JP2004361992A - Related word extracting device, related word extracting method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To properly extract related words in a certain field even when any document in another field is not prepared, or the related words do not necessarily co-occur in one document. <P>SOLUTION: An object field acquiring part 512 stores an object field inputted from an input device 3 in an object field storage buffer 521. A document reading part 514 reads a document in a document database 601, and retrieves a document having field information matched with the object field acquired by the object field acquiring part 512. The document having the matched field information is stored in a field temporary storage buffer 525, and morphemic analysis is executed to the stored document. Then, related words are tabulated and extracted from the result of the morphemic analysis by referring to a predetermined field related word notation pattern under the consideration of the rhetoric expression of the document having the field information. In this case, the related words whose appearance frequency is low can be removed by using a threshold as necessary. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a related word extraction technique, and more particularly, to a related word extraction device, a related word extraction method, and a program for extracting a related word corresponding to a predetermined field from a document database composed of documents in which field information is determined. .
[0002]
[Prior art]
Related words or synonyms are useful information for performing document search or automatic classification. Conventionally, while there has been an attempt to systematically edit related words or synonyms by hand, an attempt has been made to extract related words or synonyms by mechanical processing using an electronic computer.
For example, there is a method of using mutual information as a degree of association between words (for example, Patent Document 1). Mutual information is an index indicating the strength of association such that the two words of interest appear naturally rather than by chance. When the mutual information is low, the two words appear by chance. It is understood that it is only. In Patent Document 1, words registered in a document database are clustered, and related word information is created by this clustering. The related word of the word input by the user refers to the related word information and is presented for each cluster having a high degree of relatedness.
[0003]
Also, paying attention to the fact that mutual information can only extract so-called related words, use linguistic features to specify the degree of relevance between words, and specify relationships such as synonyms, broad words, and low words (For example, Patent Document 2). In Patent Document 2, similarity between words having a co-occurrence relationship is calculated with reference to word co-occurrence information. On the other hand, the similarity between words is calculated by referring to the linguistic features of the words. These two similarities are integrated, and when the obtained similarity is higher than a predetermined level, it is presented as having a similar relationship.
[0004]
[Patent Document 1] JP-A-2002-32394 (page 10)
[Patent Document 2] JP-A-2000-222427 (page 12)
[Problems to be solved by the invention]
In the above-described prior art, it is difficult to eliminate the relationship between words that appears in any document by the method using the frequency of co-occurring words. In addition, the method based on mutual information is devised so that the mutual information between words that appears in any document is low, but whether this mechanism functions as intended is It depends on the method of selecting a group of documents used to calculate the mutual information. For example, the words "invention" and "problem" are considered to be mutually related concepts, but when calculating mutual information using patent publications, most patent publications use the words "invention" and "problem". , The amount of mutual information tends to be relatively low. In order to increase the amount of mutual information between "invention" and "problem", it is necessary to input a document group that makes it obvious that these words co-occur. Therefore, when it is attempted to extract related words in a certain field A, if possible, a unique related word frequently appearing in a document in the field A is extracted unless documents such as a field B, a field C, etc. different from the field A are combined. There is a danger that you cannot do it. Further, as a problem common to the related art described here, there is a premise that related words co-occur. According to the above-described conventional technology, for example, it is not possible to associate the “computer” described in the document A with the “electronic computer” described in another document B.
[0005]
Therefore, the present invention has been made in order to solve the above-described problems, and when extracting related words in a certain field, a document in another field cannot be prepared. It is an object of the present invention to provide a related word extraction device, a related word extraction method, and a program that appropriately extract related words even when the related words do not co-occur.
[0006]
[Means for Solving the Problems]
In order to achieve the above object, a related-word extracting apparatus according to the present invention includes a document database having a plurality of documents, an input unit for inputting field information of the document, and a document having field information matching the input field information. Means for extracting a document from the document database, natural language analyzing means for analyzing the document extracted by the document extracting means in a natural language, and extracting field-related words related to the field of the document from the document in the document database. Field-related word notation pattern holding means for holding field-related word notation patterns, and field-related word extraction from documents extracted from a document database by referring to field-related word notation patterns based on the results of natural language analysis It is characterized by comprising word extraction means and field-related word counting means for counting the frequency of appearance of the extracted field-related words. It is set to.
