JP3425906B2 - Document retrieval apparatus, document retrieval method, and computer-readable recording medium recording a program for causing a computer to execute the method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document that describes, from a plurality of digitized documents, a document that meets an input search request and an event that is related to an event described in the relevant document. The present invention also relates to a document search device capable of performing a search, a document search method, and a computer-readable recording medium having a program for causing a computer to execute the method.

[0002]

2. Description of the Related Art Digitized documents are stored as a database in a storage device such as a hard disk, and when a search request is input by an operator, a document matching the search request is searched and displayed on a display screen or the like. A document search device for outputting is conventionally known.

[0003] Many of such devices judge whether or not the document is suitable for the search request based on whether or not the input search word is included in the document. However, although this method has a small processing load, noise (a document that includes the search word but does not match the search purpose of the operator) is easily mixed in the search result. Therefore, there is known a search method called a vector space method that can obtain a search result with less noise by paying attention to the entire meaning content of a document, not the presence or absence of a specific vocabulary.

In this search method, first, an n-dimensional vector is created that objectively and numerically expresses the semantic content of the input natural sentence and individual documents in the database. Each element value of the vector is determined according to a predetermined calculation formula based on the frequency of appearance of the corresponding vocabulary in the document.

Then, the degree of similarity between the vector of the input sentence and the vector of each document is taken as the matching degree of the document with respect to the input sentence. For example, the smaller the angle between the vectors,
The similarity is high, and the matching degree of the document with respect to the input sentence is high. Documents whose conformance exceeds a specified threshold,
Alternatively, a predetermined number of documents in descending order of relevance are usually displayed on the display or the like as relevance documents in order of relevance.

[0006]

However, in the above-mentioned prior art, the time when each relevant document is created, or the time relationship between them is not considered important. However, only the similarity between each document and the input search text is shown. Therefore, if you want to obtain some time-based knowledge from multiple documents that deal with similar subjects obtained by searching, for example, the history of a certain incident or the transition of a policy related to a certain matter, etc. It was necessary to sort the documents in order of creation date and time.

Further, in the above-mentioned prior art, the relation between the conforming document and the nonconforming document, especially the temporal relation between them is not presented to the operator. Therefore, the operator has a relationship between an event described in one conforming document and an event described in another nonconforming document, for example, whether there is a social factor that triggers this case at this time. When I wanted to analyze what kind of changes in the lives and consciousness of the citizens were brought about by the adoption of this policy at this time, I would like to separate the documents that were created at the same time as the relevant documents. There is a problem in that it is necessary to browse and find out the prominent tendencies in those documents and the events described in many of those documents exclusively by hand.

In order to solve the above-mentioned problems of the prior art, the present invention provides the operator with an easy-to-understand temporal front-rear relationship and co-occurrence relationship between events described in a plurality of electronic documents. An object of the present invention is to provide a document search device, a document search method, and a computer-readable recording medium in which a program for causing a computer to execute the method is recorded.

[0009]

[Means for Solving the Problems]
In order to achieve the object, a document search device according to the invention of claim 1 is a first document extraction means for extracting a document that matches an input search request from a plurality of digitized documents; The information about the date and time when the document extracted by the first document extracting unit is obtained, and the document created during the predetermined period before and after the date and time related to the obtained information is selected from the plurality of documents. Second document extracting means for extracting with reference to the date and time information of the document,
A plurality of documents which are not extracted by the first document extracting unit and which are created in a predetermined period before and after the date and time when the document extracted by the first document extracting unit is created Second document extracting means for extracting from documents, document similarity calculating means for calculating similarity between documents extracted by the second document extracting means, and document similarity calculating means And a document classification unit that classifies the documents extracted by the second document extraction unit by collecting documents whose similarity exceeds a predetermined threshold value.

According to the first aspect of the present invention, not only the document conforming to the input retrieval request but also the document created at the same time as the conforming document are retrieved.

