JPH01209528A - Electronic document retrieving device - Google Patents

Abstract

PURPOSE:To improve the document retrieving efficiency by securing such a constitution where a retriever retrieves the documents by means of a scale that can clearly designate the value in a retrieving mode using an index of interval scales and applies the analysis of the main component to the retrieved document in terms of another scale. CONSTITUTION:A retrieving result analyzing means 5 applies the analysis of the main component to the document retrieved by a retrieving means 4 in terms of an axis which is not included in the retrieving conditions. A main component selecting means 6 selects the main components shown as the analyzing results. An example selecting means 7 selects an example document of the main component selected by the means 6 based on the information on the retrieving conditions received from an input means 1 and the output of the means 5. Then a retrieving result display information producing means 8 produces the display information on the main component selected by the means 6 based on the output of the means 7. As a result, the information that can be clearly designated is utilized and also easily designated in accordance with the characteristics of the retrieving result. At the same time, a highly efficient retrieving key is automatically obtained to improve the retrieving efficiency.

Description

[Detailed description of the invention] (Industrial application field) The present invention relates to a search device for electronic documents.

(Prior Art) When searching for documents, images, etc., it is sometimes necessary to search based on ambiguous images. Examples include when a designer searches for something that matches his or her image, or when searching for something previously seen based on memory.

As an indexing method for such a search, a method is used in which a plurality of attributes having an interval scale property are treated as coordinate axes, and each document image is indexed as a point in a multidimensional coordinate space. In a search using this index method, search conditions are described as values on several axes.

(Problem to be solved by the invention) This method decomposes and describes an image into elements represented by several axes, but decomposing an image in this way is only possible to a certain extent. . Therefore,
Although the searcher can clearly specify values for a small number of axes, he or she cannot specify values for other axes at all, resulting in a problem in that sufficient search accuracy cannot be obtained. For example, when retrieving facial photos based on memory, facial features are broken down into elements such as eye size, eyebrow darkness, and harshness, and an interval scale for each element is determined. It is known how to configure However, since the memory of a face is retained as an overall impression, the searcher can only clearly point out the value of a few particularly conspicuous features on the factorized scale. Therefore, it has been pointed out that there is a problem that sufficient search accuracy cannot be obtained.

In order to solve the above problems, images are not decomposed into elements, but divided into multiple groups according to their overall similarity.
A method has been proposed for narrowing down the search by presenting representative examples of each group and having the searcher specify which example most closely resembles what he or she wants to search for. However, this method has the disadvantage that accurate searches cannot be performed because information on features that can be clearly pointed out by the searcher cannot be effectively used.

For example, if a feature that a searcher can clearly point out is not reflected in the grouping, there may be a case where there is no difference between representative examples with respect to the feature that a searcher can clearly point out. In such cases, the searcher may not be able to specify the relevant group, or may specify a group that does not include what he or she wants to search, requiring a great deal of effort.

■The desired document cannot be found.

Problems such as this arise.

The first purpose of the present invention is to solve such problems by utilizing information on characteristics that can be clearly specified by the searcher, analyzing portions that cannot be clearly specified, and presenting the results. The purpose is to improve search efficiency. More specifically, in a search using an index based on an interval scale, the present invention performs a search using a scale that allows the searcher to clearly specify a value, performs principal component analysis on the retrieved documents regarding other scales, By presenting the analysis results,
This is intended to improve search efficiency.

A second object of the present invention is to provide a method in which, in a search using an index based on an interval scale, principal component analysis results regarding a scale that is not used as a search condition of a retrieved document are used to indicate the browsing direction. It's about doing.

A third object of the present invention is to provide a method in which, in a search using an index based on an interval scale, a principal component analysis result regarding a scale that is not used as a search condition of a retrieved document is used as a new search element. It's about doing.

(Means for Solving the Problems) According to the present invention, there is provided a search device for a device that stores and displays gauntleted documents, which includes an input means for inputting instruction information regarding a search for the electronic document; a document storage means for storing documents; a multidimensional spatial structure storage means for storing a multidimensional spatial structure for organizing electronic documents according to a plurality of axes;
a search means for searching for a corresponding document from the documents stored in the document storage means according to the search condition input by the input means; and a search means for searching the document searched by the search means;
a search result analysis means for performing principal component analysis on axes not included in the search conditions; a principal component selection means for selecting one or more principal components from the principal components found by the search result analysis means; example selection means for selecting an example document of the principal component selected by the selection means; analysis result presentation information creation means for creating display information for the principal component selected by the principal component selection means; a display means for displaying data and a display output from the analysis result presentation information creation means; and a display means for displaying data from the document storage means and data from the search result analysis means in accordance with the output of the input means; A computerized document retrieval device is obtained, comprising a control means for supplying information on the display means via the creation means.

