JP2011242844A - Device, method, program and system for keyword extraction - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a keyword extraction device, a keyword extraction method, a keyword extraction program and a keyword extraction system which are able to accurately extract keywords on each of multiple documents contained in a document set, together with keywords in the whole document set.SOLUTION: A keyword extraction device includes: a document set keyword extraction unit 113 to extract keywords of a whole document set from the text information of the document set consisting of multiple documents; a document keyword extraction unit 114 to extract keywords of each of the documents contained in the document set from the text information of the document set, in consideration of the frequency of the appearance of the words across the whole document set; and a keyword determination unit 112 to determine keywords extracted by the document set keyword extraction unit 113 and also contained in a document and the keywords extracted by the document keyword extraction unit 114 to be the keywords of the document for each of the documents contained in the document set.

Description

  The present invention relates to a keyword extraction device, a keyword extraction method, a keyword extraction program, and a keyword extraction system.

  When it is desired to obtain an outline of a document composed of Japanese sentences, or when it is desired to prepare a search word for searching the document in advance, a method of extracting a keyword of the document is used. As a technique for extracting keywords from a Japanese document, each sentence constituting the document is decomposed into morphemes, and TF (word frequency: Term Frequency) -IDF (inverse appearance frequency: Inverse Document) for each morpheme. A technique is known in which an index representing importance such as a frequency value is calculated, and a morpheme having a high index is used as a keyword of the document.

  Also, for example, Patent Document 1 aims at classifying a plurality of documents for each topic, and extracts a keyword from the document, and based on the keyword, document aggregation for determining whether the documents handle the same topic Techniques relating to the method are disclosed.

  However, in the method of simply decomposing sentences into morphemes and calculating the importance for each morpheme, a collection of documents with a common topic, such as a slide document consisting of multiple pages, is unique to each document. There was a problem that keywords could not be extracted accurately.

  Here, in the case of a slide document, the entire slide can be regarded as an aggregate of a plurality of documents (= pages) by regarding each page as one document. Further, if a regular document is also a structured document, the entire document can be regarded as an aggregate of a plurality of documents (= structures) by regarding each structure as a single document.

  However, since each document constituting such a collection (also referred to as a document collection) is written about a common topic, when trying to extract keywords as general Japanese for each document, There is a problem that the difference in keywords between documents is reduced.

  On the other hand, when trying to extract important keywords in comparison with other documents that make up the collection, it tries to extract words that are different from those of other documents. There was a problem that keywords that expressed the topic would fall out.

  In the technique described in Patent Document 1, keywords are extracted from documents of similar topics, but keywords that consider both topic keywords common to documents and keywords specific to each book are extracted simultaneously. The problem of not being able to do so could not be solved.

  Therefore, the present invention makes it possible to extract a keyword of each document of a document set made up of a plurality of documents, and to extract a keyword that can distinguish between the documents while leaving a keyword representing the document set itself. An object of the present invention is to provide a keyword extraction device, a keyword extraction method, a keyword extraction program, and a keyword extraction system.

  In order to achieve the above object, the keyword extracting apparatus according to claim 1 includes a set word extracting unit for extracting a keyword as a whole document set from text information of the document set based on a plurality of documents, and a text of the document set. A document word extracting means for extracting a keyword in each document of the document set from information in consideration of an appearance frequency in the whole document set, and for each document in the document set, the keyword extracted by the set word extracting means And a keyword determining unit that uses a keyword included in the document and a keyword extracted by the document word extracting unit as keywords of the document.

  The invention according to claim 2 is the keyword extraction device according to claim 1, wherein the collective word extraction means stores a cost storage table storing associations between morphemes and singularities of the morphemes as Japanese. With reference to this, the specificity of each morpheme constituting each document is calculated, and keywords as a whole document set are extracted based on the specificity.

  According to a third aspect of the present invention, in the keyword extracting device according to the first or second aspect, the document word extracting means determines the appearance frequency of the morpheme in the document and the reverse appearance frequency of the morpheme in the entire document set. Based on this, keywords in each document are extracted.

  The invention described in claim 4 is the keyword extraction device according to any one of claims 1 to 3, wherein the keyword extracted by the collective word extraction unit, the keyword extracted by the document word extraction unit, Is provided with keyword display means for displaying the information in an identifiable manner.

