JP2000250916A - Document sorting method and device and recording medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To automatically sort the documents and also to improve the sorting accuracy. SOLUTION: The words extracted from many documents belonging to a specific field and the occurrence frequencies of these extracted words are previously registered on a specific dictionary 31. Meanwhile, the words extracted from many documents belonging to the general fields of a wide range are previously registered on a general dictionary 32. Then the words are extracted from an unknown document belonging to an unknown field and the degrees of associations of each of these extracted words with a specific field of the unknown document and also with the general fields of a wide range are decided by referring to both dictionaries 31 and 32. These two degrees of association are compared with each other to decide whether the unknown dictionary can be sorted as that of a specific field.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document classification method and apparatus for classifying a document whose field is unknown into a specific field, and a recording medium on which a program for executing the classification method is recorded.

[0002]

2. Description of the Related Art Conventionally, a process for classifying an unknown document to which a field to which it belongs into a specific field is manually performed. In other words, documents belonging to a specific field are collected, words are extracted from the collected documents, and data necessary for discriminating a specific field related to the extracted words (for example, the appearance of words frequently used in documents belonging to the specific field) Frequency, etc.)
Is prepared in advance, a word is extracted from the unknown document to be classified, the extracted word is referred to the previously prepared data, and the unknown document is classified into a specific field based on the reference result. A series of processes for determining whether or not the data can be classified to have been manually performed.

Recently, a large number of electronic documents have been distributed along with the construction of computer networks due to the spread of personal computers, word processors, and the like, as well as electronic documents such as CD-ROM dictionaries and newspaper articles. The situation is increasing. In such a situation where a large number of documents exist, it is impossible to deal with the problem by manual document classification, and development of an automatic document classification method is desired.

[0004]

In the conventional method, it is necessary to manually collect a large number of documents belonging to a specific field and classify words in advance. In addition, in order to improve the classification accuracy, it may be necessary to specify a specific word from these documents, but such manual classification often makes the classification criteria unclear. Furthermore, new words are not taken into account, and it is not possible to flexibly respond to dynamic changes in a specific field.

[0005] In response to the need for automatic document classification, several techniques have been proposed for automatically classifying documents. For example, JP-A-7-114572 discloses that
A feature vector of each word of a document belonging to a specific field is generated and prepared in advance, and a feature vector of each word included in a document to be classified is obtained with reference to the prepared feature vector. A method for generating a document feature vector from the feature vector of each word contained in the document and classifying the document based on the similarity between the feature vector of the generated document and the feature vector of a representative document belonging to the specific field is shown. Have been.

In the document classification method disclosed in Japanese Patent Application Laid-Open No. 7-114572, only the similarity with a document in a specific field is evaluated. Classification cannot be performed correctly due to the influence of high words, and when some documents that are categorized in advance and that are not related to the specific field are mixed, There is a problem that classification accuracy is low, for example, correct classification cannot be performed. Therefore, improvement of the classification accuracy in automatic document classification is desired.

[0007] The present invention has been made in view of the above circumstances, and of course, it is possible to automatically classify documents, as well as a document classification method and apparatus that can improve the classification accuracy, It is an object of the present invention to provide a recording medium on which a program for performing the classification method is recorded.

Another object of the present invention is to provide a document classification method and apparatus which can flexibly cope with a dynamic change in a specific field where a new word appears.

[0009]

According to a first aspect of the present invention, there is provided a document classification method, wherein a relevance of an unknown document whose field is unknown to a specific field is obtained, and the unknown document is assigned to the specific field based on the obtained relevance. In a document classification method for determining whether to classify, a step of extracting a plurality of words from the unknown document, and for each extracted word, the appearance of the word in a document belonging to the specific field prepared in advance Referring to the frequency and referring to the frequency of appearance of the word in a document belonging to a general field, which is prepared in advance, and referring to the result of reference to the frequency of appearance in the specific field and general field in each extracted word. Determining a degree of association.

[0010] The document classification method according to claim 2 is based on claim 1.
In the method, a word extracted from an unknown document is incorporated into the appearance frequency in the specific field or the general field based on the result of the classification after the classification is performed on the unknown document.

