JPH08339383A - Document retrieving device and dictionary preparing device - Google Patents

Abstract

PURPOSE: To excellently generate a retrieval condition even if the independent words and accessory words of a retrieval request are numerous when a retrieval condition is automatically generated from the retrieval request of document data inputted by a natural language. CONSTITUTION: When the retrieval request of document data is inputted in a request input means 4 by a natural language, a language analysis means 5 extracts independent words and accessory words from this retrieval request. Because a condition generation means 6 converts each of accessory words into each of plural preliminarily set operators, combines each operator with the corresponding independent word and generates a retrieval condition, a data retrieval means 7 retrieves document data from a database 2 by this retrieval condition. Because even plural accessory words are converted into each of proper operators and even many independent words are properly combined by the operators, an optimum retrieval condition is generated from the retrieval request of the natural language inputted by a user.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document retrieval device for retrieving document data from a database and a dictionary production device for producing a pre-division dictionary before and after the document retrieval device.

[0002]

2. Description of the Related Art A document retrieval apparatus such as a document management system or an image management system has a database in which document data, image data, etc. are stored in advance as a plurality of document data, and desired document data can be retrieved from this database. You can search and output. Such data search is generally executed by setting search conditions that combine words and operators, but setting such search conditions is complicated, and general users cannot find appropriate search conditions. In some cases, it cannot be set.

In order to solve such a problem, Japanese Patent Laid-Open No. 6-
In the document retrieval device disclosed in Japanese Patent No. 75996, when a user inputs a retrieval request in natural language, the retrieval request is expanded into a synonymous expression and then the document data is retrieved. More specifically, when the user inputs a document data search request in natural language, the search request is morphologically analyzed.
Next, the morphologically analyzed search request is expanded into a synonymous expression, and the morphologically analyzed search request and the synonymous expression generate a search condition. Document data in the database is searched by matching with this search condition, and the searched document data is read from the database.

Japanese Patent Laid-Open No. 6-75996 discloses the following method for generating a search condition after expanding a natural language search request into synonymous expressions.

1. Decomposition of search request If the search request is a compound word of "A + B" or a phrase of "B of A" and "A to B", the synonyms are "A and B" or "A".
Call it "B".

2. Combining Search Requests When the search request is a phrase “B of A” or “B of A”, the synonymous expression is a compound word of “A + B”.

3. Deletion of auxiliary verbs and particles If the search request is “A do”, “A + auxiliary verb” or “A + particle”, the synonymous expression is “A”.

4. Deletion of inflectional endings When a search request is a term that ends with a term (verb / adjective / adjective verb), the derivation of the inflectional ending of the term is made synonymous.

[0009]

In the above document retrieval apparatus, since the retrieval request input in natural language is expanded into the synonymous expression, the document data can be retrieved in a wider range than the user's setting.

However, the above-described document search device corresponds only to the case where the search request is expanded into a synonymous expression to generate the search condition, when the search request is "A + B", "B of A", and "B of A". Therefore, if the search request consists of a large number of words, it is impossible to generate an appropriate search condition.

Further, the search request as described above is "A".
When the search condition is generated by decomposing it into “B”, only “A”, “B”, and “A and B” are generated as the search condition, so that it is not possible to widely respond to the user's search request.

[0012]

According to another aspect of the present invention, there is provided a document search device, which comprises a database in which a plurality of document data are stored in advance, and request input means for inputting a search request for document data in natural language. , Providing a language analysis means for extracting an independent word and an adjunct word by a language analysis from a search request input in natural language, and individually extracting each adjunct word to one of a plurality of preset operators A condition generating means for generating a search condition by converting and combining with a corresponding independent word is provided, and a data searching means for searching the document data from the database according to the generated search condition is provided.

According to the document search apparatus of the invention described in claim 2,
In the document search apparatus according to the first aspect of the present invention, the operators of the condition generation means are set to AND, OR, and NOT.

According to the document retrieval apparatus of the invention described in claim 3,
The document search device according to claim 1 or 2, further comprising a compound word detecting means for detecting a compound word from the independent word extracted from the search request, and dividing the detected compound word into a plurality of constituent words. A dividing means is provided.

According to another aspect of the document search apparatus of the present invention,
In the document retrieval apparatus according to the invention of claim 3, a constituent word dictionary in which constituent words of the compound word are stored in advance is provided, and the compound word dividing means refers to this constituent word dictionary to divide the compound word into a plurality of constituent words. To do.

According to the document search apparatus of the invention described in claim 5,
In the document retrieval apparatus according to the invention as set forth in claim 3, there is provided a divided character dictionary in which the character strings located at the division points of the compound word are stored in advance, and the compound word dividing means refers to the divided character dictionary to generate a plurality of compound words. It is divided into the constituent words of.

According to the document retrieval apparatus of the invention described in claim 6,
In the document search device according to the invention of claim 3, there is provided a division propriety table in which the propriety of the combination of the character types positioned before and after the division point of the compound word is preset, and the compound word division means is referred to with reference to this division propriety table. Splits a compound word into multiple constituent words.

According to the document retrieval apparatus of the invention of claim 7,
In the document search device according to the invention of claim 3, a pre-division dictionary is provided in which a character positioned immediately before a division point of a compound word and a character positioned immediately after the division point of the compound word are individually stored in advance, and the pre-division dictionary is referred to. The compound word dividing means divides the compound word into a plurality of constituent words.

According to the document search apparatus of the invention described in claim 8,
In the document search device according to the invention as set forth in claim 7, the probability of being located at the end and the probability of being located at the beginning of the word are set for each of the characters stored in the pre-division dictionary and the compound word is referred to with reference to these probabilities. The dividing means divides the compound word into a plurality of constituent words.