  According to a fifth aspect of the present invention, in the keyword extracting device according to any one of the first to fourth aspects, the text information of the document set is input by a plurality of documents input by the user or by the user. Based on documents extracted from electronic files.

  The invention described in claim 6 further includes an input means for inputting a plurality of documents or electronic files, and an output means for displaying the extracted keywords in the keyword extracting device according to claim 5. Is.

  The keyword extraction method according to claim 7 includes: a set word extraction process for extracting a keyword as a whole document set from text information of a document set based on a plurality of documents; and a document set from text information of the document set. Document word extraction processing for extracting keywords in each document in consideration of the appearance frequency in the entire document set, and keywords included in the document among the keywords extracted by the set word extraction processing for each document in the document set And keyword determination processing using the keywords extracted by the document word extraction processing as keywords of the document.

  According to another aspect of the present invention, there is provided a keyword extraction program that extracts from a text set of a document set based on a plurality of documents a set word extraction process for extracting a keyword as a whole of the document set and a text information of the document set. A document word extraction process for extracting a keyword in each document of the document set in consideration of the appearance frequency in the whole document set, and a keyword extracted by the set word extraction process for each document in the document set. The keyword determination process using the included keyword and the keyword extracted by the document word extraction process as the keyword of the document is executed.

  The keyword extraction system according to claim 9 is a keyword extraction system including a client terminal connected to the keyword extraction device via a network, and the keyword extraction device includes text information of a document set based on a plurality of documents. A word extraction unit for extracting a keyword as a whole document set, and a word extraction for extracting a keyword in each document of the document set in consideration of the appearance frequency in the whole document set from the text information of the document set. And, for each document in the document set, a keyword determination unit that uses a keyword included in the document among keywords extracted by the set word extraction unit and a keyword extracted by the document word extraction unit as a keyword of the document And the client terminal has a plurality of documents Input means for inputting, those having an output means for displaying the keyword keyword determining means has determined, the.

  According to the present invention, it is possible to accurately extract a keyword of each document constituting a document set together with a keyword of the entire document set from a document set including a plurality of documents.

It is an example of the hardware block diagram of a keyword extracting device. It is an example of the schematic block diagram of a keyword extraction system. It is an example of a functional block diagram of a keyword extraction device and a client terminal. It is a flowchart which shows the outline | summary of a keyword extraction process. It is an example of a document input screen. It is a flowchart which shows the detail of a keyword determination process. It is a flowchart which shows the detail of a set word extraction process. It is explanatory drawing of a cost storage table. It is a flowchart which shows the detail of a document word extraction process. It is an example of a keyword display screen. It is a functional block diagram of the keyword extraction apparatus and client terminal which concern on other embodiment. It is another example of a keyword display screen. It is another example of a keyword display screen. It is a functional block diagram of the keyword extracting device which concerns on other embodiment.

  Hereinafter, the configuration according to the present invention will be described in detail based on the embodiment shown in FIGS.

The keyword extracting apparatus according to the present embodiment is a set word extracting unit (set word extracting unit 113) that extracts a keyword as a whole document set from text information of a document set based on a plurality of documents (hereinafter also referred to as a document set). ) And a document word extracting means (document word extracting unit 114) for extracting a keyword in each document of the document set from the text information of the document set in consideration of the appearance frequency in the entire document set, and each document in the document set A keyword determination unit (keyword determination unit 112) that uses a keyword included in the document among keywords extracted by the collective word extraction unit and a keyword extracted by the document word extraction unit as a keyword of the document; It is equipped with.
Is.

  More specifically, first, the keywords of the entire document set are extracted by decomposing the document of the entire document set consisting of Japanese into morphemes and calculating an index representing specificity as Japanese for each morpheme, On the other hand, the keywords of each document constituting the document set are extracted by calculating an index representing the specificity with respect to the entire document set, and the keywords in the entire document set are included in the document. In this case, the keyword is also used as the keyword of the document.

(Keyword extraction device / keyword extraction system)
An example of a hardware configuration diagram of the keyword extracting apparatus according to the present embodiment is shown in FIG. The keyword extraction device 100 is composed of, for example, a general-purpose server, workstation, personal computer, and the like, and includes a CPU 131, a memory 132, a display device (output means) such as a display screen 134 via a display adapter 133, a printer, A serial port 136 for connecting an external input / output device 135 such as a FAX or a scanner, a storage device (ROM or the like) 137, and a bus 140 for interconnecting input devices (input means) such as a keyboard 138 or a pointing device 139 are included. . In addition, many devices such as a voice interface 141 and a network interface 142 including a wireless LAN can be connected. For example, through the network interface 142, an electronic file transfer such as e-mail and FTP, and a network service 143 such as WWW can be used.

