JP2003288352A - Information analytic display device and information analytic display program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem in a conventional device that the display of a set of data by classifying and arranging it on the space in a form suitable for the set of data requires a large computational effort, and the result is hard to see. <P>SOLUTION: In a set of data expressed by matrix, the lesser data of row and column are arranged on space by a self-organizing method, and the other data is arranged by use of only this arrangement result, whereby the set of data can be classified and arranged with a less computation effort. Further, since one data can be used as a label by displaying both the data, a result easy to see can be obtained. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to data analysis and analysis result display using an electronic computer. In particular, the present invention can be applied to display document search results classified by keywords and to analyze and display the relationship between customers and products in market analysis.

[0002]

2. Description of the Related Art A word-document relation matrix is often used for document classification and analysis. This is a matrix that takes words in rows and documents in columns and records how many times the word appears in the document. See FIG. You can retrieve the vector representation of a word by extracting each row of this matrix, or the vector representation of a document by extracting each column. Therefore, the distance between certain two words A and B can be defined by the distance between the vectors, the cosine of the vector, or the inner product of the vectors. Similarly, the distance between documents can be defined by vector distance, cosine, inner product, and the like. That is, the distance between words can be expressed by comparing vectors having documents as elements, and the distance between documents can be expressed by comparing vectors having words as elements. The distance between the document and the word can also be defined by how many times the word appears in the document.

Now, as an example of data that can be expressed in a matrix like this, there is a relationship between a customer and a purchased product in marketing. A matrix with rows of products and columns of customers,
The relationship between a customer and a product can be recorded by recording which product and how much a certain customer has purchased. See FIG. Also in this case, the vector expression of each customer or the vector expression of each product can be extracted.
Since the vector of a certain customer indicates the product preference of the customer, it can be said that the customers whose vector distances are close to each other have similar preferences. Also in this case, the customer distance can be expressed by a vector having the product as an element, and the product distance can be expressed by a vector having the customer as an element.

In this example, documents and words are in a row and column relationship in a matrix, and customers and products are in a similar relationship. Although there are many combinations of data having such a relationship, the relationship between a document and a word will be described below as an example.

With the spread of IT technology and the Internet,
The number of computerized documents continues to grow explosively. For example, the number of newspaper articles and patent documents has reached an enormous amount even with existing publications, and it is certain that the number will continue to increase in the future. In order to effectively use these documents, it is indispensable to have a search / classification / analysis means capable of accurately selecting a target document.

As a means for classifying documents as search results, the following methods already exist when roughly classified.

(1) The first is to create a classification standard in advance and classify documents according to this classification standard. FIG. 17 is a flowchart showing the outline of the operation and processing of this method. As a preparatory step, first, a classification standard is manually created (1701). Once a classification standard is created, it can be used universally and repeatedly. Next, the document is searched (1702), and the document set of the search result is automatically classified according to this classification criterion (170
3) The results are displayed for each classification (1704). This method is suitable for newspaper articles whose classification can be decided in advance.

(2) The second is to arrange a set of documents in space only by the distance between the documents. By repeating the calculation until the arrangement converges, self-organizing classification becomes possible. In order to realize this method, SOM (self-organizing map)
(Reference book: T. Kohonen: "Self-Organizing Map" Springer Fairlark Tokyo ISBN4-431-
70700-X (1996)) or arrangement by spring model (reference: Peter Eade)
s: "A heuristic for graph
drawing ", congressus number
antim, Vol 42 (1984)) (References of examples applied to document analysis: Isamu Watanabe: "Visual Text Mining", Journal of Artificial Intelligence Vol. 16 No. 2 (200)
1)) etc. are famous.

Of these, the spring model is an arrangement method intended for undirected graphs (graphs having no direction), but it can be applied to classification arrangement of documents and words. For example, when arranging a document, the document is regarded as a graph contact point, and this contact point is considered to be connected by a spring according to the document distance (or similarity). FIG. 28 shows an example of the initial state.

In FIG. 28, each contact point is a document, and jagged lines schematically show the spring. Then, the stable state of the system composed of this contact and the spring, that is, the state in which each spring is settled to a position close to the original length where it does not extend or contract too much is obtained. As a result, similar documents are placed closer and dissimilar documents are placed far away. Such an example is shown in FIG.

In the example of FIG. 29, it is necessary to visually understand that documents A, B, and C are similar to each other, and document D is a document that is not similar to any of documents A, B, and C. You can

[0012] The SOM and spring model methods can be arranged in a manner suitable for the document set of the search result, so that flexible classification can be performed for each search. Since this method uses self-organizing classification, the classification result does not always follow the standard that can be seen by people. Therefore, the resulting clusters are usually labeled.