According to a second aspect of the present invention, there is provided a document retrieval method, which comprises a first document extracting step of extracting a document that matches an input retrieval request from a plurality of digitized documents.
The information about the date and time when the document extracted by the first document extracting step is acquired, and the document created during a predetermined period before and after the date and time according to the acquired information is selected from the plurality of documents. A second document extracting step of extracting with reference to the information regarding the dates and times of the plurality of documents, and a document which is not extracted by the first document extracting means and is extracted by the first document extracting step. A second document extracting step for extracting a document created in a predetermined period before and after the date and time when the document was created, and the documents extracted by the second document extracting step. The document similarity calculation step of calculating the similarity and the second document extraction step by grouping together the documents whose similarity calculated by the document similarity calculation step exceeds a predetermined threshold value. Characterized in that it contains, and document classification step of classifying a more retrieved document.

According to the second aspect of the present invention, not only the document conforming to the input retrieval request but also the document created at the same time as the relevant document are retrieved.

Further, the recording medium according to the invention of claim 3 records a program for causing a computer to execute the method described in claim 2, so that the program becomes machine-readable, whereby the recording medium according to claim 2 is obtained. The operation of can be realized by a computer.

[0014]

BEST MODE FOR CARRYING OUT THE INVENTION With reference to the attached drawings, preferred embodiments of a document retrieval apparatus, a document retrieval method, and a computer-readable recording medium in which a program for causing a computer to execute the method is recorded according to the present invention. The form will be described in detail.

(Embodiment) First, a hardware configuration of a document search apparatus according to an embodiment of the present invention will be described. FIG. 1 is a block diagram showing the hardware configuration of the document search device according to the present embodiment.

In FIG. 1, 101 is a CPU that controls the entire system, 102 is a ROM that stores a boot program, 103 is a RAM used as a work area of the CPU 101, and 104 is an HD (under the control of the CPU 101). An HDD (hard disk drive) that controls reading / writing of data from / to a hard disk 105 and an HD that stores data written under the control of the HDD 104 are shown.

Reference numeral 106 denotes an FDD (floppy disk drive) for controlling the reading / writing of data from / to the FD (floppy disk) 107 under the control of the CPU 101, and 107 is a detachable device for storing the data written under the control of the FDD 106. FD of each.

Reference numeral 108 denotes a display for displaying a window relating to data such as a document, an image, and functional information, including a cursor, an icon or a tool box,
Reference numerals 109 denote interfaces (I / F) that are connected to the network NET via the communication line 110 and control the interface between the network NET and the inside.

Reference numeral 111 denotes a keyboard provided with a plurality of keys for inputting characters, numerical values, various instructions, and the like.
Is a mouse for moving the cursor, selecting a range, moving a window, changing a size, selecting an icon, moving, etc., 113 is a scanner for optically reading an image, and 114 is a window for printing the contents displayed in the window. 115 is a removable recording medium CD-
Reference numeral 116 denotes a ROM, and reference numeral 116 denotes a CD-ROM drive for controlling the reading of data from the CD-ROM 115. Further, reference numeral 100 denotes a bus for connecting the above-mentioned respective parts.

Next, the functional configuration of the document search device according to the embodiment of the present invention will be described. FIG. 2 is a block diagram functionally showing the configuration of the document search device according to the present embodiment. In FIG. 2, the document search device according to the present exemplary embodiment includes a file storage unit 200, a vector calculation unit 201, an input unit 202, and a first document extraction unit 20.
3, the second document extraction unit 204, and the document classification unit 205
And a document rearrangement unit 206 and a display unit 207. The vector calculation unit 201 also includes a query vector creation unit 201a and a document vector creation unit 201b.
And a vector similarity calculation unit 201c.

The file storage unit 200 is specifically realized by the hard disk 105 and the hard disk drive 104 which is a reading device of the hard disk 105, and stores a plurality of digitized documents. These documents are standardized in advance in a predetermined file format, specifically the S-JIS format.