[0007]
Next, a related word extraction method according to the present invention extracts a document in a specific field from a document database having a plurality of documents, and extracts a related word corresponding to the specific field from the extracted document. And an input step of inputting field information of the document, a document extracting step of extracting a document having field information matching the input field information from the document database, and A field-related word notation pattern holding step for holding a field-related word notation pattern for extracting a related field-related word, and a document extracted from a document database by referring to the field-related word notation pattern based on the result of natural language analysis A field-related word extracting step of extracting the field-related words from the field, and a field for counting the appearance frequency of the extracted field-related words It is characterized by comprising a communication word aggregation step.
[0008]
Further, the program of the present invention extracts a document in a specific field from a document database having a plurality of documents, and outputs a related word extraction device that extracts a related word corresponding to the specific field from the extracted document. An input function for inputting field information, a document extraction step of extracting a document having field information that matches the input field information from a database, and field-related words related to the field of this document are extracted from documents in the document database A field-related word notation pattern holding function for storing field-related word notation patterns for performing a search, and field-related word notation patterns are referred to based on the results of natural language analysis, and field-related words are extracted from documents extracted from a document database. Realize a field-related word extraction function and a field-related word counting function that counts the frequency of appearance of extracted field-related words. It is characterized in that.
[0009]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings.
The related-word extraction device according to the present invention stores a document group of a plurality of fields in a magnetic disk device in advance, and when a predetermined field is input from an input device, selects a document matching the field information from the document group. Then, words related to the field information (hereinafter referred to as "field-related words"; the same applies hereinafter) are extracted from the selected document and displayed on the display device. Each document in the document group of a plurality of fields stored in the magnetic disk device holds field information in a fixed format.
[0010]
Next, a related word extracting apparatus according to the present invention will be described with reference to FIGS.
FIG. 1 is a block diagram showing a configuration of a related word extraction device according to an embodiment of the present invention.
This related word extraction device selects a document corresponding to a predetermined input field from a document group, focuses on a specific rhetorical expression found in the selected document, and extracts a field-related word based on the rhetorical expression. Extract.
The related word extraction device 1 includes a control device 2, an input device 3, a display device 4, a memory 5, and a magnetic disk device 6, and each unit is connected to each other via a bus 7.
[0011]
The control device 2 is a central processing unit (CPU), reads an OS (operating system) stored in the magnetic disk device 6 and a predetermined program into a program unit 51 described later, and reads the entire related word extraction device 1. And the data transfer process between the devices.
The input device 3 is used to input a character string, various data, and a command, and includes a keyboard, an OCR, a pen, a mouse, a tablet, or a touch panel.
[0012]
The display device 4 displays data such as data input by the input device 3, instructions to the user from the related word extraction device 1, and data such as a finally obtained related word extraction result, for example, from a CRT or a liquid crystal display. Be composed.
As shown in FIG. 2, the memory 5 includes a program unit 51 for reading and storing a predetermined program from the magnetic disk device 6 for the control device 2 to execute various controls and processes, and necessary for each process. The buffer unit 52 temporarily stores data.
[0013]
FIG. 3 is a block diagram showing a configuration of the buffer unit 52 of the memory 5 of the related word extraction device 1. The buffer unit 52 includes a target field storage buffer 521, a document storage buffer 522, a morphological analysis result storage buffer 523, a field related word notation pattern storage buffer 524, a field temporary storage buffer 525, and a related word counting buffer 526. , And a temporary variable storage buffer 527.