This is shown in the flow chart of FIG. That is, the document is first searched (180
1) self-organize and arrange retrieved documents (180
2) and divide the cluster into clusters based on the placement result (18
03), label each cluster (180
4) Finally, the placement result and the label are displayed (180
5). Japanese Patent Laid-Open No. 8-263514 is an example of a method in which SOM is applied as the self-organizing method, and the result of using SOM is often displayed as a set of cells as shown in FIG. The result using the spring model is shown in FIG.
It is often displayed as the arrangement of data on the space like.

(3) The third method is to classify documents according to how close they are to the keyword. FIG. 19 is a flowchart showing the outline of the operation and processing of this system. First, a document is searched (1901), and a person gives a keyword to the document of the search result or a keyword is automatically extracted (1902), and the keyword is arranged at a fixed point in space (1903). Place each document in the search results in the same space depending on which keyword is closest (1
904), and finally the placement result is displayed (1905). Japanese Patent Laid-Open No. 2000-76279 is an example of this method. (4) Further, the vectorized documents are clustered into an appropriate number, and the mapping means is applied only to the cluster centers that represent the clusters, so that the documents have a certain dimensional space according to the perspective of semantic content. The method of arranging the above is disclosed in Japanese Patent Application Laid-Open No. 10-171823. In this method, first, the document to be analyzed is vectorized by the vectorization unit 3503, and the vectorized document is classified into clusters by the clustering unit 3504. Next, the cluster center extraction means 35
The vector representing each cluster is extracted by 05,
The cluster center mapping unit 3506 arranges the cluster centers in a low-dimensional space while keeping the distance between them as much as possible. The documents included in each cluster are arranged based on the arrangement position thus determined and the vector classification result obtained by the clustering unit 3504. At the time of this document arrangement, a comparison is made with the cluster center of the cluster arranged in the vicinity of the cluster to which the arranged document vector belongs.

[0015]

However, the classification means of the above (1) can only classify according to a predetermined classification standard. This may be good for categorizing newspaper articles as economics and sports, but when you search for documents, you come across situations in which search results must always be categorized on a new axis. Even if sports are classified as professionals and amateurs, another classification axis will be needed for the Olympics where professionals can participate. Since the classification changes depending on the situation, there is a limit to the method of creating a predetermined classification standard.

In order to perform the self-organizing classification, the classifying means (2) has to repeatedly perform calculation between all documents until they settle at appropriate positions (until they converge). When the number of documents to be classified becomes enormous, it is very costly to calculate them until their positions converge, which is not very practical.

FIG. 28 shows a spring model, for example.
Was a model of four contacts. FIG. 30 is a schematic diagram in which a spring is drawn in a straight line for a model of eight contact points. As can be seen in this figure, the number of springs has been quadrupled by simply doubling the number of contacts from four to eight. When all the N documents are connected by springs, the number of springs is N × (N−
1) ÷ 2, so the number of springs is on the order of N squared.

Even if a document can be arranged in the space as shown in FIG. 20 by using the spring model, what kind of cluster it will be a delicate problem. Then, even if this can be divided into clusters as shown in FIG. 21, a label (character string) meaning the cluster for each cluster
Is not always attached properly. This is because this cluster is for calculation of multidimensional vectors, and therefore there is no guarantee that it is a classification that is easy for humans to understand. Even if an attempt is made to extract and display labels after classification from the title of a document as in the invention disclosed in Japanese Patent Laid-Open No. 2000-82068, the labels of clustered documents are different from each other, or they are displayed on different clusters. If there are many documents with the same title, an appropriate label cannot always be extracted. Therefore, after classification and placement, the labeling problem alone cannot be solved without costly calculations. The same applies to SOM.

The classification means (3) is based on the premise that the keywords are fixedly displayed and the distances between them are set at equal intervals. When a person attaches this keyword, it is not possible to set an appropriate number of appropriate classification words.
That is, for example, the six keywords selected there are not necessarily opposite words suitable for classification with respect to the document set. For example, when trying to classify sports-related newspaper article documents, "baseball,""ballgames,""high school baseball,""J
There are cases in which the level of concept or abstraction is not uniform, such as “league”, or a category that is not appropriate for the document set is specified. When a keyword is extracted by a computer, even if it is possible to extract an appropriate keyword, by arranging them for a given set of documents and at equal intervals, the set of documents is different from the original feature. There is a risk of being classified into. That is, JP 2000
Assuming that six keywords are arranged in a hexagon as in the invention disclosed in -76279, only one of the six keywords originally has a unique meaning for the document set. If so, the classification arrangement is not appropriate. Further, in the above method (4), when each document is arranged, the comparison is made with the cluster center of the cluster arranged in the vicinity of the cluster to which the document belongs. not going. Therefore, even if a document vector classified into a certain cluster actually has characteristics similar to those of the cluster centers other than the neighborhood, the influence from the cluster centers other than the neighborhood is ignored, and It is difficult to obtain mapping results that reflect the characteristics of various documents. Furthermore, the document is not labeled when it is placed, and the displayed placement result may be difficult for the user to visually identify. In order to realize the labeled display, it is necessary to perform a costly calculation such as a label determination based on the result of the data arrangement and a correspondence between the label and the data.