According to a third aspect of the present invention, there is provided a document classifying apparatus for determining the relevance of an unknown document whose field is unknown to a specific field, and determining whether or not to classify the unknown document into the specific field based on the obtained relevance. A first registration unit that registers in advance a frequency of appearance of a word extracted from a document belonging to the specific field; and a second registration unit that registers in advance the frequency of appearance of a word extracted from a document belonging to a general field. Registering means, means for extracting a plurality of words from the unknown document, and for each extracted word, referencing the frequency of occurrence of the word registered in the first registering means and the second registering means. Means for determining the degree of association in accordance with

According to a fourth aspect of the present invention, there is provided a document classification apparatus.
In the method described above, after performing the classification judgment on the unknown document, the first registration unit or the second registration unit additionally registers a word extracted from the unknown document based on the classification judgment result. It is characterized by having done.

According to a fifth aspect of the present invention, the relevance of an unknown document whose field is unknown to a specific field is determined, and it is determined whether or not the unknown document is classified into the specific field based on the obtained relevance. Program code means for causing the computer to extract a plurality of words from the unknown document in a computer-readable recording medium having a program recorded thereon, and each extracted word is prepared in advance. Program code means for causing the computer to refer to the frequency of occurrence of the word in a document belonging to the specific field while referring to the frequency of occurrence of the word in a document belonging to the specific field. And determining the degree of relevance according to the reference result of the frequency of occurrence of the extracted words in the specific field and the general field. The characterized by having a program code means for causing the computer.

According to a sixth aspect of the present invention, there is provided the recording medium according to the fifth aspect, wherein the program code means for causing the computer to extract a plurality of words from the document belonging to the specific field, and the program medium extracted from the document belonging to the specific field. Program code means for causing the computer to register the frequency of occurrence of each word, program code means for causing the computer to extract a plurality of words from documents belonging to the general field, and documents belonging to the general field And program code means for causing the computer to register the frequency of appearance of each word extracted from.

[0015] In the invention according to claims 1, 3, 5, and 6,
The information on the frequency of appearance of words extracted from a large number of documents belonging to a specific field is registered in advance as a dictionary, and the information on the frequency of appearance of words extracted from many documents in a wide range of general fields is also registered in advance as a dictionary. deep. Then, words are extracted from the unknown documents to be classified. For each extracted word, reference is made to the frequency of appearance in the specific field and general field from the dictionary, and the relevance of the unknown document to the specific field is determined according to the reference result. In the present invention, when calculating the degree of association, not only information (appearance frequency) related to a specific field but also information (appearance frequency) related to a wide range of general fields is referred to.

As described above, according to the present invention, when determining the degree of relevance of an unknown document to a specific field, documents belonging to a wide range of general fields are also referred to. The effects of words used in the field can be removed, words that are closely related to the specific field can be weighted heavily, and it is not necessary to manually specify a specific word, thereby improving classification accuracy.

Further, even if some documents which are not related to a specific field are mixed in a document which has been classified as belonging to a specific field in advance, the words of the documents irrelevant to the specific field are included. Because the relative frequency of words in a wide range of general fields is relatively low and it can be determined that the relevance to a specific field is small, most documents that have been classified in a specific field in advance can be identified. If you can assume that you belong to the field, collect documents manually beforehand,
Classification processing is not required, and it is hard to be affected by the mixing of the irrelevant document, so that the classification accuracy can be improved.

In the inventions according to claims 2 and 4, after the classification of an unknown document is determined, words extracted from the unknown document are additionally registered in the appearance frequency of the dictionary according to the classification result. Therefore, a newly appearing word is automatically added, so that it is possible to flexibly cope with a dynamic change in a target specific field.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be specifically described below with reference to the drawings showing the embodiments. FIG. 1 is a block diagram showing the configuration of the document classification device of the present invention. The document classification device according to the present invention includes, before the document classification process, a dictionary construction processing unit 1 for constructing a dictionary necessary for the document classification process;
The document classification processing unit 2 classifies an unknown document whose field is unknown into a specific field by referring to the dictionary constructed in advance.

The dictionary construction processing unit 1 includes a word extraction unit 11 for extracting words from a document belonging to a specific field or a document belonging to a wide range of general fields, and a word extracted from a document belonging to a specific field and its appearance frequency. The dictionary for a specific field 31
A dictionary registering unit 12 for a specific field registered in a specific dictionary 31 (hereinafter abbreviated as a specific dictionary 31) (hereinafter abbreviated as a specific dictionary registering unit 12)
And a general field dictionary registration unit 13 (hereinafter, referred to as a general field dictionary 32) that registers words extracted from documents belonging to the general field and their appearance frequencies in a general field dictionary 32 (hereinafter abbreviated as a general dictionary 32).
(Abbreviated as general dictionary registration unit 13).

