” and “column”.
(1) Start processing from i = 1 row for square matrix T of n rows and n columns (2) Among elements ti1,..., Tin of row vector ti of T, i ≦ j ≦ n and the largest (3) If i ≠ j, replace the i-th column with the j-th column of T. (4) If i <n, set i = i + 1 and repeat the process from (2).

However, in the above algorithm, since the position of the column determined in the upper row cannot be changed in the lower row, the maximum element may not be located diagonally in the lower row. Therefore, a maximum value priority one-sided diagonal sort algorithm that performs diagonalization in order from the element having the maximum value in the square matrix T as described below may be employed.
(1) Prepare a list L = {} for storing determined matrix numbers (2) Find the maximum element tij of T for which i ≠ lk and j ≠ lk for all matrix numbers lk in L (3) When i ≠ j, the i-th column and the j-th column of T are exchanged. (4) When | L | <n, L = L∪ {i} and the process is repeated from (2).

In FIG. 7B, the values of the elements in the diagonal area 70 are the largest, and the values are displayed in the line classification item column 54 so that the values are in descending order from the upper left to the lower right of the diagonal area 70. This is an example in which sorting is performed for both the classified items displayed and the classified items displayed in the column classified item column 52. As an algorithm for realizing such a sort, there is a double-sided diagonal sort shown below.
(1) Start processing from k = 1 for a square matrix T of n rows and n columns (2) Find an element tij having the maximum value among all T elements for which k ≦ i and k ≦ j holds ( 3) When k ≠ i, the kth and ith rows of T are switched. (4) When k ≠ j, the kth and jth columns of T are switched. (5) When k <n, k = Repeat the process from (2) as k + 1

  When the user presses the narrowing down instruction button 68, the matrix generation unit 16 generates matrix data in which the classification items to be displayed in the matrix display area 51 are narrowed down. The number of classification items displayed as a result of narrowing down may be a fixed value, or may be automatically determined by providing a threshold value for the numerical value calculated in sorting. Further, the user may be allowed to designate by displaying a gauge or the like that allows the user to set the number with a pointing device.

  Similar to the sort instruction button 67, the user may be able to select a method for narrowing down in the submenu. The sorting algorithm as described above may be used for narrowing down. For example, when the narrowing-down method of “dispersion degree (ascending order)” is selected in the submenu, the above-mentioned dispersion degree is calculated for each row (or column), and a predetermined number of rows (or columns) whose values are lower. ) Only. You may make it delete the row | line | column and column which do not have the dot which should be displayed, ie, a document set does not exist.

  The matrix generation unit 16 adds and displays text data representing specific information related to the document set represented by each dot on the matrix in the matrix display area 51 on the matrix display screen 60 shown in FIG. FIG. 8 shows a matrix display screen 60 when information related to a document set represented by dots is displayed. In the state where the matrix is displayed in the matrix display area 51, when the area of the dot 83 is designated by the pointing device, the horizontal guide line 80 and the vertical guide line 82 intersecting with the dot are displayed. With these guide lines, the user can grasp to which classification item the designated dot belongs.

  Further, a horizontal direction information display area 86 is displayed in the vicinity of the horizontal direction guide line 80, and a vertical direction information display area 84 is displayed in the vicinity of the vertical direction guide line 82 in a pop-up window or the like. In the horizontal direction information display area 86, the classification item name of the line to which the designated dot belongs, the specific numerical value represented by the dot, the total value of the numerical values of the elements belonging to the classification item, and the like are displayed. In the example of FIG. 8, the classification item name is “camera”, the number of documents represented by dots is “6”, and the total number of documents is “21”. Similar information is also displayed in the vertical direction information display area 84. The “number of documents” is replaced with a ratio, the number of occurrences of phrases, and the like depending on the type of data to be displayed.

  With such a screen configuration, even if there are many classification items and the matrix 50 becomes a complicated figure, it is possible to immediately know which classification item each dot represents. In addition, it is possible to acquire a local numerical value with an easy operation while grasping the overall tendency with dots. As in the case of designating the dot area, by designating each classification item, information on the classification item, for example, the number of all documents belonging to the classification item, information obtained in the classification process, and the like are displayed. You may make it do.

  In a state where a certain dot 83 area is selected, the user further selects a candidate “text” 62 a displayed in the classification item selection area 62, thereby displaying the text of the document set represented by the dot 83. A display example of the text will be shown later. At this time, the matrix generation unit 16 specifies identification information of documents belonging to the product set stored in a memory or the like when the matrix is generated based on the classification item. Then, the data for displaying the text is generated by pasting or linking the text data read from the storage unit 12 based on the identification information to the template data for display prepared separately. When there are a large number of applicable documents, functions for scrolling and paging are provided as appropriate. By selecting each category item instead of selecting a dot, the texts of all documents belonging to the category item may be displayed.

  Further, after displaying the text, the data can be saved in, for example, the csv format. As a storage instruction input and storage procedure, a general method for data storage can be used. By displaying and saving the text of a desired document set in this way, the user can check not only the numerical aspects of the classification result but also the actual state of the document. Even when a document contained in a huge database is repeatedly searched by a search query by following the steps of first classifying and checking the trend with dots and finally obtaining the desired document itself In comparison, it is possible to reach the desired document much more efficiently.

  Next, a classification method provided by the information display device 10 in the present embodiment and information obtained by displaying it as a matrix will be exemplified. As described above, the classification method includes a method of classifying each attribute when the attribute originally associated with the document exists, a method of assigning the document to a predetermined classification item by a predetermined method, and similarity between documents. One example is clustering that generates a group of documents (cluster). Here, a classification method for a predetermined classification item and a clustering method will be briefly described. However, the classification method in the present embodiment is not limited to that described here, and any of the generally proposed methods may be selected.

(Classification into predetermined classification items)
In this classification method, classification items (hereinafter referred to as categories) and related word groups (hereinafter referred to as profiles) are prepared in advance, and similarities between documents and categories are determined by word groups and profiles extracted from each document. This is a method of determining the degree and assigning the document to a category having a high degree of similarity. For example, when it is desired to classify newspaper articles into the categories of “politics”, “economy”, and “sports”, a profile is prepared that includes words such as “baseball”, “soccer”, and “game” related to “sports”. Each word constituting the profile is weighted according to its importance.

  The specific method is as follows. That is, a phrase is first extracted from a processing target document by morphological analysis. And unify the notation of words with the same content but different notations. Next, based on the weight vector of those words and the weight vector of the profile corresponding to the arrangement order, the similarity between the processing target document and each category is calculated using the vector space method. The former weight vector is A = {w_a1, w_a2, w_a3,..., W_aN}, and the latter weight vector is B = {w_b1, w_b2, w_b3,. Here, N is the number of words / phrases extracted from the document to be processed, and w is the importance of words / phrases derived by the TF-IDF (Term Frequency-Inverse Document Frequency) method. At this time, the similarity sim (B, A) is as follows.

  The similarity sim (B, A) is calculated for each category, and the processing target document is classified into the category having the highest similarity.