The present invention is to solve the above-mentioned problems of the prior art, and is to classify the document set itself by using only the information of the document set, but it is given at high speed. This is to perform labeling suitable for the set of documents and according to the actual conditions so that the user can easily use it. Moreover, it aims at providing the apparatus which implements the method.

[0021]

In order to achieve this object, according to the present invention, when the relation between two data sets can be expressed by the relation between the rows and columns of a matrix, the data contained in one of the data sets As an object A, data included in the other set is a data object B, and its operation is shown in FIG. First, using only the data object A, the data object A is arranged in a space (for example, a space of three dimensions or less that can be visually grasped by humans) from the relationship between the distances while maintaining the distance relationship (101). Using only the distance relationship between data object A and data object B, the distance relationship between data object B and data object A is maintained in space without using the distance relationship between data objects B. While arranging (102), the data objects A and B are finally displayed (103).

That is, of the two types of data objects, a small number of data objects are designated as data object A, and this data object A is arranged by, for example, self-organizing classification, and a large amount of data is generated based on this arrangement. The data object B is arranged using the arrangement of the data object A.

Therefore, in the information analysis display device of the present invention,
With the data object A arrangement, the data object B, which is a large amount of data, can be arranged in space at high speed. Further, by using the layout of the data object A as a label indicating the data attribute of the data distribution, it is possible to narrow the search results of documents and analyze the relationship between data. As a result, the labeling and the data mapping can be realized at the same time, and since the mapping result is labeled, the data analysis display can be performed in a format that is visually easy for the user to understand.

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION (First Embodiment) A first embodiment of the present invention will be described below. The first embodiment is a document classification device that searches documents and classifies and arranges the results.

FIG. 2 is a block diagram of the apparatus of the first embodiment.

As shown in FIG. 2, this apparatus has a means 201 for inputting search conditions, an output means 202 for outputting a search result and a classification arrangement result, and a search means 203 for performing a search.
Search result storage means 204 for storing search results, document data storage means 205 for storing document data, word data storage means 206 for storing word information, and matrix storage means 207 for storing word and document matrix data. A distance calculation means 208 for calculating distances between words, between documents, and between word documents using a matrix, keyword extraction means 209 for extracting keywords from a document set, and those data from distances between data. Is provided in the space, and the space storage means 211 for storing the space information. The analysis arrangement means 210 is composed of word data mapping means 210a and document data mapping means 210b. The word data mapping unit 210a is described as a label mapping unit in the claims of the present application, and the document data mapping unit 210b is described as a data mapping unit in the claims of the present application.

In the matrix storage means 207, words and documents are recorded in a matrix of rows and columns as shown in FIG.
The element (i, k) of the matrix is the number of times the word i appears in the document k.

FIG. 4 is a flow chart showing the outline of the use and operation of this device. (Step 401) First, the user inputs search conditions from the input means 201. Then (step 402), the search means 203 performs a search and stores the document set of the search result in the search result storage means 204. (Step 403) The keyword extraction means 209 extracts a keyword from this search result. The method of Japanese Patent Laid-Open No. 11-25108, which is a known technique, is used to extract the keyword (or characteristic word). (Step 404) The distance calculation means 208 calculates the distance between keywords for the extracted keywords based on the information of the matrix storage means 207, and the word data mapping means 210a makes the two-dimensional space or three-dimensional space of the space storage means 211. Place in space. The spring model, which is an existing method, is used to place the keywords in the space. Details will be described later. Keywords placed in the space are expressed by the notation of the word. FIG. 5 shows an example in which only keywords are arranged in a two-dimensional space. (Step 405) Next, the distance calculation unit 208 calculates the distance between the arranged keyword and the document obtained as a result of the search in Step 402, based on the information in the matrix storage unit 207, and the document data mapping unit 210b creates a space. The storage means 211 is arranged in a two-dimensional space or a three-dimensional space. Details will be described later. FIG. 6 shows an arrangement example in the two-dimensional space. The circles in this figure are individual documents. (Step 406) Finally, the arranged result is output from the output means 202.

Arrangement of keywords in the space in step 404 will be described.