The S-JIS format is a format similar to a structured document such as SGML. For example, the <body> tag is followed by the body of the document, the <title> tag is followed by the document title, and <au>.
The thor> tag is followed by the author name of the document, the <createddate> tag is followed by the document creation date and time, and the <pubdate> tag is followed by the publication date and time of the document (such as the date and time of publication in the publication). Has been done.

When adding a document of a different format, such as a document of a unique format created by a commercially available word processing software, to the file storage unit 200, the file format must be converted prior to the addition. The document search device according to the present embodiment has a function (filter) for converting files of various formats into the S-JIS format, but since it is not the central content of the present invention, a detailed description will be given. Omit it.

The file storage unit 200 includes a document vector creation unit 201, which will be described later, in addition to the above-mentioned document files.
b holds a plurality of search files used when creating the document vector of those documents. These search files describe information such as the frequency of appearance and the document in which all words and phrases included in the above document appear.

The vector calculation unit 201 has a function as "document similarity calculation unit" in claim 1, and includes a query vector creation unit 201a and a document vector creation unit 201.
b and a vector similarity calculation unit 201c.

The query vector creating unit 201a creates a query vector that numerically expresses the meaning of the meaning based on a search sentence input from the input unit 202 described later. The structure of the vector will be described later.

The document vector creation unit 201b creates a document vector of each document stored in the file storage unit 200 based on the search file stored in the file storage unit 200.

Here, the document vector is a vector having the same number of element values as all the words and phrases included in all the stored documents, and the meaning of the document is defined by the characteristics of the respective element values corresponding to the respective words and phrases. You can understand the contents numerically.

More simply, each element value of the document vector is determined by the frequency with which the corresponding phrase appears in the document. For example, if a phrase in the document appears once, the element value in the document vector corresponding to the phrase is "1", and if it appears 10 times, "10".
And, if it has not appeared at all, it becomes “0”.

However, for example, even if a document includes many words such as “bank”, if other documents in the same database also include many words, the degree to which the word is characteristic for the document is not. It is considered low. vice versa,
If a document contains only a few words "bank", but the word does not appear in any of the other documents,
It is considered that the word simply expresses the feature of the document when compared with other documents.

In view of such a situation, in the present embodiment, each element value of the vector is not the appearance frequency of a simple word, but
The calculation is performed in consideration of the statistical characteristics of the appearance location of the phrase, that is, the distribution status between documents or within one document. As described above, the file storage unit 20
Since these pieces of information are described in the search file stored in 0, the document vector creation unit 201b creates a document vector of each document by referring to these search files.

The vector similarity calculation unit 201c calculates the similarity between the query vector created by the query vector creation unit 201a and the individual document vectors created by the document vector creation unit 201b, or the document classification unit 2 described later.
The similarity between the document vectors created for the document input from 05 is sequentially calculated. Specifically, the vector similarity can be calculated by a predetermined calculation formula based on the inner product of them.

In order to calculate the similarity of the documents input from the document classification unit 205 described later, the document vector of each document and the document vector of each document are not actually used for every combination of those document vectors. All are regarded as one document, so to speak, and the degree of similarity with the document vector created for the document is calculated.

That is, for example, there are four documents a1, a2, a3, and a4 input from the document classification unit 205, and the respective document vectors are v1, v2, v3, and v4, and these four documents are regarded as one document. If the document vector of the document at that time is V, V and v1,
The similarity of four sets of document vectors of V and v2, V and v3, and V and v4 is calculated.

In this way, the vector calculation unit 201, which is designed only for calculating the similarity between the query vector and the individual document vectors, is also used for calculating the similarity between the individual document vectors. This is because Even if the document vector similarity is calculated with the above combination, it is substantially the same as v1 and v2, v1 and v3, and v1.
And v4, v2 and v3, v2 and v4, v3 and v4, which is the same as calculating the similarity for six sets of document vectors.