(Classification by clustering)
Clustering is a method of forming a cluster of documents based on similarity between documents without preparing a predetermined classification item. Clusters can also be viewed as topics, that is, topics. In this case as well, words are extracted by morphological analysis for each document, and the expressions of the words having the same contents are unified. Then, word weight vectors for all documents are generated in a corresponding order, and the similarity is calculated for each combination of documents. This similarity may also be the value given by Equation 2. After that, for example, when the degree of similarity exceeds a certain threshold, the documents are considered to be similar, and the same cluster is generated.

  Among the phrases extracted from the document, a phrase that characterizes the cluster is extracted as a representative phrase of the cluster. Also, the most characteristic phrase among the representative phrases is determined as the cluster name. For example, the mutual information with the cluster is calculated for each word, and several words / phrases with the highest value are used as representative words / phrases and the word / phrase with the highest value is used as the cluster name. The mutual information MI (t, C) between the word t and the cluster C is expressed by the following equation.

  Here, P (t) is a probability that a document including the word t appears, and is defined by (the number of documents including the word t / the total number of documents). P (C) is the probability that a document belonging to cluster C will appear, and is defined by (number of documents belonging to cluster C / total number of documents). P (t, C) is a probability that a document belonging to the cluster C including the word t appears, and is defined by (the number of documents of the cluster C including the word t / the total number of documents). The cluster names determined in this way are displayed in the column classification item column 52 and the row classification item column 54 of the matrix 50 in FIG.

  Next, specific examples and effects of applying each classification method to the display format of the matrix 50 will be described. In the description of each example, the matrix is represented by a table to simplify the illustration, but each numerical value in the table is represented by a dot on the matrix. The number of rows and columns constituting the matrix in each table is 2 to 3 for simplicity, but is not limited thereto.

(Profile diagnosis using clustering)
In Table 1, the classification items of the rows are “category 1” and “category 2” which are prepared categories, and the cluster classification items are “cluster A” and “cluster B” which are cluster names obtained as a result of clustering. Is a matrix. The purpose of this matrix display is to confirm whether an appropriate profile and category relationship is set. That is, it examines whether or not different categories are assigned to similar documents by examining the criteria for assigning the original document category when preparing the profile and the category.

  The clustering performed at this time is set to be the same as the number of categories. In such a matrix, numerical values 1 to 4 are represented by dots. For example, as the numerical value 1, the number of documents nA_1 of the document set classified as “cluster A” among the documents of “category 1” is displayed as a result of performing clustering on all original documents when preparing the category. Numerical values 2 to 4 are displayed in the same manner. Alternatively, the ratio of the number of documents nA_1 of the document set belonging to cluster A out of the number of documents n1 of all document sets belonging to category 1, that is, nA_1 / n1 may be set to a numerical value 1.

  By displaying such a matrix, it is possible to discover those classified into different categories in spite of similar documents, and to identify problems inherent in the category assignment criteria.

(Diagnosis of profiles using constituent words)
Table 2 shows a case in which “profile 1”, “profile 2”, and “profile 3”, which are a plurality of profiles prepared corresponding to a plurality of categories, are used for both the row classification item and the column classification item. Matrix. The purpose of this matrix display is to check the similarity between categories. For example, when there are categories “charge” and “battery” and the profiles of these categories are similar, there is a possibility that a document to be classified into the “charge” category may be classified as “battery”. In such a case, the classification accuracy can be improved by performing profile tuning such as integrating similar profiles.

  In such a matrix, numerical values 1 to 9 are represented by dots. For example, as the numerical value 2, the similarity between “profile 1” and “profile 2” is displayed. For example, a word sharing rate is calculated as an index representing the degree of similarity.

  The word sharing rate str is a value defined below as a ratio of sharing a word having a positive weight between the profiles C and D.

  Here, str (C, D) = str (D, C) does not necessarily hold. Further, the word sharing rate represented by the same profile combination, that is, the numerical values 1, 5, and 9, is 1. By changing the way in which the weight of the word i is given in Equation 4, the viewpoint of the word sharing rate can be changed. In general, the weight vector element w i can take a real value from −1 to 1. On the other hand, when the presence / absence of a word / phrase is given to wi as a binary value of 1 and 0, str becomes an index representing the ratio of shared words. Further, when the appearance frequency of a word is given to wi, str becomes an index indicating the ratio of the appearance frequency of the shared word.

  As an index representing the degree of similarity, in addition to the word sharing rate str, a cosine similarity or a cohesion degree may be employed. The cosine similarity is obtained by applying the two profile weight vectors to Equation 2. The cohesion degree rel is a value defined as follows according to the number of words constituting the profile p1 and the profile p2.

  Here, a is the number of words constituting the profile p1, b is the number of words constituting the profile p2, and c is the number of words commonly existing in the profile p1 and the profile p2.

  Note that the cosine similarity and cohesion are the same values for the same profile combination, so a matrix as shown in Table 3, which is a triangular matrix, is used. Here, the element represented by “−” does not display a dot. Also in this case, the combination of the same profiles, that is, the cosine similarity and the word sharing rate represented by the numerical values 3, 5, and 7 are 1.

(Understanding the relationship between correct answer category and classification category)
Table 4 shows the category “category 1” and “category 2” which are the prepared categories for both the row classification item and the column classification item. However, the row classification item is a case where the row classification item is correctly classified by visual observation or the like. The classification item is a case where mechanical classification is performed by a classifier or the like. The former is described as “category 1 (correct answer)”, the latter as “category 1 (classification)”, and the like. The purpose of this matrix display is to visualize the relationship between mechanically classified and correct answers.

  In Table 4, the row and column classification items are the same. In such a matrix, numerical values 1 to 4 are represented by dots. For example, as the numerical value 1, among the document number n1 (correct answer) of the document set classified as “category 1 (correct answer)”, the document number n1 (classification) _ (of the document set classified as “category 1 (classification)” is shown. The correct answer) ratio, n1 (classification) _ (correct answer) / n1 (correct answer) is displayed. Similarly, values 2 to 4 are displayed. At this time, the numerical value 1 and the numerical value 4 can be considered as the recall rate of the classifier for the correct answer.

  Alternatively, the numerical value n1 (correct answer) _ (classification) of the documents classified as “category 1 (correct answer) out of the document number n1 (classification) of documents classified as“ category 1 (classification) ”as the numerical value 1 , N1 (correct answer) _ (classification) / n1 (classification) may be displayed. Numerical values 2 to 4 display similar values. At this time, the numerical value 1 and the numerical value 4 can be considered as the precision of the classification correct answer by the classifier.

  In a state where a matrix representing the reproduction rate is displayed with dots, when the user clicks on the matrix display area 51 with a pointing device which is the input unit 20, the display may be switched to a matrix representing the matching rate. Switching from the precision ratio to the recall ratio may be performed in the same manner. By displaying such a matrix, it is possible to visually grasp which categories are misclassified by the classifiers that are frequently misclassified, and to gain knowledge in developing a profile adjustment policy. Obtainable.

(Time series analysis of classification results)
Table 5 shows time category units such as the creation date and the registration date stored in association with the document, with the category items in the row being “category 1” and “category 2” which are prepared categories. It is a matrix in the case of “time series unit A” and “time series unit B”. The time series unit is a name indicating a certain period such as 2006, 2007..., First half, second half or the like. The purpose of this matrix is to grasp changes in the document over time for each category.