  FIG. 2 shows a schematic configuration diagram of the keyword extraction system. The keyword extraction system 300 is configured by connecting the client terminal 200 to the keyword extraction device 100 via a network 301 such as an intranet or the Internet. As the client terminal 200, for example, a general-purpose personal computer can be used and has the same hardware configuration as that in FIG.

  The keyword extraction apparatus 100 in the keyword extraction system 300 extracts a keyword specific to each document and a keyword as a whole document set for a plurality of documents input by the user at the client terminal 200 and displays the result on the client terminal 200. Is.

  Specifically, for example, the user inputs a plurality of documents on the browser 201 of the client terminal 200, and the documents are transmitted to the keyword extracting device 100 via the network 301. The keyword extracting device 100 will explain below. As described above, the keyword extraction process for the document set is executed, the extraction result is displayed on the client terminal 200, and the user can check the extracted keyword on the browser 201.

  Next, an example of a functional block diagram of the keyword extracting device and the client terminal according to the present embodiment is shown in FIG. The keyword extraction apparatus 100 determines, as the processing unit 110, a keyword for each of a document receiving unit 111 that receives a document as text information from the client terminal 200 via the network 300, and a plurality of documents received by the document receiving unit 111. A keyword determination unit 112 for extracting a keyword as a whole document set composed of a plurality of documents, and for each document in the document set composed of a plurality of documents, the document is compared with the whole document set The document word extraction unit 114 for extracting words characteristic of the keyword as a keyword, the keyword display unit (keyword display unit) 115 for displaying the keyword determined by the keyword determination unit, and the storage unit (DB unit) 120 as a word Holds the cost of expressing the high specificity of Japanese as Japanese And a strike storage unit 121. In addition, since the some document in this embodiment extracts a common keyword, it is natural that it is suitable for a topic (content) to have a common part.

  The client terminal 200 receives the display result from the keyword display unit 115 and displays the result via the browser 201. Note that the client terminal 200 is not indispensable, keyword extraction processing is executed for a document input from an input device such as the keyboard 138 of the keyword extraction device 100, and the display result of the keyword display unit 115 is displayed on the display screen of the keyword extraction device 100. You may make it display on display apparatuses, such as 134 (refer FIG. 14 mentioned later).

(Keyword extraction process)
Next, an outline of the keyword extraction process (keyword extraction method according to the present invention) executed by the keyword extraction device 100 will be described with reference to the flowchart of FIG.

  First, from the document input screen 202 displayed on the display device of the client terminal 200, the user inputs a plurality of documents as text as shown in FIG. The plurality of documents are received as text information (text information of a document set) by the document receiving unit 111 (S101).

  The keyword determination unit 112 passes the document set to the set word extraction unit 113 and acquires a keyword for the entire document set as a return value (S102). Further, the keyword determination unit 112 passes the same document set to the document word extraction unit 114, and acquires a keyword in each document as a return value (S103).

  Then, the keyword determining unit 112 determines a keyword for each document based on each keyword acquired from the collective word extracting unit 113 and the document word extracting unit 114 (S104). Finally, it is passed to the keyword display unit 115 together with the original documents, and the keyword display unit 115 displays the keyword extraction result (S105).

  Further, the keyword determination process by the keyword determination unit 112 will be described with reference to the flowchart of FIG. The keyword determination unit 112 first obtains text information of a plurality of documents input by the user from the document reception unit 111 (S201). Here, the number of documents to be received is n.

Next, the text information of each document is passed to the set word extraction unit 113, and keywords word _{0,1 to} word _{0, p} as the entire document set are obtained as return values (S202, corresponding to S102 in FIG. 4). Note that the subscript 0 indicates a keyword of the document set, and p indicates the number of keywords of the document set determined by the set word extraction unit 113 (described later).

Further, the text information of the same document is transferred to the document word extraction unit 114, and the keywords word _{i, 1 to} word _{i, q} of each document are obtained as return values (S203, corresponding to S103 in FIG. 4). Note that the subscript i indicates the document number, i = 1,..., N. Q indicates the number of keywords extracted from one document determined by the document word extraction unit 114 (described later).