Regarding the distance between keywords (words),
It can be calculated from the matrix storage means 207. That is, these distances can be defined from the vector representation of words by the distance between vectors, the cosine between vectors, or the inner product of vectors. The distances between all the keywords to be arranged are calculated by the distance calculating means 208 in advance and are temporarily held in the form of a triangular matrix as shown in FIG.

Then, all the keywords are connected by a spring having a length corresponding to the distance calculated from the vector relation. In the initial state, the keyword is in an appropriate position, for example, 2
Arrange them at equal intervals on the same circumference in dimensional space. FIG. 28 or FIG. 30 schematically shows this state.
Each spring has no force when the length is a distance dm calculated from the relationship of vectors, but when it is longer than that, an attractive force is exerted, and when it is shorter, a repulsive force is exerted. This state is shown in FIG.
Shown in. In FIG. 32, the horizontal axis represents the distance and the vertical axis represents the strength of the attractive force. As the distance increases, the attractive force also increases. If the distance is shorter than the proper distance dm, a negative attractive force, that is, a repulsive force works. In other words, in the initial state, the springs are properly arranged regardless of the force of the springs, so each contact (word) has a force.

Next, with respect to the system composed of this contact and the spring, each contact is slightly moved to move the contact to a position where the system is stable, so that the entire system is in a stable state. This series of calculations will be described with reference to FIG. (Step 3301) First, the distance calculation means 208 calculates the distance between keywords based on the information in the matrix storage means 207, and temporarily holds it in the form of a triangular matrix as shown in FIG. (Step 3302) The keyword is regarded as a system in which the contact points are connected by a spring having a length corresponding to the distance, and the contact points are arranged at the initial positions in the two-dimensional (or three-dimensional) space. The initial position can be anything,
If it is a two-dimensional space, it should be placed on the circumference of the circle, and if it is a three-dimensional space, it should be on the same spherical surface. (Step 3303) The following steps 3304 and 3305 are repeated R times a prescribed number of times. (Step 3304) The following steps 3305 are processed for all the contacts. (Step 3305) For one contact i, the forces of all springs applied to the contact are calculated and combined. Then, the resultant force is directed in a certain direction, and the contact is moved in that direction by a minute distance k × α (r) × f according to the magnitude of the resultant force. f is the magnitude of the resultant force. k is a constant for converting the force into the distance.

Α (r) used in step 3305 is a parameter that decreases as the iteration progresses in accordance with the number of iterations from step 3303. For example, it is given by the following formula.

[0034]

## EQU1 ## α (r) = 1- (r / R)

Let R be the number of repetitions and r be a variable representing the number of repetitions. By using such parameters, the moving distance becomes shorter each time the iteration progresses, and as a result, the contact points converge to a position where the entire system is stable.

In the keyword arrangement in step 404, the existing method S is used instead of the spring model.
OM may be used.

The above is a detailed description of step 404. Next, the document arrangement in step 405 will be described.

The distance from the word to the document can be calculated from the matrix shown in FIG. The distance between the word p and the document q can be defined as a size proportional to the reciprocal of the matrix element (p, q). That is, if the word p in a document q appears many times, the matrix element (p, q) becomes large, and the reciprocal thereof becomes small. That is, it is closer. However, when a certain document b is much larger than a certain document a (the number of words is large), it may not be possible to simply compare. Therefore, the matrix elements require normalization processing such as division by the size of each document. These processes are calculated by the distance calculating means 208 based on the information in the matrix storing means 207.

The document placement in step 405 is calculated by applying the spring model. here,
Let us consider a system in which each document from already arranged words is connected by a spring whose length depends on the distance between the word and the document. An example of extracting one document is shown in FIG. In FIG. 7, the spring is schematically shown by a straight line. The difference of this system from the system of Fig. 28 and Fig. 30 is that
In the system of FIG. 7, all the words have already been arranged and are fixed. So you need to move just one document to find the document alignment where the spring is stable. On the other hand, FIG.
In a system like, we had to decide the position where the system (spring) was stable while moving all the contacts.

This calculation will be described with reference to FIG. (Step 3401) The distance calculation unit 208 calculates the distance from each word to the document based on the information in the matrix storage unit 207, and temporarily stores it. (Step 3402) The following process is performed for all documents. (Step 3403) Initially arrange the document. The initial placement can be anywhere. (Step 3404) The following step 3405 is repeated T times a prescribed number of times. (Step 3405) The forces of all springs applied to the document are calculated and combined. Then, the resultant force is directed in one direction, and in that direction, a small distance k depending on the magnitude of the resultant force.
The contact is moved by xα (r) × f. f is the magnitude of the resultant force. k is a constant for converting the force into the distance.
α (r) is a damping parameter similar to that used in step 3305.