  In this case, as the numerical value 1, the ratio of the document number nA_1 of the document set belonging to the “time series unit A” out of the document number n1 of the document set classified as “category 1”, that is, a value of nA_1 / n1 is displayed. . The same applies to the numerical values 2 to 4. By displaying this value, it is possible to grasp the time when the appearance ratio is high for a certain category. Alternatively, as the numerical value 1, the ratio of the number of documents n1_A of the document set classified into “category 1” out of the number of documents nA of the document set belonging to “time series unit A”, that is, a value of n1_A / nA may be displayed. Good. The same applies to the numerical values 2 to 4. By displaying this value, it is possible to grasp a category having a high appearance ratio for a certain period. Also at this time, the matrix may be switched and displayed when the user clicks on the matrix display area 51 as described above.

  By displaying such a matrix, for example, topics that have increased rapidly at a certain time can be easily grasped, and by analyzing only the document set by text mining, etc., analysis focused on the topic of interest is efficient. Can be done well. Further, by comparing with the time when an event such as an advertisement is performed, the influence of the event on the topic can be grasped. Furthermore, it is possible to follow changes in the number of documents that were not classified into any category, and considering the timing of new profile adjustments based on the trend, or checking the documents and finding opinions that lead to new product proposals You can do it.

  When the matrix in Table 5 is used in the company, each category is assigned to the department in the company in advance, and when the user clicks the classification item display area 55 in the row, the matrix re-calculated for each department to which each category belongs is displayed. You may do it. As a result, it is possible to confirm the change over time of the document number ratio for each department, and for example, it is possible to grasp the tendency of the number of complaints for each department.

  In the same manner as in Table 5, it is possible to display the results of classification according to all attributes such as the sex and occupation of the created person. For example, it is possible to grasp a tendency for an attribute that tends to have a high ratio or a category that tends to have a distribution with a similar ratio for each attribute.

(Understanding clustering results)
Table 6 is a matrix in which the row classification items are cluster names “cluster 1” and “cluster 2”, and the column classification items are cluster names “cluster A” and “cluster B”. Here, a clustering method for generating a row classification item and a clustering method for generating a column classification item are different from each other. The purpose of this matrix is to compare clustering results from different clustering techniques.

  Here, we compared the results when the number of clusters is large and small, or the results of clustering in document units and clustering after dividing into smaller topic units using an external topic divider Compare etc. In this case, for example, as the numerical value 1, for example, the ratio of the document number nA_1 of the document set classified into “cluster A” by another method out of the document number n1 of the document set classified as “cluster 1” by one method, that is, The value nA_1 / n1 is displayed. The same applies to the numerical values 2 to 4.

  By displaying such a matrix, for example, when the number of clusters to be set is increased, it is possible to visually grasp the variation of each cluster, such as a cluster that is distributed over many clusters or a cluster that is not so distributed. At this time, a cluster in which the document set does not change even when the number of clusters to be set is increased may be made inconspicuous by displaying dots in gray. As a result, it is possible to easily grasp only the cluster distribution in which the document set has changed.

  In addition, when a single document includes a plurality of topics, it is possible to check the distribution of the topics. It is possible to grasp what topics are often inquired at the same time when the processing target document is an inquiry sentence from a customer, or what kind of continuity of inquiries is made. it can.

(Understanding the relevance of topics using representative phrases of clusters)
In Table 7, the classification items of the rows are “representative phrases of cluster 1” and “representative phrases of cluster 2” that are representative phrases of each cluster, and the classification items of the columns are “cluster 1” and “cluster 2” that are cluster names. "Is a matrix. For example, the maximum number of representative words of the cluster is set to 5 words. In view of the display space and the like, the notation on the matrix may display only the cluster name even in the row classification item.

  While the above-mentioned grasping of the clustering result is performed by displaying “documents” that belong to “cluster 1” and “cluster A” in common, in this case, the relation of the topic via the “representative phrase” of the cluster The purpose is to understand sex. For example, the ratio of the document set including the representative words / phrases of cluster 1 among the documents belonging to cluster 2 is calculated. Such a document set is a document that mainly mentions the topic of cluster 2 but also touches the topic of cluster 1. That is, when there are many such documents, it can be considered that cluster 1 and cluster 2 are highly related topics. As described above, since the representative word / phrase of a cluster is a word / characteristic characterizing the cluster, it uses the property that it does not often appear in documents belonging to other clusters.

  In the matrix of Table 7, the numerical value 1 is the number of documents n1 (cluster) of the document set classified as “cluster 1” out of the number n1 (phrases) of the document set including at least one representative word of “cluster 1”. The ratio of _ (phrase), that is, a value of n1 (cluster) _ (phrase) / n1 (phrase) is displayed. The same applies to the numerical values 2 to 4.

  A practical example of using such a matrix will be described below. For example, the cluster name of cluster 1 is “finder”, its representative word is “finder / motion / outdoor”, the cluster name of cluster 2 is “liquid crystal”, and its typical word is “liquid crystal / screen / monitor”. . At this time, it is assumed that a high value is indicated by the numerical value 2 in addition to the numerical value 1 that is the value of the cluster 1 itself among the elements of the row having “representative phrase of the cluster 1” as a classification item. In other words, the representative phrase of cluster 1 “finder” appears in a relatively large number in documents belonging to cluster 2 “liquid crystal”.

  Therefore, when a dot area representing the numerical value 2 is selected and the text of the document set is displayed as described above, for example, “It is difficult to see the liquid crystal in“ outdoor ”, so take a picture with an eye like an analog camera”, “ When shooting something with “movement”, you can get a sentence such as “The viewfinder is better than the LCD.” That is, it is assumed that “Finder / motion / outdoor”, which is a representative word of cluster 1 “Finder”, represents the topic of “Finder”, and by looking at the distribution of the word, mainly describing “Liquid Crystal” You can check documents that touch the topic of “Finder”.

  FIG. 9 shows a configuration example of a screen in the display unit 18 when the text of the document is displayed in the matrix configuration of Table 7. The text display screen 98 includes a legend display area 90, a text display area 92, a highlight display area 94, and a document information display area 96. This display is, for example, a case where the text is displayed by selecting a dot with a numerical value 3 that represents a document set classified as cluster 1 “mobile” and including a representative phrase of cluster 2 “large”. At this time, the text display screen 98 highlights which words in the text are representative words of which cluster by color.

  First, in the legend display area 90, a legend of the color of the frame for emphasizing each cluster is displayed. In the figure, a white frame 90a is used for the representative word / phrase of the cluster “mobile”, and a black frame 90b is used for the representative word / phrase of the cluster “large”. Of course, other colors may be used. A text is displayed in the text display area 92, and a text in which a white frame 90a and a black frame 90b for emphasis are applied to the representative words / phrases included in the text are displayed in the highlight display area 94. In the text of the figure, “size” is surrounded by a black frame 90b as a representative phrase of the cluster “large”, and “weight” and “portability” are surrounded by a white frame 90a as representative phrases of the cluster “mobile”. It is displayed. The document information display area 96 displays information stored in association with the displayed document, that is, the file name including the document, the gender of the creator, age, occupation, creation date, and the like. When there are a plurality of documents to be displayed, the pages may be switched over and displayed over a plurality of pages.