The document classification processing unit 2 includes a word extraction unit 21 for extracting a word from an unknown document to be classified whose field is unknown,
With reference to the specific dictionary 31 and the general dictionary 32, the degree of relevance of each word of the extracted unknown document to the specific field and general field is obtained. Further, the specific field and general field of the unknown document are obtained from the obtained degree of relevance of each word. And a classification determining unit 22 that determines whether or not to classify the unknown document into a specific field based on the determined degree of relevance.

Next, the operation will be described. Before performing the unknown document classification process, the specific dictionary 31
And the general dictionary 32 are prepared and prepared in advance. FIG.
9 is a flowchart showing a procedure for creating such a specific dictionary 31 and a general dictionary 32.

A dictionary construction processing unit 1 converts a document belonging to a specific field into a dictionary.
(Step S1), and the word is extracted from the document by the word extracting unit 11 (step S2). As a method of this word extraction, a method of performing a morphological analysis on a document and extracting a word equivalent to a noun from the result of the morphological analysis can be used. For example, when the specific field is "soccer" and the document "Joint World Cup co-hosting by Japan and South Korea is expected to have a large economic effect" is input to the word extraction unit 11, "Japan-Korea", Five words (nouns) of “World Cup”, “Co-host”, “Economic effect”, and “Expectation” are extracted.

The extracted words are registered in the specific dictionary 31 as words in a specific field by the specific dictionary registration unit 12 (step S3). Specifically, in the case of a word that is not registered in the specific dictionary 31, the word is newly registered with the appearance frequency set to 1, and in the case of a word that is already registered in the specific dictionary 31, the appearance frequency is Is incremented by one.

It is determined whether the registration processing for all the documents belonging to the specific field has been completed (step S4).
If a document still remains (S4: NO), the same processing (S1 to S3) is repeated for the next document belonging to the specific field.

When the processing for all the documents belonging to the specific field is completed (S4: YES), the document belonging to the general field is input to the dictionary construction processing unit 1 (step S5).
The word extraction unit 11 extracts words from the document, similarly to a document belonging to a specific field (step S6). The extracted words are registered in the general dictionary 32 as words in a general field by the general dictionary registration unit 13 in the same manner as in the case of words extracted from a document belonging to a specific field (step S7).

It is determined whether the registration processing for all documents belonging to the general field has been completed (step S8).
If a document still remains (S8: NO), the same processing (S5 to S7) is repeated for the next document belonging to the general field. If the processing has been completed for all documents belonging to the general field (S8: YES), the entire processing is terminated.

Table 1 below is a table showing an example of the frequency of appearance of the word w in the specific dictionary 31 and the general dictionary 32 which have been registered in this way. The specific field is “soccer”, f (w) is the frequency of appearance of each word w in the specific field,
F (w) represents the frequency of appearance of each word w in the general field, d represents the amount of documents used to construct the specific dictionary 31, and D represents the amount of documents used to construct the general dictionary 32. Note that the number of documents used, the number of characters in the used documents, the number of extracted words, and the like can be used to represent the document amount.

[0029]

[Table 1]

Next, the operation of the classification process of the unknown document will be described with reference to the flowchart of FIG.

An unknown document is input to the document classification processing unit 2 (step S11), and words are extracted from the document by the word extraction unit 21 (step S12). Note that the word extraction method at this time is the same as the method in the word extraction unit 11.

For each of the extracted words, the classification determining unit 22
First, referring to the specific dictionary 31 and the general dictionary 32, the appearance frequency in the specific field and the general field is obtained (step S13). Since the frequency of appearance registered in the specific dictionary 31 and the general dictionary 32 increases in proportion to the amount of each document used to construct the dictionary, the frequency of appearance f (w), F
It does not make sense to compare (w) directly. Then, the appearance frequency f is calculated for each of the document amounts d and D used for the dictionary construction.
(W) and F (w) are normalized as the appearance frequency.
Specifically, the appearance frequency of the word w in a specific field is represented by q
(W), the appearance frequency of the word w in the general field is Q (w)
In this case, they are determined as in the following equations (1) and (2), respectively. q (w) = f (w) / d (1) Q (w) = F (w) / D (2)