The above is a detailed description of step 405.

The feature of the present invention is the order of step 404 and step 405. That is, first, a small number of keywords are arranged by using a spring model or the like, and then a large number of documents are arranged based only on the positional relationship with already fixed keywords. The distance between documents is not calculated.

In general, when a document is searched from a large database, the number Q of documents as a result of the search often reaches several hundreds. If the distance between documents is calculated by the existing method and self-organized classification is performed by a spring model or the like, it is basically necessary to calculate Q × Q times. Moreover, the distance calculation must be repeated while moving a small distance for each document until a position where the mutual distances are balanced, that is, a stable state of the system composed of springs is found. Therefore, Q × Q × R times (R Is repeated until it converges). Q is in the order of hundreds and R is in the order of hundreds to tens of thousands.

Moreover, the display result is a classification result of only the document as shown in FIGS. 20 and 22, and the user will not be able to understand it without proper labeling. For this labeling, assuming that the number of labels is P, P × Q × S
If calculation of the order of times (S is a constant) is required, calculation of the order of (Q × Q × R + P × Q × S) is required for these calculations.

In the case of the present invention, when compared with the above, the label corresponds to the keyword, that is, the number of keywords is P, and the self-organizing arrangement is P ×.
Since it is necessary to calculate P × Q × T times (T is a constant) in order to arrange Q documents using fixed keywords P × R times, these calculations require (P × P × R + P × Q ×
T) order calculations are required.

By the way, the number of users that can be used as the classification keyword is about 10 to 30 at most. That is, assuming that the number of documents is on the order of hundreds and the keywords are on the order of tens,

[0047]

[Equation 2] Q = 10 × P

The relationship of can be assumed. Therefore, in the existing method, in contrast to the calculation of the number of times of the order of (100 × P × P × R + 10 × P × P × S), in the present invention, (P ×
The calculation is completed with a smaller number of times (P × R + 10 × P × P × T) times. Although it depends on S and T, the calculation amount is about 1/10 to 1/100. Naturally, the memory used can also be small.

By the way, in FIG. 6, the document is simply arranged around the word. In order to present information to users in an easy-to-understand manner, emphasize words that attract more documents, that is, vector distance calculations using matrices (distance, cosine, dot product, etc.) that are closer to the document. It is also possible to display it. This is because by using the matrix storage unit 207 and the distance calculation unit 208 in FIG. 2, it is possible to immediately calculate which word is close to how many documents. FIG. 15 shows an example in which a keyword attracting more documents, that is, a number of documents equal to or larger than a certain threshold is highlighted. As a result, it is possible to analyze by what keyword the document set is classified.

On the contrary, it is also possible to highlight the keywords that have not attracted many documents. This is because by using the matrix storage unit 207 and the distance calculation unit 208 of FIG. 2 as in the previous example, it is possible to immediately calculate which word is close to how many documents. Keywords that have not attracted many documents are effective words for narrowing the search results when using them in a search, so what words should be searched for by highlighting them? Can show the guidelines. FIG. 16 shows an example in which keywords that have not attracted many documents are highlighted.

The data once classified and arranged can be selected by the GUI as shown in FIG. 23 and used for further analysis. For example, as shown in FIG. 24, in addition to the operation / processing of re-classifying and arranging only the selected document set, re-search is performed using only the selected word set as a search word, and keyword extraction is performed only from the selected document set. It is possible to perform redisplay and display, delete the selected document set or word set, and perform reclassification arrangement and re-search.

Further, it is possible to select a word and highlight a document close to the word (FIG. 25) or display a range of documents close to the word (FIG. 26).

It is also possible to dynamically change the document layout by selecting a word and moving the word on the layout (FIG. 27).

Further, in the present embodiment, about 10 to 20 keywords are arranged and then the document is arranged. However, in order to analyze about several tens of documents, for example, only the document is first subjected to the self-organizing method. You may arrange | position, and you may arrange | position several dozen to several hundred keywords only using the document arrange | positioned after that. In this case, keywords can be clustered by the arranged documents. Alternatively, by clustering keywords from documents, they can be used for idea support. In this way, by allocating a small number of data objects by the self-organizing method first, and then arranging another type of data object that is larger than that, relying on the data objects arranged after that, data with a small amount of calculation can be Placement / analysis is possible.

As described above, in the present embodiment, only keywords are first arranged in a space in a self-organizing manner, and then each document is arranged only from the distance from the arranged keywords. Therefore, the calculation can be completed at a far lower cost than the calculation with all combinations of documents, and moreover, it is possible to arrange appropriately for each document, and the use of words as labels makes the arrangement result easy to see. Effect is large.