  With the above configuration, it is possible to grasp topics that are strongly related to only a specific topic or topics that are widely related to a plurality of topics. Then, it is possible to narrow down and confirm documents having commonality with respect to the points written from the two viewpoints of “topic 1” and “topic 2”. In addition, it is possible to grasp the tendency that “topic 1” and “topic 2” are described in relation to each other by dots indicating numerical values.

(Cluster specific expression analysis)
In Table 8, the row classification items are “cluster 1” and “cluster 2” that are cluster names, and the column classification items are “specific expression category A” and “specific expression category B” that are the category names of specific expressions. A matrix of cases. Here, the unique expression identifies a thing or number such as a proper noun such as a product name, organization name, place name, person name, date, time, period, amount, quantity, URL (Uniform Resource Locator), e-mail address, telephone number, etc. Information. Therefore, “proprietary expression category A” or the like is a specific proper noun or the like or a set thereof. For example, in addition to proper nouns such as “Yamada” and “Tanaka”, “person names” may be used as a set including them. The purpose of this matrix is to grasp how the unique expressions are distributed in each cluster.

  Here, as a numerical value 1, for example, a ratio of the document number nA_1 of the document set from which the “specific expression category A” is extracted out of the document number n1 of the document set classified as “cluster 1”, that is, a value of nA_1 / n1. indicate. The same applies to the numerical values 2 to 4. By displaying such a matrix, it is possible to cross-analyze topics and specific expressions, and to easily obtain knowledge about topics closely related to regions, persons, organization names, etc., and other topics it can.

(Time series analysis of words)
Table 9 uses “word / phrase 1” and “word / phrase 2” which are words / phrases extracted from a document as row classification items, and time series units such as creation date and registration date stored in association with the column classification items in the document. This is a matrix in the case of “time series unit A” and “time series unit B”. The processing for extracting a phrase from a document may be performed by a device external to the information display device 10. At this time, the extracted phrase and each document are stored in association with each other in the storage unit 12 of the information display device 10. The time series unit is the same classification item as described in (Time Series Analysis of Classification Results). The purpose of this matrix is to analyze the words that appear in the document in time series.

  Of the extracted words / phrases, the number of words / phrases to be displayed in the classification item of the matrix row in Table 9, that is, the number of rows, is set in advance. For example, it is assumed that the top 20 words are frequently used. Here, as a numerical value 1, for example, a ratio of the number nA_1 of documents belonging to “time series unit A” out of the number n1 of documents including “word 1”, that is, a value of nA_1 / n1 is displayed. The same applies to the numerical values 2 to 4.

  If the document to be processed is a customer inquiry or complaint about a product, and the text of the document is, for example, “An error has occurred in printing”, the words “print”, “error”, “occurrence” are extracted, This is a line classification item. The matrix in Table 9 displays the change over time in the number of documents containing these words. Thereby, for example, it becomes easy to perform analysis while paying attention to words that have increased rapidly at a certain time, and it becomes possible to quickly grasp changes in customer voices and problems.

(Phrase opinion analysis)
Table 10 uses “noun phrases 1” and “noun phrases 2” which are noun phrases extracted from a document as classification items in a row, and “adjective phrase A” which is an adjective phrase extracted from a document as a column classification item. It is a matrix in the case of “adjective phrase B”. The column classification item may be a verb phrase instead of an adjective phrase, or a combination of an adjective phrase and a verb phrase. Similar to (time-series analysis of words / phrases), processing for extracting noun phrases, adjective phrases, and verb phrases from a document may be performed by a device external to the information display device 10. The purpose of this matrix is to visualize the relationship between extracted noun phrases and adjective phrases or verb phrases, that is, a list of phrases together with the number of documents.

  Here, as a numerical value 1, for example, the ratio of the number of appearances nA_1 of “noun phrase 1” and “adjective phrase A” to the number of occurrences of “noun phrase 1” that depends on the adjective phrase, that is, nA — 1 / n1. Is displayed. The same applies to the numerical values 2 to 4. The same numerical value is also displayed when the column classification item is a verb phrase or a combination of an adjective phrase and a verb phrase. By displaying the display switching button, the matrix can be switched and displayed. If there are a large number of noun phrases that depend on adjective phrases, such as line classification items, the upper limit of the number of noun phrases to be displayed in advance is displayed so that only the noun phrases with the highest number of occurrences n1 are displayed. Set it.

  By displaying such a matrix, it is possible to display a list of dependencies appearing in the document, and to easily grasp the tendency of the points of the document such as what is written and how. For example, if the document to be processed is an answer to a questionnaire about a product, the dependency relationship between the noun phrase “liquid crystal” and adjective phrases such as “difficult to see”, “large”, “small”, etc. is extracted. Displayed as an item. The appearance frequency is visually shown. Thereby, the impression and evaluation about "liquid crystal" can be confirmed easily.

(Analysis of opinion trends using clusters and phrases)
Table 11 shows a matrix in which the row classification items are cluster names “cluster 1” and “cluster 2” and the column classification items are phrases “phrase A” and “phrase B” extracted from the document. It is. The purpose of this matrix is to grasp the relationship between clusters and phrases.

  Here, as a numerical value 1, for example, a ratio of the number of appearances nA_1 of “phrase A” included in the document set to the number of documents n1 of the document set belonging to “cluster 1”, that is, a value of nA_1 / n1 is displayed. The same applies to the numerical values 2 to 4. Phrases that are column classification items can have a hierarchical structure. In this case, as described above, the hierarchy can be switched so that the matrix can be displayed. For example, an aggregated number for each modified noun phrase or adjective phrase is displayed, or the hierarchical structure is expanded to display the modified noun phrase and the adjective phrase together as column classification items. When there are a large number of different phrases, the upper limit of the number of phrases to be displayed is set in advance so that only the phrases with the highest number of occurrences are displayed.

  By displaying such a matrix, it is possible to obtain the relationship between clusters and phrases, such as what phrases frequently appear in a cluster, and document points such as what is written about what topic. Can be grasped more accurately. For example, when the document to be processed is a response to a questionnaire regarding a camera, the number of phrases such as phrase A “dust-easy” and phrase B “dust-easy” appears in the document set belonging to cluster 1 “main body”. If there are many, it will be easily understood that there is a problem that the camera body is likely to be dusty.

(Attribute analysis using phrases)
In Table 12, “Phrase 1” and “Phrase 2” are phrases extracted from the document as row classification items, and “Attribute A” and “Attribute A” are stored as attribute values stored in association with the document. This is a matrix in the case of “attribute B”. The purpose of this matrix is to understand the relationship between phrases and attributes.

  Here, as a numerical value 1, for example, a ratio of the number nA_1 of “phrase 1” in the document set belonging to “attribute A” to the number n1 of “phrase 1” in the document set to be processed, that is, a value of nA_1 / n1. indicate. The same applies to the numerical values 2 to 4. By this numerical value, it is possible to grasp an attribute having a high appearance ratio for a certain phrase. Alternatively, as a numerical value 1, the ratio of the number of appearances nA_1 of “phrase 1” in the document set belonging to “attribute A” to the number of occurrences nA of phrases in the document set belonging to “attribute A”, that is, a value of nA_1 / nA is displayed. Also good. With this numerical value, a phrase having a high appearance ratio can be grasped for a certain attribute value.