Further, according to the present invention, there is provided a search device for a device that stores and displays digitized documents, which includes an input means for inputting instruction information regarding search and selection of the digitized documents, and a search device for storing and displaying digitized documents. a storage means; a multidimensional spatial structure storage means for storing a multidimensional spatial structure for organizing electronic documents according to a plurality of axes; a search means for searching for relevant documents; a search result analysis means for performing principal component analysis on axes not included in the search conditions on the documents searched by the search means; and a principal component obtained by the search result analysis means. a principal component selection means for selecting one or more principal components from the above; and an example selection means for selecting an example document of the principal component selected by the principal component selection means;
analysis result presentation information creation means for creating display information of the principal component selected by the principal component selection means; and a corresponding document selected from the documents searched by the search means according to the document selection information input by the input means. a display means for displaying the data from the document storage means and the display output from the analysis result presentation information creation means; a display means for displaying the data from the document storage means and the data from the search result analysis means; A computerized document retrieval apparatus is obtained, comprising a control means for controlling according to the output of the input means and supplying the analysis result presentation information on the display means via the analysis result presentation information creation means.

Further, according to the present invention, there is provided a retrieval device for a device that stores and displays digitized documents, comprising an input means for inputting instruction information regarding search and extraction of the digitized documents; a storage means; a multidimensional spatial structure storage means for storing a multidimensional spatial structure for organizing electronic documents according to a plurality of axes; a search means for searching for relevant documents; a search result analysis means for performing principal component analysis on axes not included in the search conditions on the documents searched by the search means; and a principal component obtained by the search result analysis means. a principal component selection means for selecting one or more principal components from the principal component selection means; an example selection means for selecting an example document of the principal component selected by the principal component selection means; and a principal component selected by the principal component selection means. analysis result presentation information creation means for creating display information; document extraction means for selecting a corresponding document from the documents searched by the search means based on the document extraction information input by the input means; and the document storage means. display means for displaying data from the document storage means and display output from the analysis result presentation information creation means; and display means for displaying data from the document storage means and data from the search result analysis means according to outputs from the input means, and displays the analysis results. An electronic document retrieval device is obtained, comprising a control means for supplying presentation information onto the display means via a presentation information creation means.

(Operation) In a search using an index based on an interval scale, the present invention performs a search using a scale that allows the searcher to clearly specify a value, performs principal component analysis on other scales on the retrieved documents, and calculates the principal components. Visually display representative examples and their locations. An example of display information on the display means is shown in FIG.

In this example, the first to third principal components are displayed. As a representative example of each principal component, a reduced version of the document closest to each of the maximum, median, and minimum principal component scores is presented.

Further, this display information can be used to input document selection instruction information. For example, in FIG. 4, when the position information of the star on the axis indicating the first principal component is input from the input means, the principal component score closest to the position indicated by the input position in the search document set is input. The document is displayed.

Furthermore, this display information can be used to input search conditions. For example, in the state shown in Figure 4, if the information of the rectangle indicated by a solid line on the axis indicating the second and third principal components is input, the scores of the second and third principal components fall within the specified range. A certain document is extracted.

The principal component obtained here is a composite of the scales for which the searcher cannot clearly specify the value, and at the same time, it is the scale along which the variance of the search result documents is the largest, and the overall It is the one that best represents the characteristics. Therefore, even if the searcher cannot specify the value on each decomposed scale, it is expected that it will be possible to specify the principal component score, which is the comprehensive characteristic value. Furthermore, since the principal components are orthogonal to each other, using principal component scores as search conditions is the most efficient method.

In addition, since the principal component is a composite of multiple scales,
It is a measure that itself has no name.

Usually, the principal components are interpreted from the contribution rate of the underlying scale, etc.
It often gives itself a name. However, the interpretation and naming of principal components is done subjectively and is often unreasonable. Therefore, in the present invention, the scale of each principal component is shown using an example. This method does not provide any interpretation to the obtained principal components and does not use abstract information such as words as a medium, so problems such as individual differences in interpretation and naming do not occur. Furthermore, since concrete information such as example documents can be used, it is easy to understand, and
It is possible to absorb individual differences and temporal fluctuations in the interpretation of divided scales.