(Second Embodiment) The second embodiment of the present invention will be described below. The second embodiment is a market data analysis device.

FIG. 8 is a block diagram of the apparatus of the second embodiment.

As shown in FIG. 8, this apparatus has means 801 for inputting search conditions, output means 802 for outputting search results and classification and arrangement results, search means 803 for searching,
Data temporary storage means 8 for storing data such as search results
04 and customer data storage means 80 for storing customer data
5 and product data storage means 806 for storing product information
And a queue storing means 8 for storing queue data of products and customers
07, distance calculation means 808 for calculating distances between products, customers, and product customers using a matrix, sales data storage means 809 for storing sales data, and space between these data based on distances between data. It is provided with analysis arrangement means 810 for arrangement on the top and space storage means 811 for storing space information. Here, the analysis arrangement unit 810 includes a product data mapping unit 810a and a customer data mapping unit 810b. The product data mapping unit 810a is described as a label mapping unit in the claims of the present application, and the customer data mapping unit is described as a data mapping unit in the claims of the present application.

In the matrix storage means 807, customers and products are recorded in a matrix of rows and columns as shown in FIG.
The element (i, k) of the matrix is the total number of purchases of the item i by the customer k. Further, it is assumed that the sales data storage unit 809 records when and which customer purchased which product.

FIG. 10 is a flow chart showing an outline of the use and operation of this device. (Step 1001) The user first inputs the search condition from the input means 801. The search condition is, for example, “Recent 3
The Top 20 Best Selling Products Per Month And The Customers Who Buy It "
And so on. (Step 1002) Then, the search means 803 searches the sales data storage means 809 and stores the set of products and customers as the search result in the data temporary storage means 804. (Step 1003) For the searched products, the distance calculation unit 808 calculates the distance between the products based on the information in the matrix storage unit 807, and the product data mapping unit 810a.
Are arranged in the two-dimensional space or the three-dimensional space of the space storage means 811. An existing method such as a spring model or SOM (self-organizing map) is used for arranging the product in the space. Products placed in the space are represented by product names. FIG. 11 shows an example in which only products are arranged in a two-dimensional space. (Step 1004) Next, the distance calculation unit 208 calculates the distance between the placed product and the customer as a result of the search in Step 1002, based on the information in the matrix storage unit 807,
The customer data mapping means 810b is the space storage means 81.
It is arranged in the two-dimensional space of 1 or the three-dimensional space. Two
FIG. 12 shows an example of arrangement in the dimensional space. The circles in this figure are individual customers. (Step 1005) Finally, the output means 802 outputs the arranged result.

In this way, the method of the present invention determines the first arrangement of the data represented by the matrix by using only the row data set or the column data set, and the other data only the already arranged data. By deciding the placements in and, the data analysis placement in the space is possible with less calculation. Even in the marketing data of customers / products, it is possible to cluster a large number of customers to be analyzed by purchased products. If only customer data is classified by self-organizing clustering or the like, it becomes difficult to understand as shown in FIG. 13, but by displaying the purchased product name, it is possible to display them in an easy-to-understand manner as shown in FIG. Becomes

Of course, in marketing analysis, etc.,
With respect to the data displayed here, by selecting a part of the data display window with the mouse as shown in FIG. 14, it will be possible to take out only the data of that part for further detailed analysis.

As described above, in the present embodiment, only the products are first placed in the space, and then each customer data is placed only from the distance to the placed product. Therefore, it is possible to finish the calculation at a far lower cost than calculating with all the combinations of customers, and it is possible to arrange appropriately for each customer, and by using the product as a label, the arrangement result is easy to see and its practical use Effect is large.

[0064]

As described above, according to the present invention, with respect to a data set that can be expressed in a matrix such as words and documents, products and customers, the smaller data is arranged in the space first, and the arrangement is used. By allocating the larger amount of data in the space, it is possible to realize an appropriate arrangement with a small amount of calculation, and by arranging both data in the space, one of the data can be used as a label. It is possible to obtain an easy-to-understand data arrangement result. Further, since the comparison with all the labels is performed when arranging the data of the larger amount, the influence of the labels arranged not only in the vicinity but also in the distance is also taken into consideration. As a result, it becomes possible to perform quick and accurate data analysis and data classification, and its practical effect is very large.

[Brief description of drawings]

FIG. 1 is a flowchart showing an outline of processing of the present invention.

FIG. 2 is a block diagram showing the configuration of a document classification device according to the first embodiment of the present invention.

FIG. 3 is a conceptual diagram showing a relationship matrix between a document and words according to the first embodiment of the present invention.