According to the document retrieval apparatus of the invention described in claim 9,
In the document retrieval apparatus according to the seventh aspect of the present invention, the compound word dividing unit divides only a compound word consisting of a preset specific character type into a plurality of constituent words.

According to a tenth aspect of the document search apparatus of the present invention, in the document retrieval apparatus of the first, second, third, fourth, fifth, sixth, seventh, eighth or ninth aspect, a combination of words that are synonyms Is provided with a pre-stored synonym dictionary, provided with a synonym detection means for detecting the word stored in this synonym dictionary from the independent word extracted from the search request, and the detected independent word to another synonym A synonym expanding means for expanding is provided.

A dictionary creating apparatus according to the invention of claim 11 is
A word acquisition unit that acquires a word from document data, a character classification unit that classifies the characters that make up the acquired word, an occurrence frequency that appears in a word for each of the classified characters, and a position at the end of the word Frequency calculating means for calculating the tail frequency and the head frequency located at the head, and probability calculation for calculating the tail probability that is the ratio of the tail frequency to the appearance frequency and the head probability that is the ratio of the head frequency for each character. Means and
A character setting means for setting a character having a high end probability as a character positioned immediately before the division point of the compound word in the pre-division dictionary and setting a character having a high start probability as a character positioned immediately after the division point of the compound word, Have.

According to a twelfth aspect of the dictionary creating apparatus of the present invention, in the dictionary creating apparatus of the eleventh aspect of the present invention, a character determining unit for determining a calculated character having a low appearance frequency is provided to detect a character having a low appearance frequency. Probability setting means for replacing the tail probability and the head probability with the mean value of the tail probabilities and the mean value of the head probabilities of all characters are provided.

[0024]

In the document searching apparatus according to the first aspect of the present invention, when a search request for document data is input to the request input means in natural language, the language analysis means uses the search request input in natural language to perform independent language analysis by language analysis. And the adjunct. The condition generating means individually converts each of the extracted auxiliary words into one of a plurality of preset operators and combines them with the corresponding independent word to generate a search condition. The data retrieval means retrieves the document data from the database according to the condition. Even if there are multiple adjunct words, these are converted to appropriate operators individually, and even if there are many independent words, these are combined appropriately by the operators so that the optimum search conditions are generated from the natural language search request entered by the user. It

According to the document search apparatus of the invention described in claim 2,
Since each of the plurality of attached words extracted from the search request is individually converted into one of AND, OR, and AND NOT, the search request desired by the user can be appropriately reflected in the search condition.

According to the document retrieval apparatus of the invention described in claim 3,
When the compound word detecting means detects a compound word from the independent word extracted from the search request, the compound word dividing means divides the detected compound word into a plurality of constituent words. Is a compound word that is not suitable for data retrieval,
This is divided into optimum constituent words for data retrieval.

According to the document retrieval apparatus of the invention described in claim 4,
When the compound word dividing unit divides the compound word into a plurality of constituent words, since the constituent words of the compound word preset in the constituent word dictionary are referred to, the compound word is divided into the expected constituent words.

According to the document search apparatus of the invention described in claim 5,
When the compound word dividing unit divides the compound word into a plurality of constituent words, it refers to the character string located at the dividing point of the compound word preset in the divided character dictionary. Is divided by.

According to the document search apparatus of the invention described in claim 6,
When the compound word dividing unit divides the compound word into a plurality of constituent words, the compound word is not suitable because it refers to the possibility of the combination of the character types located before and after the dividing point of the compound word set in advance in the dividing possibility table. It is not divided at the position of the combination of character types.

According to the document search apparatus of the invention described in claim 7,
When the compound word dividing unit divides the compound word into a plurality of constituent words, it refers to the character positioned immediately before the compound word division point and the character positioned immediately after the compound word preset individually in the pre-division dictionary. The compound word is split according to the characters before and after the expected split point.

According to the document search apparatus of the invention described in claim 8,
When the compound word dividing unit divides the compound word into a plurality of constituent words, it refers to the probability of being located at the end and the probability of being located at the beginning of the word set for each of the characters stored in the dictionary before and after the division. The compound word is divided according to the probabilities of the characters before and after the division point.

According to the document search apparatus of the invention described in claim 9,
Since the compound word dividing means divides only a compound word consisting of a preset specific character type into a plurality of constituent words, useless processing of word division is performed for a compound word such as a katakana word whose division point cannot be determined by the preceding and following characters. Can be prevented.

In the document retrieval apparatus according to the present invention, when the synonym detecting means detects a word stored in the synonym dictionary from the independent word extracted from the search request, the detected independent word is synonymous. Since the expanding means expands to another synonym, even if the independent word of the search request input by the user is a word unsuitable for data search, if a synonym exists in this,
This synonym is also used for data search.

In the dictionary creating apparatus according to the eleventh aspect of the present invention, when the word acquiring means acquires a word from the document data,
The character classification means classifies the characters forming the acquired word. For each of the separated characters, the frequency calculation means calculates the frequency of appearance in the word, the end frequency of the word at the end, and the head frequency of the word at the beginning. The probability calculation means calculates a tail probability that is the ratio of the tail frequency to the appearance frequency and a head probability that is the ratio of the head frequency. The character setting means sets a character having a high end probability to a character positioned immediately before the division point of the compound word,
Since a character with a high start probability is set as a character located immediately after the division point of the compound word in the before-and-after-dictionary dictionary, the character before and after the division point of the compound word is located in this before-and-after-dictionary dictionary. Characters and are properly stored.

In the dictionary creating apparatus according to the twelfth aspect of the present invention, when the character determining means determines the calculated character having a low appearance frequency, the tail probability and the head probability of the character having a low appearance frequency are calculated for all the characters. Since the probability setting means replaces the average value of the tail probabilities with the average value of the head probabilities, the probability with low accuracy is not set in the pre-division dictionary.