(Example) Hereinafter, the present invention will be explained with reference to the drawings.

FIG. 1 is a block diagram showing a first embodiment of the present invention. The input means 1 of this embodiment is, for example, a pointing device such as a mouse or a trackball. It is also possible to use an input means such as a keyboard, with a slight sacrifice in operability. This allows you to specify or change document search/extraction conditions and input document selection information. The document storage means 2 is a means for storing a plurality of electronic document data, and is composed of, for example, an optical disk device or a magnetic disk device.

The multidimensional spatial structure storage means 3 is a means for storing axis numbers, axis centers, and number information of electronic documents corresponding to each position on each axis for a plurality of axes. The search means 4 is controlled by a control means 10 described later, and the input means 1
This means searches for a corresponding document from the documents stored in the document storage means 2 according to the search conditions inputted by.

The search result analysis means 5 performs principal component analysis on the documents searched by the search means 4 regarding axes that are not included in the search conditions. FIG. 5 is a functional block diagram showing one embodiment of the search result analysis means 5. As shown in FIG. The correlation matrix creation means 51 of this embodiment creates a correlation matrix regarding axes that are not included in the search conditions. The principal component calculation means 52 is a means for calculating principal components from the correlation matrix created by the correlation matrix creation means 51. Calculating the contribution rate of the obtained principal components is
This is contribution rate calculation means 53.

The principal component selection means 6 is means for selecting principal components to be presented as analysis results. Specific examples of methods for selecting principal components include selecting principal components with a contribution rate greater than a predetermined value, selecting principal components whose cumulative contribution rate reaches a predetermined value, etc. Alternatively, there is a method in which the user manually selects it.

The example selection means 7 is means for selecting example documents of the principal components selected by the principal component selection means 6 based on the search condition information output from the input means 1 and the output of the search result analysis means 5. Specific examples of how to select example documents include dividing the range of each principal component score into several ranges and selecting the document closest to the center of each divided range, and selecting the document with the maximum score for each principal component. , the median value, the method of selecting the document closest to the minimum value, etc.

FIG. 6 shows an example of the detailed processing of the example selection means 7 when the range of each principal component score is divided into several ranges and the document closest to the center of the divided range is selected. It is expressed as a flowchart. First, the first principal component is set as the applicable principal component, and in step 72, the applicable principal component score of the search result document is calculated. Next, in step 73, the retrieved document set is sorted in ascending or descending order for the corresponding principal component scores and set in the document number/principal component score table. Next, in a range division step 74, the range of the relevant principal component is divided. Using the sorting result in step 73, the search result document set is divided in step 75 according to which of the divided ranges it is included. Next, set the center of each range in step 7
Calculate by 6. In selection step 77, step 7
Using the division information created in step 5, select the document closest to the center of gravity for each range, and use this as an example document of the corresponding principal component. In step 78, the division information and example document information of the corresponding principal component are set in the example document table. Step 79
If there is a next principal component, it is used as the relevant principal component and the process returns to step 72. If there is no next principal component, a processing end signal is output.

The analysis result presentation information creation means 8 is means for creating display information regarding the principal component selected by the principal component selection means 6 based on the output of the example selection means 7. That is, the analysis result presentation information creation means 8 is a means for setting principal component numbers and scale information for displaying headings, and display position information of exemplary documents, and its actual operation is controlled by a control means 10, which will be described later.

The display means 9 displays the data from the document storage means 2 and the display output from the analysis result presentation information creation means 8, and the display means 9 is constituted by an ordinary CRT display or the like.

Further, the control means 10 is a means for controlling a series of search processing, and controls the data from the document storage means 2 and the data from the search result analysis means 5 according to the output of the input means 1, and the means for creating analysis result presentation information. Display means 9 via 8
It is a means of supplying

FIG. 7 is a flow chart showing an example of a detailed procedure of the control means 10. First, when a data input signal is given from the input means, an operation instruction is given to the search means 4. When a processing end signal is given from the search means 4, an operation instruction signal is given to the search result analysis means 5. When a processing end signal is given from the search result analysis means 5, an operation instruction signal is given to the principal component selection means 6. Similarly, after the operation of the principal component selection means 6 is completed,
The selection means 7 is operated. When the operation of the item selection means 7 is completed, the first principal component of the example document table set by the means is set as the corresponding principal component. Axis display step 10
6, the corresponding principal component number, axis line, and scale information are displayed. Next, in step 107, the first example document of the relevant principal component is determined to be the relevant document. At step 108, information on the relevant document is read from the document storage means 2. Step 109
If it is determined in step 110 that the size of the example document is too large for the displayable area, it is reduced to an appropriate size in step 110.