The target field storage buffer 521 is for storing the field information acquired by the target field acquisition program 512. The document storage buffer 522 stores the document read from the magnetic disk device 6 by the document reading program 514. The morphological analysis result storage buffer 523 is for storing the result of the morphological analysis of the document stored in the document storage buffer 522 by the morphological analysis program 516. The field-related word description pattern storage buffer 524 stores the field-related word description pattern dictionary 603 read from the magnetic disk device 6 by the field-related word description pattern reading program 511. The field temporary storage buffer 525 is for the field extraction program 515 to extract field information to which the document belongs from the document stored in the document storage buffer 522 and store it. The related word counting buffer 526 is a work area used when the field related word counting program 518 counts the appearance frequency of the field related words, and is used for storing temporary variables such as variables for looping the program. is there.
[0014]
As shown in FIG. 4, the magnetic disk device 6 extracts a document database 601 composed of a plurality of documents, a morphological analysis dictionary 602 to which the morphological analysis program 516 refers when performing morphological analysis, and extracts a field-related word. Related word notation pattern dictionary 603 in which a pattern for performing the operation is stored, and a data storage for storing a program for starting the related word extraction device 1 or newly created data, including an OS (operating system). An area 604 and a related word extraction program 605 storing a program for the control device 2 to execute various controls and processes are recorded.
[0015]
FIG. 5 is a block diagram showing the configuration of various programs stored in the related word extraction program 605 in the magnetic disk device 6. The related word extraction program 605 includes a field related word notation pattern reading program 6051, a target field acquisition program 6052, a display program 6053, a document reading program 6054, a field extraction program 6055, a morphological analysis program 6056, and a field related word. It comprises an extraction program 6057, a field-related word counting program 6058, an initialization program 6059, and a field information comparison program 6060.
[0016]
The field-related word notation pattern reading program 6051 is for reading the field-related word notation pattern dictionary 603 stored in the magnetic disk device 6 into the buffer unit 52. The target field acquisition program 6052 is for acquiring field information input by a user to extract field-related words. The display program 6053 is for displaying data input by the input device 3, instructions from the related word extraction device 1 to the user, and extracted field-related words on the display unit 4. The document reading program 6054 is for reading one document from a group of documents stored in the magnetic disk device 6 into the buffer unit 52. The field extraction program 6055 is for extracting field information to which the document belongs by reading the document stored in the buffer unit 52. The morphological analysis program 6056 refers to a morphological analysis dictionary 602 stored in the magnetic disk device 6 and performs morphological analysis of a document stored in the buffer unit 52. The field-related word extraction program 6057 is for extracting field-related words from the document read into the buffer 52 with reference to the field-related word notation pattern read into the buffer 52. The field-related word counting program 6058 is for counting the frequency of appearance of the extracted field-related words. The initialization program 6059 is for initializing the setting state of each device when the power of the related word extraction device 1 is turned on. The field information comparison program 6060 is for comparing the target field acquired via the input device 3 with the field information acquired according to the field extraction program 6055.
[0017]
FIG. 6 is a diagram illustrating an example of a document group forming the document database 601 of the magnetic disk device 6. As shown in FIG. 6, each document of the document group used in the embodiment of the present invention describes the field to which the document belongs following "field:". In the embodiment of the present invention, as shown in FIG. 6, an example of a document having a structure of field information and text information corresponding to the field information is shown, but a technical field is specified according to an actually designated format. There are many documents. For example, a patent gazette is given a technical classification such as an international patent classification, FI or F-term. In addition, it is considered that in-house technical documents held by the company may be given an original technical classification in order to improve the convenience of reference at a later date. In the field information of each document in the document database 601, it is assumed that one classification code may be hierarchized such as a large classification, a middle classification, and a small classification as in the patent gazette. , The expression “field: program → subroutine definition” means “program” as “major category” and “subroutine definition” as “middle category”.