FIG. 4 is a flowchart showing an outline of operations / processes according to the first embodiment of the present invention.

FIG. 5 is a schematic diagram showing an example of a placement result of only words in the first embodiment of the present invention.

FIG. 6 is a schematic diagram showing an example of an output result according to the first embodiment of the present invention.

FIG. 7 is a conceptual diagram showing document layout calculation according to the first embodiment of the present invention.

FIG. 8 is a block diagram showing a market data analysis device according to a second embodiment of the present invention.

FIG. 9 is a conceptual diagram showing a relationship matrix between a product and a customer according to the second embodiment of the present invention.

FIG. 10 is a flowchart showing an outline of operations / processes according to the second embodiment of the present invention.

FIG. 11 is a schematic diagram showing an example of an arrangement result of only products according to the second embodiment of the present invention.

FIG. 12 is a schematic diagram showing an example of an output result according to the second embodiment of the present invention.

FIG. 13 is a conceptual diagram showing that data is difficult to see when a product name is not displayed in the second embodiment of the present invention.

FIG. 14 is a schematic diagram showing a GUI operation of an output result in the second embodiment of the invention.

FIG. 15 is a schematic diagram showing an example of word highlighting according to the second embodiment of the present invention.

FIG. 16 is a schematic diagram showing an example of word highlighting according to the second embodiment of the present invention.

FIG. 17 is a flowchart showing a process outline for the first example of the document classification method according to the conventional technique.

FIG. 18 is a flowchart showing a process outline of the second example of the document classification method according to the conventional technique.

FIG. 19 is a flowchart showing a processing outline for the third example of the document classification method according to the conventional technique.

FIG. 20 is a schematic diagram showing an example of the result of classification and arrangement of data and documents based on the conventional spring model.

FIG. 21 is a schematic diagram showing an example of cluster classification of data and document classification / arrangement result examples by a conventional spring model.

FIG. 22 is a schematic diagram showing an example of a classification result display using SOM of the related art.

FIG. 23 is a schematic diagram showing a GUI operation of an output result in the first embodiment of the invention.

FIG. 24 is a schematic diagram showing a reclassification placement operation for a subset according to the first embodiment of the present invention.

FIG. 25 is a schematic diagram showing related document highlighting processing according to the first embodiment of the present invention.

FIG. 26 is a schematic diagram showing a related document range display process according to the first embodiment of the present invention.

FIG. 27 is a schematic diagram showing word movement, document dynamic rearrangement operation, and processing according to the first embodiment of the present invention.

FIG. 28 is a schematic diagram showing a physical model of a spring model and its initial arrangement.

FIG. 29 is a schematic diagram showing a physical model of a spring model and an example of the final result thereof.

FIG. 30 is a schematic diagram showing a physical model in the case where the spring model has eight contacts, and an initial arrangement of the physical model.

FIG. 31 is a schematic diagram showing a matrix that stores the result of calculating the distance between words in the first embodiment of the present invention.

FIG. 32 is a schematic diagram showing the relationship between distance and force in a spring model.

FIG. 33 is a flowchart showing word placement processing according to the first embodiment of the present invention.

FIG. 34 is a flowchart showing a document arrangement process according to the second embodiment of the present invention.

FIG. 35 is a flowchart showing an outline of processing regarding the fourth example of the document classification method according to the conventional technique.

[Explanation of symbols]

201 Input means 202 output means 203 Search method 204 Search result storage means 205 document data storage means 206 word data storage means 207 matrix storage means 208 Distance calculation means 209 keyword extraction means 210 Analysis Arrangement Means 211 Spatial storage means 801 Input means 802 output means 803 Search method 804 data temporary storage means 805 Customer data storage means 806 Product data storage means 807 matrix storage means 808 Distance calculation means 809 Sales data storage means 810 Analysis Arrangement Means 811 Spatial storage means

Claims (14)