Since each vector is a numerical representation of the meaning and content of the search sentence or document, the similarity between the vector and the search sentence that is the source of the vector and the meaning and content of the document (in other words, The degree of matching of the document with respect to the search sentence), or the degree of similarity of the semantic content between the documents that are the sources of the vector.

The degree of similarity between the search sentence and the document, or the similarity between the document and the document calculated as described above, is determined by the functional unit that has input an instruction to the vector calculation unit 201 to perform the above-described processing, and more specifically, as will be described later. The first document extraction unit 203,
Alternatively, it is output to the document classification unit 205.

The input unit 202 is specifically the keyboard 1
11 and a mouse 112 to input various commands from the operator. To search for documents related to a certain item from the documents stored in the file storage unit 200,
The operator may search only the documents that match the search text together with the natural text search text, or create documents at the same time as those documents (not necessarily at the same time, but created at a time close to them). Specified documents) are also specified.

In the following, as an example of the former, (1) when the search sentence "Apollo 11 landed on the moon" is instructed and only documents matching the search sentence are instructed, 2) In addition to the search statement "Gross domestic product stagnation", the description will be made separately for each processing content when an instruction is issued to search for matching documents and documents created at the same time as those documents.

When the input command is any of the above (1) and (2), the input unit 202 outputs the input command to the first document extracting unit 203 described later. If the input command is neither of the above (1) and (2), the input command is output to another appropriate functional unit (not shown). Is determined and output destinations are distributed by, for example, a command analysis unit (not shown).

The first document extraction unit 203 has an input unit 202.
The search sentence included in the command input from is extracted and output to the vector calculation unit 201. The vector calculation unit 201 creates the query vector and the document vector as described above. Then, the degree of similarity between them is calculated, and the processing result is output to the first document extracting unit 203.

The first document extraction unit 203 extracts one of the documents stored in the file storage unit 200 based on the processing result input from the vector calculation unit 201 and a preset reference. For example, 20 documents in order from the one having the highest matching degree with respect to the search sentence, or all the documents having the matching degree exceeding a predetermined threshold value are extracted. The document extracted by the first document extracting unit 203 in this way is hereinafter referred to as a “matching document”.

In the present embodiment, the extraction criterion is assumed to be held in advance in the first document extracting section 203, but when the operator inputs a search sentence or the like, for example, the top 20 or the like. It may be possible to input the criteria together.

Further, the first document extraction section 203 searches the file storage section 200 to obtain the title of the relevant document. As described above, the document in the file storage unit 200 is S-
Since it is unified in JIS format, the title of the document can be obtained by searching the <title> tag in the document.

Here, by the processing according to the above-mentioned instruction (1), the titles of "Apollo 11 lunar landing", "Apollo 10 launch", "Apollo 13 failure report", and "Apollo 1 fire" are respectively given. 4 documents, above (2)
According to the above-mentioned processing according to the instruction of “GDP, 11.2
% Negative growth ”“ GDP growth rate in 1997 D Research Institute-
It is assumed that three documents having the titles of "0.5% prediction" and "" Japan growth rate 1.1% "IMF revised downward" have been extracted.

If the instruction input from the input unit 202 is the above (1), the first document extracting unit 203 uniquely identifies the file such as the title and file name (or the identification number) of the matching document. Anything is possible as long as it is possible. The same applies to the following)
Output to 6. If the instruction input from the input unit 202 is the above (2), the title and file name of the matching document are sent to the document sorting unit 206, and only the file name is described later in the second document. It outputs to each extraction part 204.

The second document extraction unit 204 searches the file storage unit 200 by the file name input from the first document extraction unit 203, and acquires the creation date and time of the relevant documents from the <createdate> tag. Then, the documents created during a predetermined period before and after the creation date and time of each relevant document, for example, three days of the current day of creation and the day before and the next day of the creation date are stored in the file storage unit 200. c
Search by referring to the reateddate> tag.

In the present embodiment, the predetermined period to be searched is held in advance in the second document extracting section 204, but when the operator inputs a search sentence or the like, for example, It is also possible to make it possible to input together the designation such as the day of creation of the matching document and the days before and after that.