Next, for each word, the degree of relevance to a specific field and the degree of relevance to a general field are calculated (step S14). Specifically, assuming that the degree of relevance of a word w to a specific field is r (w) and the degree of relevance to a general field is R (w), calculation is performed according to the following equations (3) and (4), respectively.
However, when Q (w) + q (w) = 0, r
Both (w) and R (w) are set to 0. r (w) = q (w) / {Q (w) + q (w)} (3) R (w) = Q (w) / {Q (w) + q (w)} (4)

Next, the degree of relevance with the specific field and the degree of relevance with the general field are calculated for the unknown document (step S15). Note that the degree of relevance to the specific field and the degree of relevance to the general field in the unknown document are the sum of the inter-language degrees for each word. Specifically, when the relevance of the unknown document X to a specific field is s (X) and the relevance to a general field is S (X), the following equations (5) and
It is calculated according to (6). It is assumed that n words w _i (i = 1, 2,..., N) have been extracted from the unknown document X.

[0035]

(Equation 1)

The magnitude of the calculated degree of relevance s (X) to the specific field and the degree of relevance S (X) to the general field are compared (step S16). If s (X)> S (X), then (S
16: YES), it is determined that the unknown document X can be classified into the specific field (step S17). On the other hand, if s (X) ≦ S (X) (S16: NO), it is determined that the unknown document X cannot be classified into the specific field (step S18).

A description will now be given of a document classification using a specific unknown document as an example. The specific field is “soccer”, and the information registered in the specific dictionary 31 and the general dictionary 32 is
It is shown in Table 1.

It is assumed that the unknown document Y is "Platter thinks that the relationship between Japan and Korea will be improved by co-hosting the World Cup." For the unknown document Y, five words of “World Cup”, “Co-host”, “Japan-Korea”, “Relationship improvement”, and “Platter” are extracted. Note that, among these words, “platter” is not registered in Table 1 at all, and is excluded from the following calculation.

With reference to the frequency of appearance f (w) of each word and the document amount d registered in the specific dictionary 31, the appearance frequency q (w) of each word is obtained according to the above equation (1). (World Cup) = 24/100 = 0.24, q (co-host) = 6/100 =
0.06, q (Japan and South Korea) = 5/100 = 0.05, q (relationship improvement) = 0
/100=0.00. Further, by referring to the appearance frequency F (w) and the document amount D of each word registered in the general dictionary 32 and calculating the appearance frequency Q (w) of each word according to the above equation (2), Q ( World Cup) = 3/1000 = 0.003, Q
(Co-host) = 26/1000 = 0.026, Q (Japan and South Korea) = 86/1000 =
0.086, Q (relationship improvement) = 46/1000 = 0.046.

Then, the relevance r (w) of each word to “soccer” is calculated according to the above equation (3).
(World Cup) = 0.24 / (0.003 + 0.24) = 0.99, r
(Co-host) = 0.06 / (0.026 + 0.06) = 0.70, r (Japan and South Korea) =
0.05 / (0.086 + 0.05) = 0.37, r (relationship improvement) = 0.00 /
(0.046 + 0.00) = 0.00.

Therefore, when the relevance s (Y) of unknown document Y to “soccer” is calculated according to the above equation (5),
It looks like this: s (Y) = r (World Cup) + r (co-host) + r (Japan-Korea) + r (relationship improvement) = 0.99 + 0.70 + 0.37 + 0.00 = 2.06

On the other hand, similarly, when the relevance S (Y) of the unknown document Y to the general field is obtained according to the above equations (4) and (6), the following is obtained. S (Y) = R (World Cup) + R (Co-host) + R (Japan-Korea) + R (Relationship improvement) = 0.01 + 0.30 + 0.63 + 1.00 = 1.94

Accordingly, s (Y)> S (Y), and it is determined that the unknown document Y can be classified as a specific field “soccer”.

As an example of the unknown document Z, "it is said that no improvement in relations can be expected because economic effects cannot be expected in both Japan and Korea." For this unknown document Z, four words of “Japan-Korea”, “economic effect”, “expectation”, and “relationship improvement” are extracted.