[0036]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A first embodiment of the document retrieval apparatus of the present invention is shown in FIG.
1 to 3 will be described below. First, the document search device 1 of the present embodiment has a document database 2 and a data processing unit 3, as shown in FIG. For example, the document database 2 may be HD (Hard Disk) or MO (Magneto).
to optical disk), and the data processing section 3 is composed of a personal computer or a workstation. Document database 2
A large number of document data are stored in advance, and this document data is composed of a text file expressed in natural language.

The data processing section 3 has a request input section 4 which is a request input means. The request input section 4 includes:
A language analysis unit 5, which is a language analysis means, is connected. The language analysis unit 5 includes a condition generation unit 6 which is a condition generation unit.
Is connected, and the condition generation unit 6 is connected to a search processing unit 7 which is a data search means. This search processing unit 7 is connected to the document database 2,
A result output unit 8 which is result output means is connected.

The request input unit 4 has, for example, a data input device such as a keyboard capable of inputting and operating character data, and inputs a search request for document data in natural language. The language analysis unit 5, as will be described later in detail,
Morphological analysis and syntactic analysis, which are linguistic analyses, detect an independent word and an adjunct word from a natural language search request, and analyze the dependency relation of a plurality of clauses formed by these words. Then, by connecting the clauses in accordance with this dependency relationship, as shown in FIG. 2, a syntactic analysis tree is generated from the natural language search request.

The condition generating section 6 has three "and, or, and not" preset as operators, and each of the extracted adjuncts is set as one of the three preset operators. The search conditions are generated as search condition trees by individually converting and combining them with corresponding independent words as shown in FIG. This condition generation unit 6 has an operator correspondence table (not shown), and in this operator correspondence table, "or (parallel word) → or""(case particle) → and" The correspondence between the natural language adjunct and the three operators is preset. Note that "A and knot B" means having "A" and not having "B".

The retrieval processing unit 7 retrieves document data from the document database 2 according to the produced retrieval condition, and reads the retrieved document data from the document database 2. The result output unit 8 is, for example,
It has a data output device such as a display or a printer, and outputs document data by displaying or printing.

In the document retrieval apparatus 1 of this embodiment having such a configuration, when a retrieval request for document data is input in natural language, retrieval conditions are automatically generated and document data is retrieved from the document database 2. . Therefore, this processing operation will be sequentially described below.

First, when the user considers a document data retrieval request and inputs it in the request input section 4 in natural language,
The language analysis unit 5 extracts an independent word and an adjunct word from the search request by morphological analysis and syntactic analysis, and analyzes the dependency relation of the clause of the search request. More specifically, if a search request is input in natural language as a “light emitting type image display device capable of adjusting lightness or color tone”, then “lightness, color tone, adjustment, light emitting type image display device” as independent words are sequentially displayed. “Or, can” is extracted in order as an accessory word.

By analyzing the dependency relations of a plurality of clauses composed of these words, a syntactic analysis tree is generated from the natural language search request as shown in FIG. In this parse tree, each clause is represented by a terminal node, and the words that make up each clause have a rectangular framework. The target of a certain clause is the end node (hereinafter called the rightmost child node) that is reached by following the child node on the right side of the figure from the closest parent node. If this is the self node, the parent node is Go back in sequence.

More specifically, the first clause "lightness or" is the leftmost child node of the parse tree, and its constituent words are "lightness""or". “Brightness or”
Since the rightmost child node of the parent node is not itself, the destination of the above is the phrase “color tone” of the rightmost child node of the parent node. In addition, since the rightmost child node of the parent node is itself, the destination of this phrase “color tone” is the phrase “adjustable” of the rightmost child node of the parent node that has gone back one step.

Next, the condition generator 6 individually converts each of the extracted adjunct words into one of the three operators "and, or, and knot" and combines it with the corresponding independent word. By doing so, the search condition is generated. More specifically, the word selecting means executes the word selecting process, and the independent word "brightness, color tone, adjustment, light emitting image display device" is selected from the constituent words of the search request. Next, the operator selecting means refers to the operator correspondence table and executes the operator selecting process, and each of the adjuncts "or, can," is converted to the operator "or, and, and". Converted individually.

Since the operators thus converted are combined with the corresponding independent words, as shown in FIG. 3, a search condition tree in which an end node consisting of an independent word is connected to a parent node consisting of an adjunct word is searched. It is generated as a condition. When this search condition tree is expressed in text format, it becomes (((brightness or color tone) and control) and light emitting image display device), which can be handled in the same way as existing search conditions. Therefore,
The search processing unit 7 searches and reads the document data from the document database 2 according to the generated search condition, and the result output unit 8 outputs the document data by displaying or printing.

In the document retrieval apparatus 1 of this embodiment, when the user inputs a retrieval request in natural language as described above, appropriate retrieval conditions are automatically generated and document data is retrieved, so that the user retrieves. There is no need to set conditions. Moreover, even if a large number of independent words and adjunct words are present in the search request entered in natural language, each of the adjunct words is converted into an operator and combined with the corresponding independent word, so that appropriate search conditions are required. The generated document data is searched well. Furthermore, since three operators "and, or, and knot" are set as operators for generating the search condition, it is possible to meet a wide range of user search requests.

Next, a second embodiment of the document retrieval apparatus of the present invention will be described below with reference to FIG. With regard to the second embodiment, the same parts as those in the first embodiment described above are designated by the same names and reference numerals, and detailed description thereof will be omitted.

First, in the document search device 1 of the present embodiment, the condition generating section 6 which is a condition generating means has a synonym dictionary, a synonym detecting means, and a synonym expanding means. The synonym dictionary previously stores a combination of words that are synonyms such as “brightness, brightness”, and the synonym detection means requests the words stored in the synonym dictionary as a search request. Detect from independent words. Since the synonym expanding means expands the detected independent word into another synonym, as shown in the figure,
The condition generation unit 6 can generate a search condition using the expanded synonyms.