Then, the second document extraction unit 204 uses the file name of the document acquired as described above to determine which relevant document the document was created at the same time (in other words, which document was created). Information is added so that it can be distinguished whether it is a document extracted corresponding to a matching document or not, and is output to the document classification unit 205 described later.

The document classifying unit 205 includes the second document extracting unit 2
The file names input from 04 are output to the vector calculation unit 201, and instructions are given to calculate the degree of similarity between documents having those file names, that is, the degree of similarity between those document vectors. The vector calculation unit 201 calculates the degree of similarity between these documents as described above, and outputs the processing result to the document classification unit 205.

The calculation of the degree of similarity is performed between documents corresponding to the same matching document. For example, as documents created at the same time as the conforming document A, a1, a2, a3, and a4
Assuming that the four documents b1, b2, and b3 are input as documents created at the same time as the conforming document B from the second document extraction unit 204, the similarity is calculated as a.
It is performed independently in the document group of 1, a2, a3, and a4 and in the document group of b1, b2, and b3, and the degree of similarity across the document groups of documents a1, b1, a2, b3, etc. is not calculated. .

The document classification unit 205 is a vector calculation unit 20.
Based on the processing result input from 1, the documents whose similarity exceeds a predetermined threshold value are grouped into one group, so that the documents retrieved by the second document extracting unit 204 may have several contents. Categorize documents into groups of similar documents.

For example, the documents retrieved by the second document extracting unit 204 for the matching document A are a1, a2, a3.
And there are four cases of a4, and the vector calculation unit 201 determines that the similarity of each document vector is 50 for v1 and v2.
0, v1 and v3 are 20, v1 and v4 are 400, v2
And v3 are calculated to be 50, v2 and v4 are calculated to be 350, and v3 and v4 are calculated to be 100. If the threshold value held in advance by the document classification unit 205 is 300, these four documents are It is classified into two groups, a group consisting of a1, a2 and a4 and a group consisting of a3.

Then, the document classification unit 205 outputs the above processing result to the second document extraction unit 204. The second document extraction unit 204 extracts one of the groups input from the document classification unit 205 according to a preset standard. For example, one group having the largest number of documents belonging to the group or all groups including two or more documents is extracted. Here, for the matching document A, only the group including the documents a1, a2, and a4 is extracted.

Further, the second document extracting section 204 determines the documents belonging to the extracted group, that is, the documents a1 and a in the above example.
For 2 and a4, the file storage unit 200 is searched to obtain the title of the document described in the <title> tag.

Here, for example, the conforming document “GDP, 11.2% minus growth” extracted by the first document extracting unit 203 in accordance with the above-mentioned instruction (2) and the title of the document created at the same time. "TSE average stock price 18,000
It is assumed that the yen is divided, the TSE falls below 18,000 yen, the difficult problems one after another and the market is bearish, and the TSE stock price is fully weakened.

In addition, the conforming document “Fiscal 1997 GDP Growth Rate D
Regarding the “-0.5% forecast by R & D,” the title of the document prepared at the same time as the document was “Expansion of cooperation with the US in an emergency.
"Japan and the United States agree on new defense guidelines" and "Chinese foreign ministers restraint on new defense guidelines"
% Regarding “IMF downward revision”, the titles of documents prepared at the same time as the document were “Listed by five former managing directors of Y Securities” and “Arrest of five managing directors of Y securities”.
It is assumed that each has been acquired.

Then, the second document extracting section 204 outputs the title and file name of the document obtained as described above to the document rearranging section 206 described later.

The document rearranging unit 206 searches the file storage unit 200 by the file name input from the first document extracting unit 203 and the second document extracting unit 204, and
Acquires the creation date and time of those documents described in the ateddate> tag. Then, first, the titles of the matching documents input from the first document extracting unit 203 are rearranged in the order of the creation date and time.