[Claims] 1. A data set included in the set A, wherein a set having a smaller number of data sets is used as a label indicating a data attribute of a data distribution from the two data sets stored in the storage means. Label mapping means for mapping each data object A in a three-dimensional space or less while maintaining a relative distance relationship between each data object of object A, and a set having a larger number of data are targeted for data analysis. A certain set B, and data mapping means for mapping the data object B in the space while maintaining the relative distance relationship between the data object B and the data object A which are the data included in the set B. An information analysis display device, characterized in that it comprises an analysis arrangement means. 2. The data mapping means performs mapping of the data object B by fixing the arrangement position in the space of the data object A which is the result of the label mapping means. Information analysis display device. 3. The output means for visually displaying the arrangement position of at least one of the data objects A and B in the space arranged by the analysis arrangement means. Information analysis display device described. 4. The information analysis display device according to claim 1, wherein one of the data objects A and B is word data and the other is text data. 5. The information analysis display device according to claim 1, wherein one of the data objects A and B is product data and the other is customer data. 6. The data included in one of the two data sets, which is used as a label indicating the data attribute of the data distribution, is the data object A, and the other data is included in the data set to be analyzed. As data object B, data object A as line,
Data object B as a column and data object A
And matrix storage means for storing a matrix having a value indicating the relationship between B and B as elements, and distance calculation for calculating distances between data objects A and data objects A and B stored by the matrix storage means. Means, space storage means for storing a space for arranging these data objects, output means for visually outputting information on the space of the arrangement result, and analysis arrangement for arranging the data objects in the space A label mapping means for arranging the data object A in the space using only the distance relationship between the data objects A calculated by the distance calculation means, and the arrangement. , The distribution of the data object B using the distance relationship between the data objects A and B calculated by the distance calculation means. An information analysis display device, comprising: a data mapping unit that determines a position and arranges a data object B in a space, and the output unit displays the arrangement result of at least one of the data objects A and B. 7. One of the data objects A and B is word data and the other is document data, and a value indicating the relationship between the data objects A and B is used as a frequency of appearance of words in the document. The information analysis display device according to claim 6. 8. The method according to claim 6, wherein one of the data objects A and B is product data, the other is customer data, and the value indicating the relationship between the data objects A and B is used as the product purchase frequency of the customer. Information analysis display device. 9. The information analysis according to claim 3, wherein the output unit highlights a data object A having more data objects B than a threshold at a distance shorter than a predetermined distance. Display device. 10. The information analysis display according to claim 3, wherein the output unit highlights a data object A having a data object B smaller than a threshold at a distance shorter than a predetermined distance. apparatus. 11. A set A having a smaller number of data sets is used as a label indicating a data attribute of a data distribution from the two data sets stored in the storage means for data analysis display on a computer, and Set A
With respect to the label mapping step of arranging the data object A in a space of three dimensions or less while maintaining the relative distance relationship between the respective data objects of the data object A that is the data included in the set A, A large set is set as a set B which is a target of data analysis, and the data object B in the space is arranged while maintaining the relative distance relationship between the data object B and the data object A which are the data of the set B. An information analysis display program for causing a computer to execute a data mapping step and a step of visually displaying the arrangement position of at least one of the data objects A and B mapped in the space. 12. A data object A is data included in a data set used as a label indicating a data attribute of a data distribution of the two data sets for data analysis and display on a computer, and the other is data. Matrix storage for storing data contained in a data set to be analyzed as a data object B, a data object A as a row, a data object B as a column, and a matrix having values indicating the relationship between the data objects A and B as elements Means, distance calculation means for calculating distances between the data objects A stored by the matrix storage means, between data objects A and B, and space storage means for storing a space for arranging these data objects. And a data object calculated by the distance calculation means A label mapping means for mapping the data object A in space using only the distance relationship between them and the distance relationship between the data objects A and B calculated by the distance calculation means based on the arrangement thereof. Information that functions as a data mapping unit that determines the arrangement of the data object B by mapping the data object B in space and an output unit that displays the mapping result of at least one of the data objects A and B. Analysis display program. 13. An information analysis display method for performing data analysis of two data sets and displaying an analysis result, wherein a set having a smaller number of data is selected from the two data sets stored in the storage means. A set A used as a label indicating a data attribute of a data distribution, and a data object in a space of three dimensions or less while maintaining a relative distance relationship between each data object of the data object A which is the data included in the set A A label mapping step of arranging A, a set having a larger number of data as a set B which is a target of data analysis, and a relative distance relationship between the data object B and the data object A which are the data of the set B are shown. A data mapping step of arranging and arranging a data object B in the space; Visually displaying the arrangement position of at least one of the marked data objects A and B. 14. An information analysis display method for performing data analysis of two data sets and displaying an analysis result, wherein data included in one of the two data sets is data object A and the other is A matrix storing step of storing a matrix having data contained in the set as a data object B, data object A as a row, data object B as a column, and a value indicating a relationship between data objects A and B as an element; A distance calculation step for calculating distances between the data objects A and between the data objects A and B stored by the storage means, and a space storage step for storing a space for mapping the data objects A and B, Distance relationship between data objects A calculated by the distance calculation means A label mapping step of mapping the data object A by using only in space, the based on the mapping result of the data object, calculated by the distance calculation unit data object A
A data mapping step of determining the arrangement of the data object B using the distance relationship between the data objects B and B and mapping the data object B in space, and an output step of displaying the mapping result of at least one of the data objects A and B Information analysis and display method consisting of and.