[0018]
FIG. 7 is a diagram showing an example of logical information held by the morphological analysis dictionary 602 of the magnetic disk device 6. Although the morphological analysis dictionary generally has a complicated data structure in order to improve search efficiency, here, only logical information is shown in a simplified manner. The morphological analysis dictionary 602 shown in FIG. 7 holds three types of information, ie, headings, readings, and parts of speech of words. In general morphological analysis, there are also those that use fragmented parts of speech and attributes or fine connection grammar, but the morphological analysis referred to in the embodiment of the present invention uses a simplified one, It is assumed that the part-of-speech connection information is incorporated in the morphological analysis processing. The morphological analysis used in the embodiment of the present invention can be replaced with a more general and accurate one.
[0019]
FIG. 8 is a diagram illustrating an example of the field-related word notation pattern dictionary 603. There are five field-related word notation patterns from (1) to (5), and in each of the patterns, "" is a placeholder. The placeholder is a so-called wildcard, and the control device 2 extracts a predetermined byte string (the number of digits is not particularly limited) surrounded by "" as a field-related word according to the field-related word extraction program 6057. I do.
[0020]
FIG. 9 is a diagram illustrating an example of the field-related word counting buffer 526. The field-related word totaling buffer 526 is composed of an item number, a word, and a frequency. Words extracted as field-related words by the related word totaling process described below are allocated to each item number, and are stored in the document. Stored together with the frequency of appearance.
Next, the operation of the related word extraction device 1 will be described with reference to FIGS.
FIG. 10 shows processing (specifically, target field acquisition processing, related word counting processing, and related word extraction processing) from power-on of the related word extraction device 1 to extraction of field related words and termination. It is a flowchart explained systematically. FIG. 11 is a flowchart for explaining the target field acquisition process (S3) in the flowchart shown in FIG. 10, and FIG. 12 is a flowchart for explaining the related word counting process (S5) in the flowchart shown in FIG. FIG. 13 is a flowchart illustrating an operation of finally extracting as a field-related word from a target word that is a candidate of the field-related word totaled by the related word totaling process illustrated in FIG. 12.
[0021]
When the operation of the related word extraction device 1 starts, the control device 2 reads each program from the related word extraction program 605 and stores it in the program unit 51 of the memory 5 as appropriate, and then executes a predetermined process according to the program. .
That is, in FIG. 10, when the power of the related word extraction device 1 is turned on, a bootstrap activation process is executed, and a program for executing the process shown in FIG. It is executed after being loaded into the unit 51. In this process, the control device 2 initializes the setting states of various devices such as the input device 3 and the display device 4 according to the initialization program 6059 (S1). Subsequently, in accordance with the field-related word notation pattern reading program 6051, the field-related word notation pattern dictionary 603 of the magnetic disk device 6 is read and stored in the field-related word notation pattern storage buffer 524 (S2). Thereafter, the control device 2 enters a target field acquisition process (S3). Unless the control unit 2 finishes the target field acquisition process (to be described in detail later) (No in S4), the control device 2 starts execution of the related word counting process (S5). When terminating the target field acquisition process (Yes in S4), the related word extraction device 1 ends the process after shutting down, for example, storing data on the memory 5 such as information on the system in the data storage area 604. .
[0022]
Next, the target field acquisition processing will be described with reference to FIG.
In FIG. 11, the control device 2 obtains a target field necessary for obtaining a field-related word via the input device 3 of the related word extraction device 1 according to the target field acquisition program 6052 (S301). Here, in accordance with the target field acquisition program 6052, the control device 2 executes a function indicating the end of the target field acquisition process from the input device 3 (for example, a process corresponding to a process in which the user has pressed the close button on the window). It is determined whether or not it has been sent (S302). Unless the processing is completed (No in S302), the control device 2 stores the acquired target field in the target field storage buffer 521 (S303). If the processing is to be ended (Yes in S302), the control device 2 stores the value of the end (for example, binary 0 which is the value representing the end of the character string to be written in the program) in the target field storage buffer 521 (S304). Return to the caller (to S4). When the control apparatus 2 stores the value indicating the end of the target field acquisition processing in the target field storage buffer 521 (S304), in FIG. 10, the control apparatus 2 determines that the end is determined in S4 (Yes in S4), Then, the process is terminated after shutting down, for example, storing data on the memory 5 such as the information in the data storage area 604. Hereinafter, a description will be given assuming that the target field acquired according to the target field acquisition program 6052 stored in the program unit 51 by the control device 2 is “printer technology”.