For example, in response to the instruction of (1) above, the first
The creation dates and times of the above four documents extracted by the document extracting unit 203 are July 20, 1969 and 1969, respectively.
If it was May 18, 2013, April 13, 1970, and January 27, 1967, the titles of the above four documents would be "Apollo 1 Fire", "Apollo 10 Launch", and "Apollo 11".
Sort by No. Landing on the moon, "Apollo 13 trouble report".

Similarly, the creation dates and times of the three documents acquired by the first document extraction unit 203 in accordance with the above-mentioned instruction (2) are September 12, 1997 and September 25, 1997, respectively.
If the date is 17th, the titles of the above three documents are
Sort by "GDP, 11.2% negative growth", "Japan growth rate 1.1%" IMF downward revision, and "GDP growth rate D Research Institute in 1997 forecast -0.5%".

Next, the document rearranging section 206 rearranges the titles of the documents input from the second document extracting section 204.
These are first sorted in the order of the creation date and time of the corresponding relevant document, sorted by groups having similar meanings in the documents corresponding to the same relevant document, and further, among documents belonging to the same group. They are sorted in order of oldest creation date.

When the titles of the documents input from the first document extracting unit 203 and the second document extracting unit 204 are rearranged in this way, the document rearranging unit 206 associates the titles of these documents with their creation dates and times. In addition, the display unit 20 described later
Output to 7.

The display unit 207 is specifically realized by the display 108, a display memory, and the like, and displays the title of the document and the creation date and time input from the document rearrangement unit 206 on the display screen.

FIG. 3 is an explanatory diagram showing an example of a screen finally displayed by the display unit 207 when the instruction (1) is input from the input unit 202. The titles and the creation dates and times of the matching documents extracted by the first document extraction unit 203 are sorted and displayed by the document sorting unit 206 in the order of the creation dates and times.

Further, FIG. 4 shows the above-mentioned (2) from the input unit 202.
FIG. 8 is an explanatory diagram showing an example of a screen finally displayed by the display unit 207 when the instruction is input. The titles and the creation dates and times of the matching documents are sorted in the order of the creation date and time, and the titles and the creating dates and times of the documents extracted by the second document extracting unit 204 are created at the same time as the matching documents. Are displayed according to the order of creation date and time.

In the present embodiment, the documents created at the same time as the conforming document are classified into several groups that are semantically similar to each other, and the groups are conspicuous to some extent (for example, a large number of documents belong to it). Although only the documents belonging to the document are extracted and presented to the operator, all the documents created at the same time as the matching document may be displayed.

In the present embodiment, the document created at the same time as the conforming document is displayed, but the document published at the same time as the conforming document may be displayed.
In order to realize this, the second document extraction unit 204 may refer to the <pubdate> tag instead of the <createddate> tag. In addition to the above creation date and publication date, refer to the transmission date and time, the reception date and time, and the last update date and time written on any tag to display the documents that were sent at the same time as the conforming document was sent. The operator may be allowed to specify so.

The file storage unit 200, vector calculation unit 201, input unit 202, first document extraction unit 203, second document extraction unit 204, document classification unit 205, document rearrangement unit 206 and display unit 207 are ROM10 respectively
2, the RAM 103 or the hard disk 105, the floppy disk 107 or the like, the CPU 101 or the like executes the command processing according to the command described in the program recorded in the recording medium, thereby realizing the function of each unit.

Next, a series of processes of the document search device according to the present embodiment will be described. FIG. 5 is a flowchart showing a procedure of processing of the document search device according to the present embodiment. The instruction input from the input unit 202 is first analyzed by a command analysis unit (not shown), and when it is determined that the instruction is a document search from the file storage unit 200, the process according to this flowchart is started.

In the flowchart of FIG. 5, first, in step S501, the first document extraction unit 203 and the vector calculation unit 2 called by the first document extraction unit 203.
By 01, a document matching the input search sentence is extracted from the documents stored in the file storage unit 200. In the following step S502, the step S501
The titles of the documents extracted in step 3 are acquired from the file storage unit 200, and the titles and file names of those documents are output to the document sorting unit 206.