With such a configuration, the document search apparatus 1 of this embodiment also automatically generates search conditions when a document data search request is input in natural language, as in the first embodiment. To search the document data.

When the condition generating unit 6 generates the search condition from the search request, the synonym detection means detects a word stored in the synonym dictionary from the independent word of the search request, and the detected independent word is a synonym. Expanded to another synonym by the expansion means. For example, when "lightness" is present in the independent word of the search request, "brightness" is detected as a synonym, and these synonyms are connected by "OR". In this case, document data in which "brightness" is not described as a term and "brightness" is described is also searched, so that the document data can be searched well and a wide range of user search requests can be met. it can.

Next, a third embodiment of the document retrieval apparatus of the present invention will be described below with reference to FIG. With regard to the third embodiment, the same parts as those of the first embodiment described above are designated by the same names and reference numerals, and detailed description thereof will be omitted.

First, in the document search device 1 of the present embodiment, the condition generating section 6 which is a condition generating means has a character type discriminating means, a compound word detecting means, and a compound word dividing means. The character type discriminating means determines the character type of the independent word extracted from the search request (Kanji, Katakana, Hiragana, number, alphabet,
...) is determined, and the compound word detecting means detects an independent word having a changed character type as a compound word. Since the compound word dividing unit divides the detected compound word into a plurality of constituent words with the change point of the character type as a dividing point, the condition generation unit 6 causes the condition generating unit 6 to divide the plurality of divided constituent words as illustrated. The search condition can be generated by.

With such a configuration, in the document search device 1 of this embodiment, when the condition generator 6 generates a search condition from a search request, the character type of the independent word of the search request is determined by the character type determination means. An independent word of which the character type is changed is detected as a compound word by the compound word detecting means, and then divided into a plurality of constituent words by the compound word dividing means.

For example, when "light emitting type image display device" exists in the independent word of the search request, this means that the character type is "Kanji /
Since it has been changed to katakana / kanji, this is divided into compound words of "light emitting type", "image", and "display device" as compound words, and they are connected by the operator of "and". Not only the document data in which the "type image display device" is described as one term, but also the document data in which the three terms "light emitting type", "image", and "display device" are described separately are searched. Document data can be searched for, and a wide range of user search requests can be met.

In this embodiment, the constituent word is "and".
Although the connection with the operator of is illustrated, it is also possible to connect with the operator of “or”.

Further, in the present embodiment, an example is shown in which a compound word which is one independent word is divided into a plurality of constituent words which are independent words, but it is also possible to execute the expansion of synonyms at the latter stage. Is. In this case, as shown in FIG. 6, after the "light emitting type image display device" is divided into the constituent words of "light emitting type""image""displaydevice" as a compound word,
Since "image" is also expanded to the synonym "image",
Document data can be searched better.

In the above-described document retrieval device 1, since the independent word is identified as the compound word by the character type discrimination and divided into the constituent words, the processing operation is simple and the structure can be simplified. However, the present invention is not limited to the above embodiment, and the method of dividing the compound word is not limited to the discrimination of the character type.

For example, a constituent word dictionary in which constituent words of a compound word are stored in advance is provided, and the compound word dividing means of the condition generation unit 6 refers to the constituent word dictionary to divide the compound word into a plurality of constituent words. Is also possible. In this case, if the constituent word dictionary is properly set, the “display device” that is a compound word with one character type can be divided into the constituent words of “display” and “device”, so that the document is better documented. You can search the data.

When detecting a constituent word from an independent word, it can be assumed that the constituent words overlap. That is,
From the independent word "Tokyo", the constituent words "Tokyo" and "Kyoto" are detected, but these are partially duplicated. In such a case, it is possible to expand the search range of the document data using both as constituent words, but it is desirable to stop the division of the independent word to prevent the search of inappropriate document data. Also, when adopting both of the constituent words that are partially duplicated, divide the compound words by each constituent word and connect them with the "and" operator, and connect them with the "or" operator. Is also possible. For example, the independent word is "Tokyo"
In case of, it becomes ((Tokyo and Miyako) or (East and Kyoto)). In this case, the search range of the document data can be expanded while minimizing the search for inappropriate document data.

Further, a divided character dictionary in which the character strings located at the division points of the compound word are stored in advance is provided, and the compound word dividing means of the condition generation unit 6 refers to the divided character dictionary to form a plurality of compound words. It is also possible to divide it into words. For example, if the character string "indication" is set in the divided character dictionary,
Since the “display device” which is a compound word of one character type is divided into the constituent words of “display” and “device”, the document data can be searched more favorably.

Further, as shown in Table 1 below, a division permission / prohibition table in which permission / prohibition of the combination of the character types located before and after the division point of the compound word is preset is provided, and the compound word division means of the condition generation unit 6 is provided. It is also possible to divide the compound word into a plurality of constituent words by referring to the division possibility table.

[0063]

[Table 1]

In this division permission / prohibition table, "O" is set for the combination of character types that can be divided, and "X" is set for the combination of character types that cannot be divided. For example, division of a compound word is performed at the "Hiragana / Kanji" position, but not at the "Kanji / Hiragana" position.

More specifically, when a search request is input by a natural language as a "microprocessor system having an interrupt signal input / output terminal", as shown in FIG.
"Interrupt signal input / output terminal, microprocessor system" is extracted as an independent word, and "is present" as an accessory word
Is extracted. Next, as shown in FIG. 8, "NO-ARU" extracted as this adjunct word is converted into an "AND" operator, and the "interrupt signal input / output terminal" extracted as an independent word has a character type Since it is changed to "Kanji / Hiragana / Kanji", it is detected as a compound word.