  Further, when any of the classification items in the column classification item display area 53, that is, the value of the attribute is selected and the text is displayed, the text of the document set belonging to the attribute is displayed. On the other hand, when one of the classification items in the classification item display area 55 in the row, that is, the phrase is selected and the text is displayed, the text of the document set including the phrase is displayed. When a dot area in the graphic display area 57 is selected and the text is displayed, the text of the document set including the corresponding phrase and belonging to the corresponding attribute is displayed.

  Phrases, which are classification items of rows, may have a hierarchical structure, similar to (opinion trend analysis based on clusters and phrases). In this case, as described above, the hierarchy is switched so that the matrix can be displayed. When there are a large number of different phrases, the upper limit of the number of phrases to be displayed is set in advance so that only the phrases with the highest number of occurrences are displayed. Furthermore, the line classification item may be a co-occurrence frequency upper term pair instead of the phrase. The co-occurrence frequency upper term pair is obtained by arranging the word / phrase pairs appearing in one sentence in the order of frequency and acquiring the higher rank.

  By displaying such a matrix, it is possible to acquire the degree of appearance of a phrase that can grasp the point of the document, such as what is what and what is desired, in association with the attribute corresponding to the document. For example, when the document to be processed is a customer's opinion, for example, it is possible to grasp the trend of what you want to do for each gender of the customer, and to easily change what is in time series. I can grasp it.

(Affect degree time series analysis)
Table 13 shows that the classification items in the row are the noun phrases “evaluation phrase noun phrase 1” and “evaluation phrase noun phrase 2” included in the evaluation phrase, and the column classification items are stored in association with the document. , A matrix in the case of “time series unit A” and “time series unit B” which are time series units such as registration date. Here, the evaluation phrase is a phrase related to evaluation among phrases extracted from a document, and is a dependency such as a modified noun phrase “image quality” and an adjective phrase “bad”. The purpose of this matrix is to grasp the change in the degree of effect over time.

Here, as the numerical value 1, for example, the degree of effect of “noun phrase 1 of the evaluation phrase” in the document set belonging to “time series unit A” is displayed. The same applies to the numerical values 2 to 4. Here, the degree of effect is obtained by quantifying the degree of evaluation and is defined as follows.
Impression expression effect level = Degree adverb degree value x Impression expression word effect degree Modified noun phrase effect degree = Σ (Impression expression effect degree x Impression expression frequency including impression expression word) / (Affect degree obtained) Frequency of impression expression words)

  Here, Σ is the sum of the calculation results. Prior to the calculation of the effect level, the impression expression word and the effect level represented by the impression expression word are associated with each other and stored in the storage unit 12 in advance. A degree value indicating how much the degree adverb emphasizes the impression expression word is stored in the storage unit 12 in association with each degree adverb. Thereby, the degree of effect of impression expression is obtained using the above definition. The degree of effect of the modified noun phrase is, as defined, for impression expressions having the same noun phrase, adding the degree of effect of each impression expression considering the frequency, and dividing by the number of impression expressions that obtained the degree of effect. Can be obtained. This numerical value is the degree of effect per impression expression of the noun phrase, and expresses the affectivity for the object represented by the modified noun phrase.

  For example, consider a case where “cute” appears twice, “strong” appears twice, and “unbalance” appears once as an adjective phrase related to the modified noun phrase “body” in a document set. When the degree of effect of these adjective phrases is “3”, “3”, and “−1”, the degree of effect of “body” is (3 × 2 + 3 × 2 + (− 3 × 1)) / 5 = 1. 8 can be calculated.

  In addition to the matrix in Table 13, rows and columns representing the average degree of effect may be added and displayed. For example, a column is added to the right side of the matrix, and the average degree of effect of “evaluation phrase noun phrase 1” is displayed with dots in the entire period, or a row is added to the lower side of the matrix. You may display the average effect degree of all noun phrases, etc. with a dot.

  Since the degree of effect can be a positive or negative real value, the dot color may represent positive or negative, and the dot size may represent an absolute value. For example, when the degree of effect is positive, it is displayed in three colors such as blue, when it is negative, red when it is 0, and the size is determined by the absolute value. By adopting such a configuration, the transition of the degree of effect can be followed for each noun phrase in one screen, and comparison and analysis are facilitated.

  Further, when any of the classification items in the column classification item display area 53, that is, the time series unit is selected and the text is displayed, the document set including the noun phrase of the evaluation phrase among the document sets belonging to the time series unit is displayed. Make sure the text is displayed. On the other hand, when one of the classification items in the classification item display area 55 in the row, that is, the noun phrase of the evaluation phrase is selected and the text is displayed, the text of the document set including the noun phrase of the evaluation phrase is displayed. To do. When a text area is displayed by selecting a dot area in the graphic display area 57, the text of the document set belonging to the corresponding time-series unit among the document sets including the noun phrase of the corresponding evaluation phrase is displayed. Like that. At this time, the evaluation phrase included in the text is extracted and displayed as an attribute.

  FIG. 10 shows a configuration example of the screen in the display unit 18 when the text of the document is displayed in the matrix configuration of Table 13. The text display screen 108 includes a document data display area 110. The document data display area 110 includes a modified noun phrase display area 112, an adjective phrase display area 114, a text display area 116, and an attribute display area 118. This figure is a screen when, for example, a customer's comment on a camera as a product is a processing target, and a text set is displayed by selecting a document set including “shooting” as a noun phrase of an evaluation phrase. Therefore, “captured” is described in all sentences in the modified noun phrase display area 112.

  In addition, the adjective phrase of the evaluation phrase included in each document is extracted from the text and described in the adjective phrase display area 114. Thereby, it is possible to confirm at a glance the evaluation points of each document described in the text display area 116. The attribute display area 118 describes attributes such as the gender and age of the person who created the document. In this way, it is possible to check the specific contents by referring to the main text on a single screen while grasping the trend of what kind of customers are doing what kind of evaluations. Can be done efficiently.

  11 and 12 show another example of the screen configuration in the display unit 18 when the text of the document is displayed in the matrix configuration of Table 13 as well. In this example, the text is displayed by a two-stage process. First, in the document set including the noun phrase selected as the first stage, the number of appearances of the evaluation phrase including the noun phrase and the degree of effect are displayed in a list. FIG. 11 shows a screen configuration example at that time. The evaluation phrase count display screen 128 includes a modified noun phrase display area 120 and an adjective phrase display area 122. In the modified noun phrase display area 120, the noun phrase of the evaluation phrase and the number of appearances thereof corresponding to the dot selected in the matrix display of Table 13 are displayed. In the figure, the noun phrase “shooting” and the number of appearances “37” are displayed.

  The adjective phrase display area 122 displays an adjective phrase that is dependent on the noun phrase corresponding to the selected dot, the number of appearances, and the degree of effect. In the figure, “unsuitable”, “good”, and “to stable” are displayed as adjective phrases, and the number of occurrences of each is “7”, “3”, “2”, and the degree of effect is “unpopular” or “popular” "," Unpopular "is displayed. Here, the indication of the degree of effect expresses the degree of effect of each adjective phrase by words. That is, if the effect degree is a positive value, it is popular, and if it is negative, it is not popular. Further, the degree is represented by three levels of high, medium, and low. For example, an adjective phrase with an effect degree of -3 is displayed as “not popular”, and an adjective phrase with an effect degree of 4.5 is displayed as “highly popular”. These correspondences are set in advance and stored in the storage unit 12.