In step S503, it is determined which of the designated instructions is to be searched, only the matching documents are searched, or the matching documents and documents created at the same time as those are searched. To do. When it is determined that the instruction is to retrieve only the matching documents (Yes in step S503), the process proceeds to step S508.

If it is determined that the instruction is to retrieve the matching documents as well as the documents created at the same time (NO at step S503), at step S504, the second document extracting unit 204 Step S501 from the documents stored in the file storage unit 200
A document created at the same time as the relevant document extracted in step S3 is searched, and the file name of the searched document is output to the document classification unit 205.

In step S505, the document classification unit 2
The vector calculation unit 201 called in 05 calculates the similarity between the documents extracted in step S504. Then, in step S506, the documents extracted in step S504 are classified by grouping the documents whose similarity exceeds a predetermined threshold value, and the processing result is output to the second document extraction unit 204. To do.

In step S 507, the second document extraction unit 204 acquires, from the file storage unit 200, the titles of the documents belonging to any (or all) groups input from the document classification unit 205. Then, the titles and file names of these documents are output to the document rearrangement unit 206.

In step S508, the document rearrangement unit 206 causes the first document extraction unit 203 and the second document extraction unit 203 to
The document storage unit 200 searches the file storage unit 200 for the creation date and time of the document input from the document extraction unit 204. Further, in step S509, the titles and creation dates and times of those documents are rearranged in the order of creation date and time, and the processing result is displayed on the display unit 2.
It outputs to 07.

Then, in step S510, the processing result input from the document rearrangement unit 206 is displayed on the display screen, and the processing according to this flowchart ends.

As described above, according to the present embodiment, the documents conforming to a certain retrieval request, in other words, the documents describing the same or similar phenomenon or subject are displayed in the order of the creation date and time. It is possible to easily understand the history of a certain matter.

Further, since the documents created at the same time as the conforming document are classified and displayed in groups having similar meanings, it is possible to easily know what happened before and after a certain event. .

The fact that the time of occurrence of the event is close may be just a coincidence, but it may indicate some relation that has already been discovered or has not yet been discovered. By searching for the documents created at the same time as the documents, it is possible to provide useful information to the operator who is going to investigate the relation between the events.

The document search method described in this embodiment is realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. This program is a hard disk, floppy disk, CD-ROM, M
It is recorded on a computer-readable recording medium such as O or DVD, and is read from the recording medium by the computer and executed. The program can be distributed via the recording medium and a network such as the Internet.

[0082]

As described above, according to the first aspect of the present invention, the first document extracting means extracts a document matching the input search request from a plurality of digitized documents,
The second document extracting means acquires information on the date and time when the document extracted by the first document extracting means is created, and the documents created in a predetermined period before and after the date and time according to the acquired information, From the plurality of documents, information regarding the dates and times of the plurality of documents is referred to and extracted, and the document similarity calculation unit calculates the similarity between the documents extracted by the second document extraction unit, The classification unit collects the documents whose similarity calculated by the document similarity calculation unit exceeds a predetermined threshold to classify the documents extracted by the second document extraction unit. Documents that meet the search request as well as documents that were created at the same time as the relevant documents are searched, and as a result, the temporal sharing between events described in multiple digitized documents is performed. There is an effect that the document search apparatus capable of presenting an easy-to-understand relationship operator is obtained.

According to the second aspect of the present invention, the first document extracting step extracts a document matching the input search request from the plurality of digitized documents, and the second document extracting step Of the plurality of documents, the information about the date and time when the document extracted by the first document extracting step is acquired, and the document created during a predetermined period before and after the date and time according to the acquired information From the document, the document similarity calculation step calculates the similarity between the documents extracted by the second document extraction step, and the document classification step calculates the document similarity. The documents extracted in the second document extraction step are classified by grouping together the documents whose similarity calculated in the index calculation step exceeds a predetermined threshold value, so that the documents satisfy the input search request. At the same time, documents that were created at the same time as the relevant documents are also searched, and this allows the operator to easily understand the temporal co-occurrence relationship between events described in multiple digitized documents. An effect is obtained that a document search method that can be performed is obtained.