Therefore, this compound word "interrupt signal input / output terminal" is divided into a plurality of constituent words based on the change of the character type.
It is divided at the position that changes to “Kanji” but not at the position that changes to “Kanji / Hiragana.” Therefore, the “interrupt signal input / output terminal” of “Kanji / Hiragana / Kanji” is
It is not divided into "interrupt", "mi", and "signal input / output terminal", but is divided into "interrupt" and "signal input / output terminal" of "Hiragana / Kanji" as shown in FIG.

That is, since the compound word dividing means refers to the above-mentioned division permission / inhibition table by the compound word dividing means, it is possible to prevent improper division such that the kana is divided from the kanji, so that the compound word is divided into a plurality of constituent words. The dividing process can be appropriately performed. Even if the division of a compound word is controlled in this way, a "signal input / output terminal" that is a compound word with one character type
Cannot be divided into the constituent words of "signal", "input / output", and "terminal", so it is preferable to use the constituent word dictionary described above.

However, in such a constituent word dictionary, it is inevitably necessary to appropriately set a huge number of words, and it is expected that the creation thereof will be difficult. Therefore, in order to solve such a problem, as shown in Table 2 below, a pre-division dictionary is provided in which the character immediately before the division point of the compound word and the character immediately after the division point are individually stored in advance. The compound word dividing unit of the condition generation unit 6 can also divide the compound word into a plurality of constituent words by referring to the pre-division dictionary.

[0069]

[Table 2]

That is, since the compound word is divided into a plurality of constituent words, the character at the end of the constituent word is located immediately before this dividing point, and the character at the beginning of the constituent word is located immediately after the dividing point. . Since it is possible to predict a character that is likely to be located at the beginning or end of a word, if this character is stored in the pre / post-dictionary dictionary, the compound word can be properly divided into constituent words.

More specifically, as shown in Table 2 above, "input, output, end, ..." Is set as the character immediately before the division point, and "target, character, number, force" is set as the character immediately after. ... "is set, the above-mentioned" signal input / output terminal "is" No.
"Signal""Input / output" divided at the input "force / end" position
It becomes a "terminal". That is, by referring to the above-mentioned pre-division dictionary and the compound word dividing means, it is possible to favorably divide a compound word having one character type into a plurality of constituent words without setting a huge number of words.

However, also in such a pre-division dictionary, it is necessary to appropriately predict and set the leading and trailing characters of the constituent words, and its creation requires specialized ability. In order to solve such a problem, it is desirable that the pre-division dictionary is automatically created by the dictionary creating device.

An embodiment of the dictionary creating apparatus of the present invention will be described below with reference to FIG. First, the dictionary creating device 11 of the present embodiment stores characters online or offline in the pre-division dictionary 12 composed of a storage device such as an HD. Therefore, the dictionary creating apparatus 11 of the present embodiment has a document input unit 13 as a document input unit, a word acquisition unit 14 as a word acquisition unit, a character detection unit 15 as a character classification unit, and a frequency calculation unit as a frequency calculation unit. It has a unit 16, a probability calculating unit 17 which is a probability calculating unit, and a character setting unit 18 which is a character setting unit, and the character setting unit 18 is connected to the pre-division dictionary 12.

The document input unit 13 is composed of, for example, an electronic file system, and inputs document data of a text file expressed in natural language. The word acquisition unit 14 acquires a word from document data by language analysis such as morphological analysis or syntactic analysis, and the character detection unit 15
Separates the letters that make up the retrieved word. The frequency calculation unit 16 calculates an appearance frequency that appears in a word, a tail frequency that is located at the end of the word, and a head frequency that is located at the beginning of each of the classified characters, and the probability calculation unit 17 A tail probability, which is the ratio of the tail frequency to the appearance frequency, and a head probability, which is the ratio of the head frequency to the appearance frequency, are calculated for each of the separated characters. The character setting unit 18 sets a character having a high ending probability and a character having a high starting probability in the pre-division dictionary 12 as a character positioned immediately before and a character positioned immediately after a division point of a compound word.

With such a configuration, the dictionary creating apparatus 11 of the present embodiment automatically detects the characters located at the beginning or end of a word when appropriate document data is input as a sample. The characters detected in this way are set in the before-and-after-division dictionary 12 as the characters located before and after the division point of the compound word, so that the before-and-after-division dictionary 12 of the document search device 1 can be automatically created.

More specifically, when the document input section 13 inputs the document data of a text file expressed in natural language, the word acquisition section 14 causes the morphological analysis, the syntactic analysis, etc. as shown in Table 3 below. A word is acquired from the document data by the language analysis of.

[0077]

[Table 3]

Then, the character detection unit 15 classifies the characters forming the acquired word, so that the frequency calculation unit 16
For each of the separated characters, the appearance frequency of the word, the tail frequency of the word, and the head frequency of the word are calculated. For example, when attention is paid to "out" as a character, as shown in Table 3 above, the total number of appearance frequencies is "12". The end frequency, which is the total number of "outs" located at the end of the word, is "2", and the end frequency, which is the total number of "outs" located at the beginning of the word, is "8".

Therefore, as shown in Table 4 below, the probability calculation unit 17 separates the tail probability, which is the ratio of the tail frequency to the appearance frequency, and the head probability, which is the ratio of the head frequency to the appearance frequency. Calculate for each character. For example, paying attention to the "out" mentioned above, the appearance frequency is "12".
Since the tail frequency is "2", the tail probability is "2/12 = 0.17".
And the leading probability is “8/12 = 0.67”.