Then, the appearance frequency f (w) and document amount d of each word registered in the specific dictionary 31 and the general dictionary
32, the relevance s (Z) of the unknown document Z to "soccer" according to the expressions (1) to (6) with reference to the frequency of appearance F (w) and the document amount D of each word registered in the word 32. And the degree of relevance S (Z) to the general field is as follows. s (Z) = r (Japan and South Korea) + r (economic effect) + r (expected) + r (relationship improvement) = 0.37 + 0.16 + 0.53 + 0.00 = 1.06 S (Z) = R (Japan and South Korea) + R (economic effect) + R (expectation) + R (relationship improvement) = 0.63 + 0.84 + 0.47 + 1.00 = 2.94

Therefore, s (Z) <S (Z), and it is determined that the unknown document Z cannot be classified into the specific field “soccer”.

Next, another embodiment of the present invention will be described. FIG. 4 is a block diagram illustrating a configuration of the document classification device according to the embodiment. In FIG. 4, the same reference numerals as in FIG. 1 denote the same parts. FIG. 5 is a flowchart showing the procedure of the document classification process in this embodiment. In FIG. 5, the same processes as those in FIG. 3 are denoted by the same step numbers.

In this embodiment, after performing the same dictionary construction processing and document classification processing as in the above-described embodiment, the word information of the unknown document subjected to the classification processing is registered in the dictionary. After the classification of the unknown document is determined (S16, S1
(7, S18), if it is determined that the unknown document can be classified into a specific field, the extracted word of the unknown document is sent to the specific dictionary registration unit 12 and additionally registered in the specific dictionary 31 (step S1).
9) If it is determined that the unknown document cannot be classified into a specific field, the extracted word of the unknown document is registered in the general dictionary registration unit.
13 to the general dictionary 32 for additional registration (step S2
0).

For example, in the case of the unknown documents Y and Z described above, the extracted words of the unknown document Y determined to be classified as "soccer" are additionally registered in the specific dictionary 31, and the "soccer"
The extracted words of the unknown document Z that are determined to be unable to be classified in the general dictionary 32 are additionally registered in the general dictionary 32.

The word "platter" in the unknown document Y is a word which does not affect any of the specific field and the general field until the classification of the unknown document Y is determined. Since the word “platter” appears in the subsequent classification determination, it will contribute greatly to the classification of documents into the specific field “soccer”.

In the example described above, the specific dictionary 31, the general dictionary 32
Are the word appearance frequencies f (w) and F (w) and the document quantities d and D
Is registered, and the word appearance frequencies q (w), Q (w), and Q (w), F (w), and F (w), and the document amounts d and D are referred to. The relevance r (w) and R (w) of the word are calculated in order to finally obtain the relevance s (X) and S (X) of the unknown document. , Word appearance frequency q (w), Q
(W) or the degree of relevancy r (w), R (w) of the word is registered, and q (w),
With reference to Q (w) or r (w), R (w), s
(X) and S (X) may be obtained.

FIG. 6 is a block diagram showing the configuration of an embodiment of the recording medium of the present invention. The program exemplified here includes steps S1 to S20 shown in FIGS. 2, 3, and 5, and is recorded on a recording medium described below.

In FIG. 6, a recording medium 41 that is connected online to the computer 40 is, for example, a WWW (World Wide Web) server computer that is installed separately from a place where the computer 40 is installed. The program 41a is recorded as follows. When the program 41a read from the recording medium 41 controls the computer 40, the computer 40 executes the above-described dictionary construction processing and document classification processing.

The recording medium 42 provided inside the computer 40 uses a built-in hard disk drive or ROM, for example, and the recording medium 42 stores the program 42a as described above. When the program 42a read from the recording medium 42 controls the computer 40, the computer 40 executes the above-described dictionary construction processing and document classification processing.

The recording medium 43 used by being loaded into the disk drive 40a provided in the computer 40 is a transportable medium such as a magneto-optical disk, a CD-ROM, or a flexible disk. The program 43a is recorded as follows. When the program 43a read from the recording medium 43 controls the computer 40, the computer 40 executes the above-described dictionary construction processing and document classification processing.