According to the invention of claim 3, claim 2
By recording a program for causing a computer to execute the method described in (1), the program becomes machine-readable, and as a result, a recording medium capable of realizing the operation of claim 2 by a computer can be obtained. Play.

[Brief description of drawings]

FIG. 1 is a block diagram showing a hardware configuration of a document search device according to the present embodiment.

FIG. 2 is a block diagram functionally showing the configuration of the document search device according to the present embodiment.

FIG. 3 is an explanatory diagram showing an example of a screen displayed by a display unit 207 according to the present embodiment.

FIG. 4 is an explanatory diagram showing another example of a screen displayed by the display unit 207 according to the present embodiment.

FIG. 5 is a flowchart showing a processing procedure of the document search device according to the present embodiment.

[Explanation of symbols]

100 bus 101 CPU 102 ROM 103 RAM 104 HDD 105 HD 106 FDD 107 FD 108 display 109 I / F 110 communication line 111 keyboard 112 mice 113 scanner 114 printer 115 CD-ROM 116 CD-ROM drive 200 file storage 201 Vector calculator 201a Query vector creation unit 201b Document vector creation unit 201c vector similarity calculation unit 202 Input section 203 first document extraction unit 204 second document extraction unit 205 Document classification section 206 Document Sorting Unit 207 display

Continuation of the front page (56) References Japanese Patent Laid-Open No. 9-101990 (JP, A) Akihiko Yabu, Data search software Datahunter Ver. 1.1, INTERNET magazine, Japan, Impress Co., Ltd., April 1, 1990, No. 51, page 239 Information Retrieval and Utilization Software Let's Use Datahunter, Japan, Sharp Inc., July 9, 1999 , Pp. 1-16 Kazuo Sumita, Seiji Miike, Information Filtering Technology, Toshiba Review, Japan, Toshiba, January 1, 1996, Vol. 51, No. 1, pp. 42-44 Tetsuya Nasukawa, Masayuki Morohashi, Tohru Nagano, Special Issue Natural Language Processing to Expand Fields 2 Text Mining-Knowledge Discovery by Automatic Analysis of Enormous Document Data, Information Processing, Japan, Information Processing Society of Japan, April 15, 1999, Vol. 40, No. 4, pp. 358-364 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/30-17/30 419 JISST file (JOIS)

Claims (3)

(57) [Claims] 1. A first document extracting unit for extracting a document matching a search request input from a plurality of digitized documents, and a document extracted by the first document extracting unit is created. The information about the date and time that has been acquired, and extracting the documents created in the predetermined period before and after the date and time related to the acquired information from the plurality of documents by referring to the information on the date and time of the plurality of documents. Second document extracting means, a document similarity calculating means for calculating the similarity between the documents extracted by the second document extracting means, and the similarity calculated by the document similarity calculating means is a predetermined threshold. A document retrieving apparatus comprising: a document classifying unit that classifies the documents extracted by the second document extracting unit by collecting documents that exceed a value. 2. A first document extracting step of extracting a document that matches an input search request from a plurality of digitized documents, and a document extracted by the first document extracting step is created. The information about the date and time that has been acquired, and extracting the documents created in the predetermined period before and after the date and time related to the acquired information from the plurality of documents by referring to the information on the date and time of the plurality of documents. Second document extracting step, a document similarity calculating step of calculating the similarity between the documents extracted by the second document extracting step, and a similarity calculated by the document similarity calculating step having a predetermined threshold. A document retrieval method comprising: a document classification step of classifying the documents extracted by the second document extraction step by collecting documents that exceed a value. 3. A computer-readable recording medium on which a program for causing a computer to execute the method according to claim 2 is recorded.