  When the adjective phrase selection check box 124 is checked on the evaluation phrase count display screen 128, as a second stage of the text display, a document set including any of the adjective phrases, that is, a document set corresponding to the dot selected in the matrix of Table 13 Displays the text of. FIG. 12 shows a screen configuration example at that time. When the check box at the head of each adjective phrase is checked on the evaluation phrase count display screen 128, only the text of the document set including each adjective phrase is displayed.

  The text display screen 130 in FIG. 12 includes a full text display instruction area 132, an adjective phrase display area 136, a text display area 138, a cluster name display area 140, and an attribute display area 142. In the adjective phrase display area 136, the adjective phrases displayed on the evaluation phrase count display screen 128 are displayed, and in the text display area 138, a sentence having a predetermined length including the corresponding part is displayed among documents including the adjective phrases. In the figure, for example, the texts of seven documents including the adjective phrase “unsuitable” are displayed in order from the top, and the texts of three documents including the next adjective phrase “good” are displayed next. . In the text display, the noun phrase and adjective phrase of interest are highlighted by surrounding them with a frame or coloring them.

  The cluster name display area 140 displays the cluster name of the cluster to which each document belongs, for example, “large”, “room”, “moving image”, and the like. Thereby, it is possible to easily grasp what the displayed document is mainly about. Similar to the screen example of FIG. 10, the attribute display area 142 describes the gender and age of the person who created the document. When the full text display instruction area 132 displayed at the beginning of each line is clicked, the full text of the document on the selected line is further displayed.

  By making the text display in two stages, it is possible to display the text after confirming the degree of influence and the number of appearances, and it is possible to efficiently narrow down even when there are a large number of document sets. In addition, various data such as the degree of effect, the number of appearances, the cluster name, and the text can be easily associated and understood.

(Affect degree attribute analysis)
Table 14 shows the name of the attribute stored in association with the document with the column classification item being “impression expression phrase 1” and “impression expression phrase 2”, which are impression expression phrases included in the document set. This is a matrix in the case of “attribute A” and “attribute B”. The purpose of this matrix is to grasp the relationship between the impression expression used and the attribute using the degree of effect.

  Here, as the numerical value 1, for example, the degree of effect considering the frequency of “impression expression word / phrase 1” in the document set belonging to “attribute A” is displayed. The same applies to the numerical values 2 to 4. For example, it is assumed that “impression expression phrase 1” is “cute”, “impression expression phrase 2” is “unbalance”, “attribute A” is “male”, and “attribute B” is “female”. Assume that the “cute” unique effect degree is 3, and the frequencies of “cute” in documents created by “male” and “female” are 2 and 10, respectively. Further, it is assumed that the unique “unbalance” effect degree is −3, and the frequency of “unbalance” in documents created by “male” and “female” is 12 and 4, respectively. At this time, the degree of affect of each word and phrase considering the frequency is obtained by multiplying the frequency of the unique degree of effect, so the numerical value 1 is 6, the numerical value 2 is 30, the numerical value 3 is -36, and the numerical value 4 is -12.

  FIG. 13 shows a matrix when this data is represented by dots. In the matrix 50 of the same figure, the column classification item column 52 displays attribute names “male” and “female”, and the row classification item column 54 displays impression expression phrases “cute” and “unbalance”. Has been. In the graphic display column 56, the degree of effect of each impression expression word / phrase is displayed for each attribute in the size and color of the dot. Here, for the convenience of display, white and black indicate the positive and negative effect degrees.

  By displaying such a matrix, for example, “female” has a positive degree of effect, that is, the word “cute” is often used when evaluating a favorable evaluation. It is possible to grasp the relationship between the impression expression and the attribute as a tendency, such as “man” is often used.

(FAQ preparation support)
Table 15 shows “cluster 1 (query)” and “cluster 2 (query)” which are cluster names when query statements from customers are clustered as row classification items, and column classification items are response sentences to the query text. This is a matrix in the case of “cluster A (answer)” and “cluster B (answer)” which are cluster names when clustering. The inquiry sentence and the answer sentence are sentences written in an e-mail or a postcard, or a voice documented on a telephone. The number of clusters displayed in the rows and columns may be the same. The purpose of this matrix is to grasp the variation of answers to queries and the variation of queries to answers.

  Here, as the numerical value 1, for example, the ratio of the number of documents nA_1 of the document set belonging to “Cluster A (Answer)” out of the number of documents n1 of the document set belonging to “Cluster 1 (inquiry)”, that is, a value of nA_1 / n1. indicate. The same applies to the numerical values 2 to 4. This value represents the variation in the content of the response to the content of the inquiry. On the other hand, as a numerical value 1, for example, the ratio of the number of documents nA_1 of the document set belonging to “Cluster 1 (inquiry)” out of the number of documents nA belonging to “Cluster A (answer)”, that is, a value of nA_1 / nA is displayed. May be. This value represents the variation in the content of the inquiry with respect to the content of the answer. Note that these numerical values may be switched and displayed when the user clicks on the matrix display area 51.

  For example, when there is a large variation in the content of answers to a certain inquiry, it is necessary to clarify the answer criteria for such an inquiry. In this way, by displaying the variation of the answer contents with respect to the contents of the inquiry in a matrix, the improvement point on the answering side can be grasped. In addition, by displaying the variation in the content of the query with respect to the content of the response in a matrix, it is possible to grasp the variation in the terms used by the inquirer and the respondent, and to understand the similarity of the responses in order to create a FAQ from the common answers. Can be.

(FAQ search time series analysis)
Table 16 uses “search sentence 1” and “search sentence 2” which are search sentences input by the questioner in the FAQ search as row classification items, and “time series unit” which is a time series unit in which column classification items are input. A matrix when “A” and “time-series unit B” are used. Here, the search sentence is a search character string such as “zip code” or a sentence “I want to know about a 7-digit zip code”, which is input in the FAQ search system in order to search for a target document. The search sentences to be displayed in Table 16 have the highest frequency, and the number to be displayed is set in advance. The purpose of this matrix is to grasp changes with time of a search sentence inputted in the FAQ search.

  Here, as a numerical value 1, for example, a ratio of the number nA_1 of “search sentence 1” input in “time series unit A” to the number n1 of “search sentence 1” in all periods, that is, a value of nA_1 / n1. indicate. The same applies to the numerical values 2 to 4. By this numerical value, it is possible to specify the time when the input frequency of a certain search sentence is high. On the other hand, as a numerical value 1, for example, a ratio of the input number nA_1 of “search sentence 1” to the input number nA of search sentences input in “time series unit A”, that is, a value of nA_1 / nA may be displayed. With this numerical value, a search sentence having a large number of input items in a certain period can be specified. Note that these numerical values may be switched and displayed when the user clicks on the matrix display area 51.

  The classification item of the line may be the search sentence itself, or may be a classification item for each phrase extracted from the search sentence. By displaying such a matrix, it is possible to grasp, for example, search sentences with increasing inquiries, and to quickly take measures for improvement such as expanding the content corresponding to the inquiries and extracting product problems. Can stand up.