Display example document information or reduced information Step 11
1, it is displayed at a predetermined position on the display means 9.

The process proceeds to step 112, and if there is the next example document, it is determined as the relevant document in step 113, and the process returns to step 108. If there is no next example document, the process advances to step 114 to check whether there is the next principal component, and if so, it is used as the relevant principal component and the process returns to step 106. If there is no next principal component, the process proceeds to step 116, displays the current relevant document on the display means 9, sets the position information of the relevant document, and ends the process.

Next, FIG. 2 shows a second embodiment of the present invention. Here, the same numbers as the means in FIG. 1 indicate means having similar functions. The document selection means 11 is means for selecting a corresponding document from the search result documents according to the document selection instruction information inputted from the input means 1. That is, the document selection means 11 uses the document number/principal component score table created by the individual selection means 7 to select a value for the principal component specified by the document selection instruction information to a value corresponding to the specified position. Select the closest document.

FIG. 8 is a flowchart showing an example of detailed processing of the control means 10 when the input means 1 is implemented by a mouse. Steps 100 to 117 are the same processes as in FIG. In a mouse click detection step 118, it is detected whether the mouse is being clicked, and if the mouse has been clicked, the process proceeds to step 119. If the process is not deemed to be finished in step 119, the coordinates corresponding to the mouse cursor are detected in step 120. The position information of the currently displayed document set in step 117 is compared with the position information indicated by the detected coordinates, and if it is determined that the position is at a position where another document is to be displayed, the document selection means 11 , select the relevant document. Then, the process returns to step 116 for displaying the corresponding document. If, for example, the end key is pressed or the end mouse button is clicked in step 119, and the process is deemed to have ended, the operation is completed.

Next, FIG. 3 shows a third embodiment of the present invention. Here, the same numbers as the means in FIG. 1 indicate means having similar functions. The document extraction means 12 is a means for extracting a corresponding document from the retrieved document set based on the document extraction information inputted from the input means 1. That is,
The document extraction means 12 uses the document number/principal component score table created by the piece selection means 7 to extract documents whose principal component scores specified by the document extraction information are included in the specified range.

FIG. 9 is a flowchart showing an example of detailed processing of the document extraction means 12. First, in step 81, the first principal component included in the document extraction condition is determined to be the applicable principal component, and the first column of the intermediate table is determined to be the applicable column.

Next, the document number/principal component score table is referred to, and the number of the document included in the extraction target range of the corresponding principal component is set in the corresponding column of the intermediate table. In step 84, it is determined whether or not there is the next principal component, and if so, it is set as the relevant principal component, the next column in the intermediate table is set as the relevant column, and the process returns to step 83. If there is no next principal component, step 87
Proceed to.

If no document number is set in the intermediate table, an error message is displayed on the display means 9. If a document number is set in only one column of the intermediate table, its contents are set in the extraction result table. If document numbers are set only in a plurality of columns in the intermediate table, the AND result is set in the extraction result table. Once the data has been set to the extraction result table or the error message has been displayed, the process ends.

FIG. 10 is a flowchart showing an example of detailed processing of the control means 10 when the input means 1 is realized by a mouse and the extraction range is specified by a rectangle. Steps 100 to 119 are the same processes as in FIG.

In step 118, it is detected whether the mouse has been clicked, and if it has been clicked, step 119
Proceed to. If it is not determined that the process has ended in step 119, the process proceeds to step 123, where it is determined whether the document extraction conditions have been set. If the document extraction condition settings are also determined, the extraction conditions are input in step 124. Based on the input extraction conditions, a corresponding document is extracted using the document extraction means 12, and the process is terminated. If, for example, the end key is pressed or the end mouse button is clicked in step 119, and the process is deemed to have ended, the operation is completed.

(Effects of the Invention) As explained above, according to the present invention, in a search system using an index based on an attribute axis having an interval scale property, information that can be clearly specified by a searcher is utilized, and
Search clues that are easy for the searcher to specify and are efficient are automatically constructed according to the characteristics of the search results, thereby improving search efficiency. Furthermore, since the headings are visually displayed using specific examples, it is easy to grasp the characteristics of search results and use new search clues, improving the user interface.