[0023]
Next, the related word counting process will be described with reference to FIG.
12, the control device 2 reads one document from the documents stored in the document database 601 of the magnetic disk device 6 according to the document reading program 6054, and stores it in the document storage buffer 522 (S501). Thereafter, the control device 2 determines field information to which each document belongs for the documents stored in the document storage buffer 522 according to the field extraction program 6055 and stores the field information in the field temporary storage buffer 525 (S502). As described above, all of the documents in the document database 601 handled in the embodiment of the present invention shown in FIG. 6 have the character string "field:" at the beginning, followed by the field information to which the document belongs. For example, in the case of the document N in FIG. 6, field information “printer technology” is extracted, and this field information is stored in the field temporary storage buffer 525.
[0024]
Next, the control device 2 compares the target field stored in the target field storage buffer 521 with the field information stored in the field temporary storage buffer 525 according to the field information comparison program 6060 stored in the program unit 51 (S503). . When it is determined that the target field stored in the target field storage buffer 521 is different from the field information stored in the field temporary storage buffer 525 (No in S503), the control device 2 operates in accordance with the document reading program 6054. , A document different from the document in the document database 601 read in S501 is read and stored in the document storage buffer 522, and the target field stored in the target field storage buffer 521 and the field information stored in the field temporary storage buffer 525 are read. The document in the document database 601 is repeatedly read until it is determined that "matches" (Yes in S507, to S501).
[0025]
When the control device 2 determines that the target field stored in the target field storage buffer 521 matches the field information stored in the field temporary storage buffer 525 (Yes in S503), the control apparatus 2 follows the morphological analysis program 6056. At this time, the document stored in the document storage buffer 522 is morphologically analyzed, and the result of the morphological analysis is stored in the morphological analysis result storage buffer 523 (S504). In the morphological analysis performed by the control device 2, information in the morphological analysis dictionary 602 illustrated in FIG. 7 is referred to. The morphological analysis in the embodiment of the present invention has a compound word processing function. The compound word processing function represents a process of identifying a combination of a plurality of words as a new word.
[0026]
For example, when two nouns are continuous like “noun” + “noun” = “noun”, these two words are recognized as new nouns, or “prefix” + “noun” = “noun” When the noun appears following the prefix, the control device 2 recognizes the compound word according to the morphological analysis program 6056 based on a compound word creation rule such as detecting the two words as new nouns. Processing. For example, when the document N-2 in FIG. 6 is divided into morphemes, the result is as follows ("/" represents a word break). Ink-jet (noun) / method (noun) / no (case particle) / printing (noun) / device (noun) / on (particle) (noun). Furthermore, when compound word processing is performed, a series of nouns and nouns is recognized as a new noun, and the result is as follows. Ink-jet method (noun) / no (case particle) / printing device (noun) / in (particle) (noun).
[0027]
After the morphological analysis processing, the control device 2 according to the field-related word extraction program 6057, based on the result of the morphological analysis stored in the morphological analysis result storage buffer 523, stores the field-related word notation pattern stored in the field-related word notation pattern storage buffer 524. A field-related word is extracted with reference to the word notation pattern (S505). In the above example, the control device 2 matches the field-related word notation pattern (in “”) of FIG. 8 (5), and extracts “printing device” as the field-related word according to the field-related word extraction program 6057. When the field-related words are extracted, the control device 2 stores the field-related words and the field-related words in the field-related word counting buffer 526 in the document according to the field-related word counting program 6058, as illustrated in FIG. The frequency of appearance is counted, and the counting result is stored (S506). Specifically, each time a field-related word is extracted, the appearance frequency of the field-related word is incremented.