Then, for the document i (i = 1,..., N), word _{i, 1 to} word _{i, q} is a keyword specific to the document, and the document i of words _{0,1 to} word _{0, p} Are the keywords of the document i related to the topic of the document set. That is, the two types of keywords are combined and determined as the keyword of the document i (S204, corresponding to S103 in FIG. 4). Details of each process will be described below.

<Aggregate word extraction processing>
Details of the processing by the collective word extracting unit 113 will be described with reference to the flowchart of FIG. The collective word extraction unit 113 first obtains text information of the document from the keyword determination unit 112 (S301).

  Next, the following processing is executed for each document (S302 to S306). First, the document i is decomposed into morphemes (S303). The morpheme decomposition process may be a known or new method, and is not particularly limited. For example, a known morpheme analysis engine such as ChaSen or MeCab may be used as a library. it can.

  Then, the number included in the document is counted for each morpheme, and the appearance frequency of the morpheme j in the document i is set to F (i, j) (S303). Here, it is preferable that the frequency of appearance is not counted except for the part of speech that can easily become a keyword such as a noun as a keyword candidate.

Furthermore, the TF value (word frequency: Term Frequency) of each morpheme in the document is calculated by Equation 1 and added to the TF value tf (j) in the document set (S304). The TF value is a value obtained by dividing the number of appearances of a certain word (morpheme j) in a certain document (document i) by the number of variations of words appearing in the document (morpheme number).

  If the value of tf (j) on the right side of Equation 1 does not exist at the time of calculating this equation, it is set to 0. If the value of tf (j) has already been obtained by another document, the value is added to that value.

  The above processing (S303 to S304) is performed for all documents (S305, S306). As a result, tf (j) exists as many as the number of morpheme types included in the document set. However, morphemes excluded from keyword candidates are excluded.

As a result, the product of the obtained tf (j) and idf (j) that is a unique value corresponding to the morpheme j is obtained (S307, the following equation (2)).
tf · idf (j) = tf (j) · idf (j) (2)

  Here, the IDF value (Inverse Document Frequency) is a value (an index indicating the level of specificity) obtained by the difficulty of appearance of the morpheme as Japanese.

  The IDF value is preferably recorded in the storage unit 120 as a database using morphemes as keys. In the present embodiment, the cost storage table 122 as shown in FIG. 8 is recorded in the cost storage unit 121 of the storage unit 120.

  The cost storage table 122 is a dictionary-type database in which corresponding IDF values are stored using each morpheme as a key. The IDF value calculation method may be a known or new method, and is not particularly limited. For example, a set of highly general documents such as news articles is prepared, and all documents included in the set are calculated. The number obtained by dividing the number by the number of documents containing morpheme j in the document, or by using the full-text search engine on the Internet and searching the number of documents that can be searched by the search engine using morpheme j A natural logarithm can be calculated and obtained for a value obtained by dividing by a number.

The TF-IDF value (tf · idf (j)) obtained by the above equation (2) is an index representing the singularity of the morpheme j in the document as Japanese. Finally, the set word extraction unit 113 outputs p morphemes (return values, word _{0,1 to} word _{0, p} ) in descending order of tf · idf (j) as keywords of the document set (S308). Note that p is an arbitrary value, and can be, for example, a value obtained by dividing n by a constant.

<Document word extraction processing>
Details of processing by the document word extracting unit 114 will be described with reference to a flowchart of FIG. The document word extraction unit 114 first obtains text information of a document from the keyword determination unit 112 (S401).

  Next, the following processing is executed for each document (S402 to S406). First, the document i is decomposed into morphemes (S403). The decomposition process into morphemes may be performed in the same manner as the collective word extraction unit 113.

  Then, the number included in the document is counted for each morpheme, and the appearance frequency of the morpheme j in the document i is set to tf (i, j) (S403). It should be noted that, except for parts of speech that can easily become keywords, such as nouns, it is preferable that the keywords are not candidates and the appearance frequency is not counted.

Further, all the morphemes j included in the document and not excluded from the keyword candidates are set to the following expression (3) (S404).
N (j) = N (j) +1 (3)

  If N (j) on the right side of the equation (3) does not exist at the time of calculating this equation, the value is set to zero. When the above processing (S403 to S404) is performed on all documents, the number of documents including morpheme j is N (j) (S405, S406).