[Brief explanation of the drawing]

FIG. 1 is a functional block diagram showing a first embodiment of the electronic document retrieval device of the present invention. Further, FIGS. 2 and 3 are block diagrams showing second and third embodiments of the present invention, respectively. Furthermore, in order to facilitate understanding of the method of the present invention, FIG. 4 is a diagram showing a specific example of displaying principal component analysis results and its use for document selection instructions and search condition input, and FIG. 1 is a flowchart showing an example of detailed processing of the search result analysis means 5 in the embodiment, FIG. 6 is a flowchart showing an example of detailed processing of the example selection means 7 in the embodiment of FIG. 3 is a flowchart showing an example of a detailed procedure of the control means 10 in the illustrated embodiment. FIG. 8 is a flowchart showing an example of a detailed procedure of the control means 10 in the second embodiment. FIG. 9 is a flowchart showing an example of a detailed process of the document extraction means 12 in the third embodiment, and FIG. 10 is a flowchart showing an example of a detailed procedure of the control means 10 in the third embodiment. In the figure, 1... input means, 2... document storage means, 3... multidimensional spatial structure storage means, 4... search means, 5... search result analysis means, 6... main Component selection means, 7... Example selection means, 8... Analysis result presentation information creation means, 9...
- Display means, 1O... Control means, 11... Document selection means, 12... Document extraction means.

Claims (3)

[Claims] (1) A search device for a device that stores and displays digitized documents, which includes an input means for inputting instruction information regarding the search for the digitized documents, a document storage means for storing the digitized documents, and a plurality of a multidimensional spatial structure storage means for storing a multidimensional spatial structure for organizing digitized documents according to axes; and a search for searching for a corresponding document from the documents stored in the document storage means based on search conditions input by the input means. means and
a search result analysis means for performing principal component analysis on axes not included in the search conditions on the documents searched by the search means; and a search result analysis means for analyzing one or more principal components from the principal components found by the search result analysis means. a principal component selection means for selecting, an example selection means for selecting an example document of the principal component selected by the principal component selection means, and an analysis result presentation for creating display information for the principal component selected by the principal component selection means. information creation means; display means for displaying the data from the document storage means and display output from the analysis result presentation information creation means; and input means for displaying the data from the document storage means and the data from the search result analysis means. and control means for controlling according to the output of the analysis result presentation information and supplying it to the display means via the analysis result presentation information creation means. (2) A search device for a device that stores and displays digitized documents, comprising an input means for inputting instruction information regarding search and selection of the digitized documents, and a document storage means for storing the digitized documents; A multidimensional spatial structure storage means for storing a multidimensional spatial structure for organizing electronic documents according to a plurality of axes, and a search condition input by the input means to search for a corresponding document from the documents stored in the document storage means. a search means for performing principal component analysis on axes that are not included in the search conditions on the documents retrieved by the search means; principal component selection means for selecting a plurality of principal components; example selection means for selecting example documents of the principal components selected by the principal component selection means; and display information for the principal components selected by the principal component selection means. analysis result presentation information creation means; document selection means for selecting a corresponding document from documents searched by the search means according to the document selection information input by the input means; and data from the document storage means. a display means for displaying a display output from the analysis result presentation information creation means; and a display means for controlling the data from the document storage means and the data from the search result analysis means according to the output of the input means, and the analysis result presentation information creation means and control means for supplying the information to the display means via the electronic document retrieval device. (3) A search device for a device that stores and displays digitized documents, comprising input means for inputting instruction information regarding search and extraction of the digitized documents, and document storage means for storing the digitized documents; A multidimensional spatial structure storage means for storing a multidimensional spatial structure for organizing electronic documents according to a plurality of axes, and a search condition input by the input means to search for a corresponding document from the documents stored in the document storage means. a search means for performing principal component analysis on axes that are not included in the search conditions on the documents retrieved by the search means; principal component selection means for selecting a plurality of principal components; example selection means for selecting example documents of the principal components selected by the principal component selection means; and display information for the principal components selected by the principal component selection means. an analysis result presentation information creation means; a document extraction means for selecting a corresponding document from the documents searched by the search means based on the document extraction information input by the input means; and data from the document storage means. a display means for displaying a display output from the analysis result presentation information creation means; and a display means for controlling the data from the document storage means and the data from the search result analysis means according to the output of the input means, and the analysis result presentation information creation means and control means for supplying the information to the display means via the electronic document retrieval apparatus.