[0080]

[Table 4]

The character setting unit 18 regards a character having a high end probability and a character having a high head probability as a character positioned immediately before and a character positioned immediately after the division point of the compound word.
Set to 2 respectively. For example, the threshold for this setting is "0.50"
Then, the character having a higher end probability than "0.50" is set in the pre-division dictionary 12 as the character positioned immediately before the compound word, and the character having a higher start probability than "0.50" is the character positioned immediately after the compound word. It is set in the pre-division dictionary 12.
In this case, as shown in Table 4 above, the "incoming, incoming, outgoing,
"End, ..." is set as the character positioned immediately after the division point, and "go, power, child, ..." is set as the character positioned immediately before the compound word.

That is, in the dictionary creating apparatus 11 of the present embodiment, the characters before and after the division point of the compound word are automatically detected from the document data input as a sample, and the dictionary before and after the division 1
Since it can be set to 2, it is not necessary for the user to perform complicated work requiring specialized ability. In particular,
If the user inputs document data having the same content as the document data used for the search by the document search device 1 to the dictionary creation device 11 as a sample, the pre / post-division dictionary 12 can be created in an optimum form for the user's work. it can.

The dictionary creating device 11 is used as described above.
When the pre-division dictionary 12 is created by the above, the more the document data input as a sample, the more accurate the setting contents of the pre-division dictionary 12, but this increases the work load and time. In this respect, if the amount of document data is small, the work load and time are reduced, but in this case, the accuracy of the pre-division dictionary 12 cannot be expected.

If this is a problem, then
It is desirable that the dictionary creation device 11 be provided with a character determination unit that is a character determination unit and a probability setting unit that is a probability setting unit. In this case, the character determination unit determines a character with a low appearance frequency, and the probability setting unit determines the tail probability and the head probability of the character with a low appearance frequency as the average value of the tail probabilities of all the characters and the average of the head probabilities. Replace with and respectively.

That is, for a character having a low appearance frequency, the accuracy is not sufficient even if the head probability or the tail probability is calculated from the ratio of the tail frequency or the head frequency to the appearance frequency. Therefore, in such a case, the probability of the end or the beginning of the character is replaced with the average value of the probabilities of the end or the beginning of all the characters to prevent the setting of the inappropriate probability.

More specifically, in the case of Tables 3 and 4 described above, since the character "child" appears only in three of the "terminals", its leading probability is "0.00" and its trailing probability is " 1.00 ”
Is. However, this is a coincidence that the sample document data is related to electronic devices and the frequency of appearance of "child" is low, and the word in which "child" is located at the beginning is "child,
Since there is a child machine, child node, etc., its leading probability is "0.
Setting 00 ”is not appropriate.

In this case, the appearance frequency of all characters is "4.
Since the ending frequency is “0”, the ending frequency is “14”, and the starting frequency is “14”, the average value of the ending probabilities of all characters is “14/40 = 0.35”, and the average value of the starting probabilities is also “0.35”. As described above, by replacing the probability of a character having a low appearance probability with the average value as described above, it is possible to prevent the setting of an unsuitable probability and improve the accuracy of the pre-division dictionary 12.

As described above, by setting the characters located immediately before and after the dividing point of the compound word in the before-and-after-dictionary dictionary 12, the compound word dividing means of the document retrieval apparatus 1 makes the compound word into a plurality of constituent words. However, the end probability and the start probability calculated by the dictionary creation device 11 as described above are set in the dictionary 12 before and after the division together with the characters, and the compound word division means of the document search device 1 uses these. It is also possible for the compound word dividing means to divide the compound word into a plurality of constituent words by referring to the probability.

More specifically, the pre-division dictionary 12 includes
As shown in Table 4 above, since the leading probability and the ending probability are set for each character, the document retrieval apparatus 1 multiplies the probabilities of the characters before and after the compound word division point to obtain the division evaluation value. The division evaluation value is compared with a threshold value to determine whether or not division is possible. For example, when dividing a compound word “signal input / output terminal”, the divided evaluation value is calculated as follows, so the signal / number → 0.00 × 0.00 = 0.00 number / input → 0.88 × 0.70 = 0.61 input / output → 0.30 × 0.67 = 0.20 output / force → 0.17 × 0.06 = 0.01 force / end → 0.50 × 0.75 = 0.38 end / child → 0.25 × 0.00 = 0.00 These divisions When the evaluation value is compared with the threshold value "0.30", the positions of "go / enter" and "force / end" are divided points, which are divided into "signal", "input / output" and "terminal".

In the various methods of dividing the compound word, attention is paid to the characters before and after the compound word. For example, in the case of phonetic hiragana or katakana, two characters are noted. Even the dividing point is difficult to determine. In other words, in the case of a compound word composed of hiragana or katakana, the calculation process for determining the division point is useless.

That is, as described above, the document retrieval device 1
When dividing a compound word into constituent words by paying attention to the characters before and after the dividing point, the compound word dividing means should divide only the compound word consisting of a specific character type preset as Kanji into multiple constituent words. This makes it possible to omit the division of the compound word consisting of hiragana or katakana, and reduce the processing load and time.

Here, various methods are exemplified as the method of dividing the compound word into the constituent words. For example, in the method of discriminating the character type, it is not necessary to create a dictionary in advance and it is not necessary to refer to the dictionary, but a compound word with one character type cannot be divided into constituent words. On the other hand, in the method of referring to the dictionary, a compound word having one character type can be divided into constituent words, but it is necessary to create the dictionary in advance and also refer to the dictionary. In the method of referring to the dictionary, the higher the word division processing accuracy is, the more the data capacity and the processing load can be reduced. That is, the various methods described above are preferably selected according to various conditions such as the user's request and the specifications of the device.