Next, an IDF value for each morpheme j handled in the document set is obtained by the following equation (4) (S407).
idf (j) = log (n / N (j)) (4)

  The value of idf (j) calculated by the above equation (4) represents the difficulty of appearance of morpheme j, but here it is calculated by the logarithm of (number of documents in document set / number of documents including morpheme j). Therefore, in the set word extraction process (FIG. 7), the value of idf (j) represents the difficulty of appearance as Japanese, whereas in the document word extraction process, the morpheme j “appears in the document set” It is a value that represents “difficulty in nodding”.

  A TF-IDF value in each document i of each morpheme j is calculated from tf (i, j) and idf (j) obtained by the above processing (S408, similar to the above equation (2)). Here, since the value of idf (j) is a value indicating “difficult to appear in the document set”, the value of TF-IDF is more specific than the other documents in the document set. The value will be higher for higher morphemes.

Finally, the document word extraction unit 114 calculates q morphemes j (return values, word _{i, 1 to} word _{i, q} ) for each document _{i in} descending order of tf · idf (i, j). It outputs as a keyword (S408). Note that q is an arbitrary value, for example, a value obtained by dividing the number of words in each document by a certain constant. In this case, q can be a different value depending on i.

<Keyword display process>
Moreover, the keyword display means 115 displays the keyword extracted by the above process (S201-S204 of FIG. 6). FIG. 10 shows an example of the keyword display screen 203. In the example shown in FIG. 10, for the document input by the user and the extracted keywords, the keywords (keywords of the document set) in all three documents are displayed as enclosing characters, and keywords specific to each document are displayed as underlines. Two types of keywords are displayed so that they can be easily identified. Also, for example, the keywords in the whole document may be displayed in different colors, such as blue characters and keywords unique to each document in red characters.

(Other embodiments)
Another embodiment of the keyword extracting device according to the present invention will be described. In addition, the description about the same point as the said embodiment is abbreviate | omitted.

  In the above-described embodiment, an example in which the user inputs text from the document input screen 202 has been described. However, it is also preferable that the user performs input using an electronic file.

  An example of a functional block diagram of the keyword extracting device and the client terminal according to the present embodiment is shown in FIG. In order to enable input using an electronic file, it is necessary to perform processing for extracting text information from the electronic file sent from the client terminal 200.

  Therefore, as shown in FIG. 11, the keyword extraction device 100 according to the present embodiment includes a text extraction unit 116 that extracts text information from an electronic file in addition to the components of the above embodiment. In this embodiment, since the document reception unit 111 receives an electronic file instead of text information, the document reception unit 111 passes the received electronic file to the text extraction unit 116, and the text extraction unit 116 converts the electronic file into text information. The document receiving unit 111 receives the text information. The subsequent processing is the same as in the above embodiment.

  FIG. 12 shows an example of the keyword display screen 203 when the user inputs an electronic file. The example shown in FIG. 12 shows a display example in the case of handling a file in which a plurality of slides are collected as a collection of documents, in which keywords of the document collection are indicated by enclosing characters and keywords specific to each document are indicated by underlining. Yes. As a result, the user can easily know both the topic of the slide as a whole and the contents of the slide in the slide simply by looking at the keyword of each slide. .

  It is also preferable to search for a file having the keyword by using the keyword extracted by the keyword extracting device as a trigger. For example, when a single document is regarded as a document set of each page and the search result is displayed, a page having the keyword in the document is displayed as shown in FIG. By displaying other keywords, it is possible to easily know the contents of the search results. In the example shown in FIG. 13, the keyword of the entire document is made large, the keyword specific to the page is made small, and the search word is displayed as an enclosing character so that it can be easily identified. It is easy to know what the word is, what the document is, and what is written on the page to which the search keyword belongs.

  Further, as shown in FIG. 14, the text of the document is input from the input device of the keyword extraction device 100 without using the client terminal 200, and the keyword extraction result is displayed on the output device of the keyword extraction device 100. It is also preferable.

  The keyword extraction device 100 according to the present embodiment includes a document input unit 117 instead of the document reception unit 111. The document input unit 117 according to the present embodiment displays a document input screen (see FIG. 5) on which a user inputs a document on the display device of the keyword extraction device 100, and the keyword input unit 117 uses the document input by the user as text information. It is what you pass.