[0028]
Next, the control device 2 refers to the documents stored in the document storage buffer 522 to determine whether all the documents stored in the document database 601 have been read (S507), When there is a document (Yes in S507), the process is continuously executed (to S501). If there is no unprocessed document that has not been read (No in S507), the process proceeds to a related word extraction process described later (S508), and returns to the call source.
[0029]
Next, the related word extraction processing will be described with reference to FIG.
It is assumed that the contents of the field-related word totaling buffer 526 are in the state shown in FIG.
First, the control device 2 calculates the total value of the frequencies in FIG. 9 in the temporary variable storage buffer 527 according to the field-related word extraction program 6057, and substitutes it for the temporarily set variable SUM (S5081). In the example of FIG. 9, the value 626 obtained by calculating 181 + 160 + 54 + 44 + 120 + 8 + 5 + 54 is assigned to the variable SUM. In the related word extraction device 1, a threshold value is set for removing a field related word having a low appearance frequency in a document as noise in order to enhance the suitability of a result extracted as a field related word. It is assumed that this threshold is set to 5% of the variable SUM, and words whose frequency in FIG. 9 is less than 5% are excluded (S5082). Here, since 5% of 626 is 31.3, “ink ribbon” and “method” whose frequency in FIG. 9 is less than 31.3 are excluded. Next, the remaining words are displayed on the display device 4 as the field-related words required by the related word extraction device 1 (S5083). Here, “printer”, “printer”, “printing device”, “printing device”, “inkjet printer”, and “thermal transfer printer” are displayed as field-related words.
[0030]
As described above, in the embodiment of the present invention, attention is paid to characteristic rhetorical expressions often found in documents classified in detail according to fields, such as patent specifications, and the same as the target field required by the user. The field-related words are extracted from the document having the field information on the basis of this characteristic rhetorical expression.
Further, in order to enhance the suitability of the extracted field-related words, the related-word extracting apparatus 1 sets a threshold for removing field-related words having a low appearance frequency in the document, and calculates that the threshold is lower than this threshold. Remove related words.
[0031]
The present invention is not limited to the above-described embodiment, and can be implemented with various modifications without departing from the spirit of the invention. For example, it is possible to add a document in the document database 601 in the magnetic disk device 6. Further, in the embodiment of the present invention, the field information of each document in the document database 601 is determined, but the field information may not be determined. In this case, by using an existing technology for extracting field information from a predetermined document, and combining this field extraction technology with the related word extraction technology of the present invention, a more versatile related word extraction technology can be obtained.
[0032]
In the above embodiment, the document in the document database 601 having the same field as the target field input to the input device 3 is morphologically analyzed to be compared with the field-related word notation pattern. Instead, other natural language analysis may be used. For example, natural language analysis such as syntactic analysis or semantic analysis may be used.
[0033]
【The invention's effect】
As described above, according to the present invention, a document having the same field information as the desired field information is selected without depending on the co-occurrence information of words appearing in the document, and a predetermined And a related word extraction device that extracts a field related word based on the rhetorical expression of
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a related word extraction device.
FIG. 2 is a diagram showing an internal configuration of a memory 5;
FIG. 3 is a block diagram showing a configuration of a buffer unit 52 of the memory 5.
FIG. 4 is a block diagram showing a configuration of a magnetic disk device 6.
FIG. 5 is a block diagram showing a configuration of a related word extraction program 605.
FIG. 6 is a diagram showing an example of a document group constituting a document database 601.
FIG. 7 is a view showing an example of logical information held by a morphological analysis dictionary 602.
FIG. 8 is a diagram showing an example of a field-related word notation pattern dictionary 603.
FIG. 9 is a diagram showing an example of a field-related word counting buffer 526.
FIG. 10 is a flowchart for systematically explaining the overall operation of the related word extraction device 1;
FIG. 11 is a flowchart illustrating target field acquisition processing.
FIG. 12 is a flowchart illustrating a related word counting process.