[0093]

According to the document retrieval apparatus of the invention described in claim 1,
Providing language analysis means for extracting independent words and adjuncts from a search request entered in natural language by language analysis, and converting each of the extracted adjuncts individually into one of a plurality of preset operators By providing the condition generation means that generates search conditions by combining with the corresponding independent word, even if multiple adjunct words are individually converted into appropriate operators, even if multiple independent words are combined appropriately by the operator As a result, appropriate search conditions can be generated from the natural language search request input by the user, and document data can be searched well.

According to another aspect of the document search apparatus of the present invention, the operators of the condition generating means are AND, OR, AND NOT,
By setting as, it is possible to appropriately reflect the search request desired by the user in the search condition, and it is possible to widely respond to the search request of the user.

The document retrieval apparatus according to the third aspect of the present invention is provided with a compound word detecting means for detecting a compound word from the independent word extracted from the search request, and divides the detected compound word into a plurality of constituent words. By providing the word dividing means, even if the independent word of the search request entered by the user is a compound word that is not suitable for data search, it is automatically divided into constituent words suitable for data search, so that the document data can be satisfactorily improved. You can search.

According to another aspect of the document retrieval apparatus of the present invention, a constituent word dictionary in which constituent words of a compound word are stored in advance is provided.
By referring to this constituent word dictionary and dividing the compound word into a plurality of constituent words by the compound word dividing means, even if the independent word of the search request input by the user is not suitable for data retrieval,
Since this is automatically divided into constituent words suitable for data retrieval, the document data can be retrieved well.

According to the document retrieval apparatus of the fifth aspect of the present invention, there is provided a divided character dictionary in which the character strings positioned at the division points of the compound word are stored in advance, and the compound word dividing means refers to this divided character dictionary to combine the compound word dividing means. By splitting a word into multiple constituent words,
Even if the independent word of the search request input by the user is a compound word that is not suitable for data search, it is automatically divided into constituent words suitable for data search, so that document data can be searched well.

According to the document retrieval apparatus of the present invention, a division permission / prohibition table in which the permission / prohibition of the combination of the character types located before and after the division point of the compound word is set in advance is provided. By dividing the compound word into a plurality of constituent words by the word dividing means, for example, if a combination of kanji / hiragana is set to be impossible, it is possible to prevent an improper division in which a kana is divided from kanji. Therefore, it is possible to appropriately perform the process of dividing the compound word into a plurality of constituent words.

According to the document retrieval apparatus of the invention as defined in claim 7, a pre-division dictionary is provided in which the character immediately before the division point of the compound word and the character immediately after the division point of the compound word are individually stored in advance. If the compound word dividing means divides the compound word into a plurality of constituent words by referring to, and predicts characters that are likely to be located at the beginning or end of the constituent word and stores them in the dictionary before and after the division, the compound word can be well composed. Since the words can be divided and it is not necessary to predict and store a huge number of constituent words, the capacity of the dictionary can be reduced.

According to the document retrieval apparatus of the present invention, the probability of being located at the end and the probability of being located at the beginning of a word are set for each character stored in the pre-division dictionary and the probabilities are referred to. The compound word dividing means divides the compound word into a plurality of constituent words, and if the character is predicted and stored in the dictionary before and after the division together with the probability of being located at the beginning or end of the constituent word, it is divided by these probabilities. Since the compound word can be divided into the constituent words by evaluating the accuracy of the points, the compound word can be divided better.

In the document retrieval apparatus according to the ninth aspect of the present invention, the compound word dividing means divides only a compound word consisting of a preset specific character type into a plurality of constituent words. If it is set to be divided, it is possible to reduce the processing load and time by omitting the processing that is difficult to determine the division point even when focusing on two characters, such as a compound word consisting of katakana. .

The document retrieval apparatus according to the invention of claim 10 is:
A synonym dictionary in which combinations of words that are synonyms are stored in advance is provided, and a synonym detection unit that detects the word stored in this synonym dictionary from the independent word extracted from the search request is provided. By providing the synonym expansion means for expanding a word into another synonym, even if the independent word of the search request input by the user is not suitable for data search, if a synonym exists in this word, this synonym Since the word is also used for the data search, the document data can be searched well.

The dictionary creating apparatus according to the invention of claim 11
A word acquisition unit that acquires a word from document data, a character classification unit that classifies the characters that make up the acquired word, an occurrence frequency that appears in a word for each of the classified characters, and a position at the end of the word Frequency calculating means for calculating the tail frequency and the head frequency located at the head, and probability calculation for calculating the tail probability that is the ratio of the tail frequency to the appearance frequency and the head probability that is the ratio of the head frequency for each character. Means and
A character setting means for setting a character having a high end probability as a character positioned immediately before the division point of the compound word in the pre-division dictionary and setting a character having a high start probability as a character positioned immediately after the division point of the compound word, With this, since the characters located before and after the division point of the compound word can be automatically detected from the document data and set in the pre-division dictionary, the user does not need to perform such complicated work.

According to the twelfth aspect of the present invention, there is provided a dictionary creating device,
A character determining means for determining a calculated character having a low appearance frequency is provided, and the tail probability and the head probability of a character having a low appearance frequency are respectively replaced with the average value of the tail probabilities of all the characters and the average value of the head probabilities. By providing the probability setting means, the characters positioned before and after the division point of the compound word can be set together with the probability in the before-and-after-dictionary dictionary, so that the document retrieval device can appropriately divide the compound word by these probabilities. it can.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of a document search device of the present invention.

FIG. 2 is a schematic diagram showing a search request for a natural language subjected to language analysis.

FIG. 3 is a schematic diagram showing search conditions generated from a search request.

FIG. 4 is a schematic diagram showing a search condition in which a synonym is added to one of the independent words in the second embodiment of the document search device.

FIG. 5 is a schematic diagram showing search conditions obtained by dividing one compound word, which is an independent word, into three constituent words in the third embodiment of the document search device.

FIG. 6 is a schematic diagram showing a search condition in which a compound word is divided and a synonym is added in a modified example of the document search device.