  A similar matrix may be displayed with the line category items as words extracted from the question sentence. Here, the question sentence is a question included in the FAQ, that is, a “frequently asked question”, for example, a sentence such as “an error has occurred in printing”. Phrase extraction processing is performed on such sentences, and classification is performed based on the extracted phrases. In the above sentence, for example, the words “print”, “error”, and “occurrence” are extracted. The phrase extraction process may be performed by a device external to the information display apparatus 10. In this case, the extracted phrase is associated with the document and the attribute and stored in the storage unit 12. Even when the line classification item is a word extracted from the question sentence, the same effect as the search sentence can be obtained.

(Category analysis of FAQ search sentences)
In Table 17, “Search text 1” and “Search text 2”, which are search sentences input by the questioner in the FAQ search, are set as the row classification items, and “Category A” is the category name corresponding to the search text as the column classification items. ”And“ Category B ”. The purpose of this matrix is to grasp the search text input in the FAQ search as a ratio for each category. The category is information about items assigned to the created FAQ. For example, if the FAQ is for a printer, the category is “paper setting / printing” or “postcard”. By specifying a category when inputting a search sentence, the search sentence and the category can be stored in association with each other.

  Here, as the numerical value 1, for example, the ratio of the input number nA_1 of “search sentence 1” belonging to “category A” out of all the input numbers n1 of “search sentence 1”, that is, a value of nA_1 / n1 is displayed. The same applies to the numerical values 2 to 4. The classification item of the line may be the search sentence itself, or may be a classification item for each phrase extracted from the search sentence. By displaying such a matrix, it is possible to specify categories that tend to be classified even in the same search sentence, or to specify search sentences that have similar distributions for a plurality of categories. For example, it is possible to grasp a tendency such as a search term “driver” used in a plurality of categories in a search related to a computer in which category of “model” or “OS” is frequently searched. .

  As with (Time Series Analysis of Classification Results), each category is assigned to internal departments in advance, and when the user clicks on the classification item display area 55 in the row, a matrix that is re-aggregated by department to which each category belongs is displayed. You may make it do.

(Understanding the category distribution of FAQ questions)
Table 18 is a category name corresponding to a question sentence with column classification items as “question word phrase 1” and “question word phrase 2”, which are words extracted from FAQ question sentences. This is a matrix in the case of “Category A” and “Category B”. The purpose of this matrix is to grasp in what category the words included in the question sentence are distributed.

  Here, for example, a ratio of the number nA_1 of question sentences belonging to “category A” out of the number n1 of question sentences including “phrase 1 of question sentence”, that is, a value of nA_1 / n1 is displayed. The same applies to the numerical values 2 to 4. By displaying such a matrix, it is possible to grasp the distribution of the phrase included in the question sentence with respect to the category. For example, a phrase of a question sentence that belongs only to a certain category is highly likely to be a phrase that characterizes the category. Therefore, when the FAQ is created, it is useful information for determining the category to be given to the question sentence including the word / phrase.

  Also, if there are words that are distributed in a few categories and the ratio of one of them is extremely high, the question sentence that belongs to the category with a low distribution ratio may have an incorrect category. is there. In addition, there is a possibility that the question sentence uses the phrase even though it is not directly related to the contents of the category. From this point of view, this matrix display is a material for determining whether or not the category assigned in the prepared FAQ is accurate.

  Furthermore, if there are words that are distributed relatively evenly in a plurality of categories and are also category names, there is a possibility that there is a problem with the category structure in the created FAQ. For example, the question text “Error during printing” belongs to the “Print” category, the question text “Error during upgrade” belongs to the “Upgrade” category, and the question text “An unexpected error has occurred.” When “is” belongs to the “error” category, the category “error” exists even though the word “error” is used in a plurality of categories. In such a case, there is a high possibility that there is a problem with the category configuration. By displaying the matrix, the problems as described above can be easily identified.

(Time series analysis of the number of FAQ categories)
In Table 19, “Category 1” and “Category 2” are category names corresponding to the question texts of the FAQ, and the column classification items are “time series units” that are time series units for creating the question texts. This is a matrix in the case of “A”, “time series unit B”, and “time series unit C”. The purpose of this matrix is to grasp the change over time in the number of question sentences for each category.

  Here, as the numerical value 1, for example, the number nA_1 of question sentences created in “time series unit A” among the question sentences belonging to “category 1” is displayed. The same applies to the numerical values 2 to 6. Alternatively, numerical values 2 and 5 relating to the question sentence created in time series unit B and numerical values 3 and 6 relating to the question sentence created in time series unit C are the number of question sentences created in time series unit A. The amount of change from the numerical values 1 and 4 or the rate of change may be used. For example, assuming that the number of question sentences created in “time series unit A” among the question sentences belonging to “category 1” is nA_1 and the number of question sentences created in “time series unit B” is nB_1, the numerical value 2 is The amount of change is nB_1-nA_1 or the rate of change nB_1 / nA_1. As in the matrix shown in FIG. 3, the rate of change and the absolute value may be displayed simultaneously depending on the color and size of the dots.

  By displaying such a matrix, it is possible to easily grasp changes such as a sudden increase in the number of FAQs belonging to a certain category, and to find out the category in which a problem occurs or to adjust the balance between categories. Recognize the necessity and immediately recognize the occurrence of a problem. In order to obtain such an effect more reliably, a threshold value may be set in advance for the amount of change and the rate of change, and a function of notifying the user when the threshold value is exceeded may be provided. For the notification, a warning to that effect may be displayed on the screen, or an e-mail may be automatically transmitted to an address set by the user for the category in question.

(Expertise analysis)
In Table 20, the classification item of the row is “technical term 1” and “technical term 2” which are technical terms included in the document, and the column classification item is “person in charge A” who is the person in charge of creating the document. This is a matrix in the case of “person in charge B”. Here, information relating to technical terms is stored in advance in the storage unit 12 from an external dictionary or the like. The purpose of this matrix is to grasp for each creator the technical terms used in answers to inquiries and business daily reports.

  Here, as a numerical value 1, for example, a ratio of the number of times nA_1 that “person A” used “technical term 1” to the number n1 that all persons in charge used “technical term 1”, that is, a value of nA_1 / n1 is displayed. To do. The same applies to the numerical values 2 to 4. By displaying such a matrix, it is possible to grasp the level of expertise, field of expertise, and field of weakness of each person in charge, and it is possible to efficiently perform person-in-charge education and person-in-charge change.

  According to the above-described embodiment, the document set is classified by various methods specified by the user, and the information related to the classification result is visualized by displaying with dots on the matrix. Thereby, even when an advanced classification method is used, the correlation between the classification results can be intuitively grasped on one screen. In addition, since the classification items that make up the rows and columns of the matrix are set by simple operations such as drag and drop from the list display, it is possible to obtain knowledge by comparing the trends by displaying the matrix with different classification items one after another. it can. One matrix is two-dimensional information. However, if a plurality of matrices can be compared, multidimensional analysis is realized, and the amount of information obtained is greatly increased.

  Furthermore, even if an enormous database is to be processed, classification items suitable for the document can be selected, so that efficient classification processing is possible, and it is possible to efficiently narrow down until a desired document is finally obtained. In addition, since any document can be processed regardless of the presence / absence and format of the document, it can be widely used for searching documents, analyzing products, etc., and checking the validity of the classification process itself. Adjustments can also be made. Various effects obtained in each application are as described above.