  The keyword display unit 115 displays a keyword display screen (see FIG. 10) on the display device. In the example shown in FIG. 14, when the user inputs a document as an electronic file, the text extraction unit 116 may be further provided as described above.

  The keyword extraction processing by the keyword extraction device described above can be realized by a program (keyword extraction program) that causes a computer to execute keyword extraction processing. The keyword extraction program is preferably provided by, for example, downloading from the Internet and installed on a computer. Further, the present invention is also applied to an aspect of a recording medium (recording medium on which a keyword extraction program is recorded) on which a keyword extraction program is recorded so as to be executable by a computer.

  The above-described embodiment is a preferred embodiment of the present invention, but is not limited thereto, and various modifications can be made without departing from the gist of the present invention.

DESCRIPTION OF SYMBOLS 100 Keyword extraction apparatus 110 Processing part 111 Document receiving part 112 Keyword determination part 113 Collected word extraction part 114 Document word extraction part 115 Keyword display part 116 Text extraction part 117 Document input part 120 Storage part 121 Cost storage part 122 Cost storage table 131 CPU
132 memory 133 display adapter 134 display screen 135 external input / output device 136 serial port 137 storage device 138 keyboard 139 pointing device 140 bus 141 voice interface 142 network interface 143 network service 200 client terminal 201 browser 202 document input screen 203 keyword display screen 300 keyword Extraction system 301 network

JP 2006-293616 A

Claims (9)

A set word extracting means for extracting a keyword as a whole of the document set from text information of the document set based on a plurality of documents;
A document word extracting means for extracting a keyword in each document of the document set in consideration of the appearance frequency in the whole document set from the text information of the document set;
For each document in the document set, a keyword determination unit that uses a keyword included in the document among keywords extracted by the set word extraction unit and a keyword extracted by the document word extraction unit as a keyword of the document. When,
A keyword extraction device comprising:   The set word extraction means calculates a specificity of each morpheme constituting each document with reference to a cost storage table storing a correspondence between the morpheme and the specificity of the morpheme as Japanese. The keyword extracting apparatus according to claim 1, wherein a keyword as a whole of the document set is extracted based on the keyword.   The document word extracting unit extracts a keyword in each document based on the appearance frequency of the morpheme in the document and the reverse appearance frequency of the morpheme in the whole document set. The keyword extraction apparatus in any one.   The keyword display means for displaying the keyword extracted by the collective word extraction means and the keyword extracted by the document word extraction means in an identifiable manner. The keyword extraction device described in 1.   5. The text information of the document set is based on a plurality of documents input by a user or a document extracted from an electronic file input by a user. Keyword extraction device.   6. The keyword extracting apparatus according to claim 5, further comprising: an input unit that inputs the plurality of documents or the electronic file; and an output unit that displays the extracted keyword. A set word extraction process for extracting a keyword as a whole of the document set from text information of the document set based on a plurality of documents;
A document word extraction process for extracting a keyword in each document of the document set in consideration of the appearance frequency in the whole document set from the text information of the document set;
For each document in the document set, a keyword determination process using a keyword included in the document among keywords extracted by the set word extraction process and a keyword extracted by the document word extraction process as keywords of the document When,
A keyword extraction method characterized by: On the computer,
A set word extraction process for extracting a keyword as a whole of the document set from text information of the document set based on a plurality of documents;
A document word extraction process for extracting a keyword in each document of the document set in consideration of the appearance frequency in the whole document set from the text information of the document set;
For each document in the document set, a keyword determination process using a keyword included in the document among keywords extracted by the set word extraction process and a keyword extracted by the document word extraction process as keywords of the document When,
A keyword extraction program characterized by causing A keyword extraction system comprising a client terminal connected to a keyword extraction device via a network,
The keyword extraction device includes:
A set word extracting means for extracting a keyword as a whole of the document set from text information of the document set based on a plurality of documents;
A document word extracting means for extracting a keyword in each document of the document set in consideration of the appearance frequency in the whole document set from the text information of the document set;
For each document in the document set, a keyword determination unit that uses a keyword included in the document among keywords extracted by the set word extraction unit and a keyword extracted by the document word extraction unit as a keyword of the document. And comprising
The client terminal is
Input means for inputting the plurality of documents;
And an output means for displaying the keyword determined by the keyword determination means.