[0056]

As described above, according to the present invention, not only the frequency of appearance of words in a specific field but also the frequency of appearance of words in a wide range of general fields is referred to. Of words that are used in a wide range of fields can be removed, words that are closely related to a specific field can be heavily weighted, and there is no need to manually specify a specific word, thereby improving classification accuracy. Can be. Also, in documents that have been classified into specific fields in advance,
Even if some documents that are not related to the specific field are mixed in, if it can be assumed that most of the documents classified in the specific field in advance belong to the specific field, manual This eliminates the need for document collection and classification processing, and is less susceptible to the mixing of irrelevant documents, thereby improving classification accuracy.

According to the present invention, after the classification of an unknown document is determined, words extracted from the unknown document are additionally registered in a dictionary according to the classification result.
New words can be added automatically, and it can flexibly respond to dynamic changes in the target specific field.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a document classification device according to the present invention.

FIG. 2 is a flowchart illustrating a procedure of a dictionary construction process.

FIG. 3 is a flowchart illustrating a procedure of a document classification process.

FIG. 4 is a block diagram showing another configuration of the document classification device of the present invention.

FIG. 5 is a flowchart illustrating another procedure of the document classification process.

FIG. 6 is a block diagram showing a configuration of an embodiment of a recording medium.

[Explanation of symbols]

 Reference Signs List 1 dictionary construction processing unit 2 document classification processing unit 11 word extraction unit 12 specific dictionary registration unit (dictionary registration unit for specific field) 13 general dictionary registration unit (dictionary registration unit for general field) 21 word extraction unit 22 classification judgment unit 31 Specific dictionaries (dictionaries for specific fields) 32 General dictionaries (dictionaries for general fields) 40 Computers 41, 42, 43 Recording media

 ────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Koichi Entrance 4-53, Kitahama, Chuo-ku, Osaka-shi, Osaka Sumitomo Metal Industries, Ltd. F-term (reference) 5B075 ND03 NK34 NR02 NR12 PR04 PR10 QM10

Claims (6)

[Claims] 1. A document classification method for determining a degree of relevance of an unknown document whose field is unknown to a specific field and determining whether to classify the unknown document into the specific field based on the obtained degree of relevance. Extracting a plurality of words from an unknown document, and for each extracted word, refer to the frequency of occurrence of the word in a document belonging to the specific field prepared in advance, and prepare in advance a general field A step of referring to the frequency of occurrence of the word in a document belonging to the document, and a step of calculating the degree of association in accordance with the result of referring to the frequency of appearance of the extracted word in the specific field and the general field. Classification method. 2. The method according to claim 1, wherein after performing classification judgment on the unknown document, a word extracted from the unknown document is incorporated into the appearance frequency in the specific field or the general field based on the classification judgment result. Document classification method. 3. A document classification apparatus for determining the degree of relevance of an unknown document whose field is unknown to a specific field and determining whether to classify the unknown document into the specific field based on the obtained degree of relevance. First registration means for registering in advance the frequency of appearance of a word extracted from a document belonging to a specific field; second registration means for registering in advance the frequency of appearance of a word extracted from a document belonging to a general field; Means for extracting a plurality of words, means for referring to the frequency of appearance of the words registered in the first registration means and the second registration means for each of the extracted words, and means for obtaining the degree of association in accordance with the reference result A document classification device comprising: 4. After performing a classification judgment on an unknown document, the first registration means or the second registration means additionally registers a word extracted from the unknown document in the appearance frequency based on the classification judgment result. 4. The document classification device according to claim 3, wherein the document classification is performed. 5. A program for determining a degree of relevance of an unknown document whose field is unknown to a specific field, and determining whether or not to classify the unknown document into the specific field based on the obtained degree of relevance. A computer-readable recording medium, the program code means for causing the computer to extract a plurality of words from the unknown document, and each extracted word belonging to the specific field, which is prepared in advance. A program code means for making the computer refer to the frequency of occurrence of the word in the document, and which is prepared in advance and referring to the frequency of appearance of the word in a document belonging to the general field; Causing the computer to determine the relevance according to the reference result of the frequency of appearance in the specific field and the general field. Recording medium having program code means. 6. A program code means for causing the computer to extract a plurality of words from a document belonging to the specific field, and registering an appearance frequency of each word extracted from the document belonging to the specific field. Registering program code means for causing a computer, program code means for causing the computer to extract a plurality of words from a document belonging to the general field, and appearance frequency of each word extracted from a document belonging to the general field 6. The recording medium according to claim 5, further comprising program code means for causing the computer to perform the above.