  The present invention has been described based on the embodiments. This embodiment is an exemplification, and it will be understood by those skilled in the art that various modifications can be made to combinations of the respective constituent elements and processing processes, and such modifications are also within the scope of the present invention. is there.

It is a figure which shows the whole structure of the search system in this Embodiment. It is a figure which shows the example of the matrix displayed on a display part in this Embodiment. It is a figure which shows another example of the matrix displayed on a display part in this Embodiment. It is a flowchart which shows the process sequence of the matrix display by the information display apparatus in this Embodiment. It is a figure which shows the structural example of the screen displayed on a display part in this Embodiment. It is a figure which shows the example of the matrix displayed on a matrix display area, when the classification item of a line has hierarchical structure in this Embodiment. It is a figure which shows typically a mode that diagonalization sort is performed in this Embodiment. It is a figure which shows the structural example of the screen at the time of displaying the information which concerns on a document set in this Embodiment. It is a figure which shows the structural example of the screen at the time of displaying the text of a document in this Embodiment. It is a figure which shows the structural example of the screen at the time of displaying the text of a document in the effect degree time series analysis of this Embodiment. It is a figure which shows the structural example of the screen at the time of displaying the number of appearance of the evaluation phrase containing the noun phrase selected in the effect degree time series analysis of this Embodiment, and the degree of effect. It is a figure which shows the structural example of the screen at the time of displaying the text of a document in the effect degree time series analysis of this Embodiment. It is a figure which shows the example of the matrix displayed in the effect degree attribute analysis of this Embodiment.

Explanation of symbols

  10 information display devices, 12 storage units, 14 classification processing units, 16 matrix generation units, 18 display units, 20 input units, 22 matrix display units, 50 matrices, 51 matrix display areas, 52 columns of classification item fields, 54 rows Classification item field, 56 graphic display field, 60 matrix display screen, 62 classification item selection area, 66 document set designation area, 67 sort instruction button, 68 narrowing instruction button, 80 horizontal direction guide line, 82 vertical direction guide line, 84 vertical direction Direction information display area, 86 Horizontal direction information display area.

Claims (17)

A storage unit for storing a plurality of documents;
A classification processing unit that forms a plurality of document set groups by classifying a plurality of documents stored in the storage unit by a first classification method and a second classification method;
The correlation between the classification results obtained by the first classification method and the second classification method performed by the classification processing unit is related to the intersection of the document sets by expanding the two series of document sets into rows and columns. A matrix display unit for displaying numerical information as a two-dimensional matrix expressed in a predetermined figure;
An information display device comprising:   The display unit according to claim 1, further comprising a display unit that displays a list of a plurality of classification methods and receives an input from a user who selects the first classification method and the second classification method from the list. Information display device.   The said matrix display part represents either the number of the documents contained in the said product set, and the ratio of the number of the said documents by the change of the color of the said figure, The Claim 1 or 2 characterized by the above-mentioned. Information display device.   The matrix display unit is a numerical value related to the amount of change in the number of documents included in the product set belonging to each row, based on the number of documents included in each product set belonging to a predetermined column of the product set. The information display device according to claim 1, wherein the information is expressed by a change in color of the graphic. The classification processing unit classifies a plurality of documents stored in the storage unit according to a predetermined clustering method based on similarity of words included in the document,
At least one of the first classification method and the second classification method is a clustering method performed by the classification processing unit,
The matrix display unit displays a representative phrase selected according to a predetermined criterion from phrases extracted from each document when the classification processing unit performs clustering as a classification item on a heading of the two-dimensional matrix. 3. The information display device according to 1 or 2. Both the first classification method and the second classification method are the same clustering method performed by the classification processing unit,
The matrix display unit expresses a ratio of a document set belonging to another classification item including at least one representative word / phrase selected from a word / phrase extracted from a document set belonging to a certain classification item by the graphic. The information display device according to claim 5. The storage unit further stores preset fixed classification items,
The classification processing unit classifies the plurality of documents stored in the storage unit into the fixed classification items,
Both the first classification method and the second classification method are classifications into the fixed classification items performed by the classification processing unit,
The said matrix display part represents the similarity for every combination of the said fixed classification item calculated with the word extracted from the document set which belongs to each fixed classification item with the said figure, or characterized by the above-mentioned. 2. The information display device according to 2. Either of the first classification method and the second classification method is a classification method in which classification items form a hierarchical structure,
The classification processing unit performs classification corresponding to each hierarchy of the classification items,
The information display device according to claim 1, wherein the matrix display unit switches and displays the two-dimensional matrix that represents a classification result corresponding to the classification item hierarchy according to a selection instruction of a user.   The matrix display unit determines a display order of each classification item in the two-dimensional matrix so that a numerical value related to the product set becomes maximum on a diagonal line of the two-dimensional matrix. The information display device described in 1.   The matrix display unit converts at least one of numerical information represented by a graphic displayed in an area on the two-dimensional matrix selected by a user and information on classification items corresponding to rows and columns of the area into a text format. The information display device according to claim 1, further displaying.   The said matrix display part further displays the document contained in the said product set corresponding to the figure displayed on the area | region on the said two-dimensional matrix selected by the user in a text format. The information display device described in 1. A storage unit for storing a plurality of documents;
A classification processing unit that forms a plurality of document sets by classifying the plurality of documents stored in the storage unit by a predetermined classification method;
Based on the number of occurrences of words / phrases extracted from each document set classified by the classification processing unit or the number of occurrences of combinations of words / phrases, the numerical information related to the words / phrases is calculated, and the result of classification performed by the classification processing unit and A matrix display for displaying a correlation with numerical information related to a phrase as a two-dimensional matrix expressing the numerical information related to the phrase in a predetermined figure;
An information display device comprising: The storage unit further stores data associating a predetermined impression expression word with a numerical value indicating a degree of evaluation represented by each impression expression word,
The matrix display unit calculates, for each modified noun phrase included in the phrase including the impression expression word extracted from each document set, a degree of evaluation of the modified noun phrase based on the data, The information display device according to claim 12, wherein the information display device is represented by a figure. Receiving a selection input of a first classification method and a second classification method for classifying a plurality of documents from a user;
Classifying the plurality of documents with the selected first classification method and second classification method to form a two-line document set group;
The correlation between the classification results obtained by the first classification method and the second classification method is expressed as follows: the two series of document set groups are expanded into rows and columns, and numerical information relating to a product set of the document set groups is expressed in a predetermined figure. Displaying as a represented two-dimensional matrix;
An information display method comprising:   The information display method according to claim 14, further comprising a step of receiving a selection input of the type of numerical information from a user.   When a selection input of any one of the first classification method, the second classification method, and the numerical information type is newly received after the displaying step, the two-dimensional matrix corresponding to the selection input is displayed. The information display method according to claim 15, further comprising a step of switching between. A function of accepting selection input of a first classification method and a second classification method for classifying a plurality of documents from a user;
A function of classifying the plurality of documents by the selected first classification method and second classification method to form a two-line document set group;
The correlation between the classification results obtained by the first classification method and the second classification method is expressed as follows: the two series of document set groups are expanded into rows and columns, and numerical information relating to a product set of the document set groups is expressed in a predetermined figure. A function to display as a expressed two-dimensional matrix;
A computer program for causing a computer to realize the above.

Images