FIG. 13 is a flowchart illustrating a related word extraction process.
[Explanation of symbols]
1 ... Related word extraction device
2 ... control device
3 ... input device
4 Display devices
5 ... memory
6 ... magnetic disk drive
7 ... bus
51 ... Program section
52 ··· Buffer section
521: target field storage buffer
522: Document storage buffer
523: Morphological analysis result storage buffer
524... Field-related word notation pattern storage buffer
525: Field temporary storage buffer
526: Field-related word counting buffer
527 ... temporary variable storage buffer
601: Document database
602 morphological analysis dictionary
603 ・ ・ ・ Field related word notation pattern dictionary
604: Data storage area
605: Related word extraction program
6051 ・ ・ Field related word notation pattern reading program
6052 ・ ・ ・ Target field acquisition program
6053 ・ ・ ・ Display program
6054 ・ ・ ・ Document reading program
6055 ・ ・ ・ Field extraction program
6056 ・ ・ ・ Morphological analysis program
6057 ・ ・ ・ Field-related word extraction program
6058 ・ ・ ・ Field related word counting program
6059: Initialization program
6060 ・ ・ ・ Field information comparison program

Claims (9)

A document database having a plurality of documents;
Input means for inputting field information of the document;
Document extraction means for extracting from the document database a document having field information that matches the input field information,
Natural language analyzing means for analyzing the natural language of the document extracted by the document extracting means,
Field-related word notation pattern holding means for holding a field-related word notation pattern for extracting a field-related word related to the field of this document from a document in the document database,
A field-related word extraction unit that refers to the field-related word notation pattern based on the result of the natural language analysis and extracts the field-related word from a document extracted from the document database;
Field-related word counting means for counting the frequency of appearance of the extracted field-related words,
A related word extraction device comprising: 2. The related word extraction device according to claim 1, wherein the natural language analysis unit executes one of morphological analysis, syntactic analysis, and semantic analysis. 3. The related word extracting apparatus according to claim 1, wherein the field related word totaling unit removes a field related word whose appearance frequency of the totaled field related words is lower than a predetermined threshold. A related word extraction method for extracting a document in a specific field from a document database having a plurality of documents, and extracting a related word corresponding to the specific field from the extracted document.
An input step for inputting field information of the document;
A document extraction step of extracting a document having field information that matches the input field information from the database;
A field-related word notation pattern holding step of holding a field-related word notation pattern for extracting a field-related word related to the field of this document from the document in the document database,
A field-related word extraction step of referring to the field-related word notation pattern based on the result of the natural language analysis and extracting the field-related word from a document extracted from the document database;
A field-related word counting step of counting the frequency of appearance of the extracted field-related words,
A related word extraction method characterized by comprising: 5. The related word extraction method according to claim 4, wherein the natural language analysis step performs one of morphological analysis, syntactic analysis, and semantic analysis. 6. The related word extraction method according to claim 4, wherein the field related word totaling step removes a field related word in which the frequency of appearance of the totaled field related words is lower than a predetermined threshold. From a document database having a plurality of documents, to extract a document of a specific field, a related word extraction device that extracts a related word corresponding to the specific field from the extracted document,
An input function for inputting field information of the document,
A document extraction step of extracting a document having field information that matches the input field information from the database;
A field-related word notation pattern holding function for holding a field-related word notation pattern for extracting a field-related word related to the field of this document from a document in the document database,
A field-related word extraction function of referring to the field-related word notation pattern based on the result of the natural language analysis, and extracting the field-related word from a document extracted from the document database;
A field-related word counting function for counting the frequency of appearance of the extracted field-related words,
A program characterized by realizing. 8. The program according to claim 7, wherein the natural language analysis function executes one of morphological analysis, syntactic analysis, and semantic analysis. The program according to claim 7, wherein the field-related word totaling function removes a field-related word in which the frequency of appearance of the totaled field-related words is lower than a predetermined threshold.

Images