FIG. 7 is a schematic diagram showing a search request for a natural language subjected to language analysis in another modification of the document search device.

FIG. 8 is a schematic diagram showing search conditions generated from a search request.

FIG. 9 is a schematic diagram showing search conditions in which one compound word is divided into two constituent words.

FIG. 10 is a block diagram showing an embodiment of a dictionary creating device of the present invention.

[Explanation of symbols]

 1 Document Retrieval Device 2 Database 4 Request Input Means 5 Language Analysis Means 6 Condition Generation Means 7 Data Retrieval Means 11 Dictionary Creation Equipment 14 Word Acquisition Means 15 Character Sorting Means 16 Frequency Calculating Means 17 Probability Calculating Means 18 Character Setting Means

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] July 25, 1995

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 5

[Correction method] Change

[Correction content]

[Procedure Amendment 2]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 7

[Correction method] Change

[Correction content]

[Procedure 3]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 8

[Correction method] Change

[Correction content]

[Procedure amendment 4]

[Document name to be amended] Statement

[Name of item to be corrected] Claim 11

[Correction method] Change

[Correction content]

[Procedure Amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0001

[Correction method] Change

[Correction content]

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a document retrieval device for retrieving document data from a database, and a dictionary production device for producing a pre / post-division character dictionary used in this document retrieval device.

[Procedure correction 6]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction content]

According to the document search apparatus of the invention described in claim 5,
In the document search apparatus of the invention of claim 3, wherein, provided the divided character dictionary string that can become the division point of the compound word is stored in advance, a compound word dividing means with reference to the divided character dictionary compound words Divide into multiple constituent words.

[Procedure Amendment 7]

[Document name to be amended] Statement

[Correction target item name] 0018

[Correction method] Change

[Correction content]

According to the document retrieval apparatus of the invention of claim 7,
In the document search apparatus of the invention of claim 3, wherein, the character located immediately after the character located immediately before the division point of the compound word previously stored divided before and after character dictionary individually provided, this amount
The compound word dividing means divides the compound word into a plurality of constituent words by referring to the character dictionary before and after the division.

[Procedure Amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0019

[Correction method] Change

[Correction content]

According to the document search apparatus of the invention described in claim 8,
In the document search device according to the invention of claim 7, before and after division.
The probability of being located at the end and the probability of being located at the beginning of the word are set for each of the characters stored in the character dictionary , and the compound word dividing means divides the compound word into a plurality of constituent words with reference to these probabilities. .

[Procedure Amendment 9]

[Document name to be amended] Statement

[Name of item to be corrected] 0022

[Correction method] Change

[Correction content]

A dictionary creating apparatus according to the invention of claim 11 is
A word acquisition unit that acquires a word from document data, a character classification unit that classifies the characters that make up the acquired word, an occurrence frequency that appears in a word for each of the classified characters, and a position at the end of the word Frequency calculating means for calculating the tail frequency and the head frequency located at the head, and probability calculation for calculating the tail probability that is the ratio of the tail frequency to the appearance frequency and the head probability that is the ratio of the head frequency for each character. Means and
When set to divide front and rear character dictionary as a character located the head probable characters immediately after the dividing point of the compound word and sets the character position just prior to the dividing point of the compound word trailing probable character breaks judgment And means .

[Procedure Amendment 10]

[Document name to be amended] Statement

[Correction target item name] 0028

[Correction method] Change

[Correction content]

According to the document search apparatus of the invention described in claim 5,
When the compound word dividing means divides the compound word into a plurality of constituent words, the division point of the preset compound word split character dictionary
Since reference to string obtaining Ri, compound words are divided at the position of the expected string.

[Procedure Amendment 11]

[Document name to be amended] Statement

[Name of item to be corrected] 0030

[Correction method] Change

[Correction content]

According to the document search apparatus of the invention described in claim 7,
When the compound word dividing unit divides the compound word into a plurality of constituent words, it refers to the character positioned immediately before the compound word division point and the character positioned immediately after the compound word preset point in the character dictionary before and after the division. , The compound word is split according to the characters before and after the expected split point.

[Procedure Amendment 12]

[Document name to be amended] Statement

[Correction target item name] 0031

[Correction method] Change

[Correction content]

According to the document search apparatus of the invention described in claim 8,
When the compound word dividing unit divides the compound word into a plurality of constituent words, it refers to the probability of being located at the end and the probability of being located at the beginning of the word set for each of the characters stored in the character dictionary before and after the division . , The compound word is divided according to the probabilities of the characters before and after the division point.

[Procedure Amendment 13]

[Document name to be amended] Statement

[Correction target item name] 0034

[Correction method] Change

[Correction content]

In the dictionary creating apparatus according to the eleventh aspect of the present invention, when the word acquiring means acquires a word from the document data,
The character classification means classifies the characters forming the acquired word. For each of the separated characters, the frequency calculation means calculates the frequency of appearance in the word, the end frequency of the word at the end, and the head frequency of the word at the beginning. The probability calculation means calculates a tail probability that is the ratio of the tail frequency to the appearance frequency and a head probability that is the ratio of the head frequency . The dividing property determining means sets a character having a high end probability as a character positioned immediately before the division point of the compound word and a character having a high start probability as a character positioned immediately after the division point of the compound word in the character dictionary before and after the division. Therefore, the character dictionary immediately before and the character immediately after the division point of the compound word are appropriately stored in the character dictionary before and after the division.

[Procedure Amendment 14]

[Document name to be amended] Statement

[Correction target item name] 0035

[Correction method] Change

[Correction content]

In the dictionary creating apparatus according to the twelfth aspect of the present invention, when the character determining means determines the calculated character having a low appearance frequency, the tail probability and the head probability of the character having a low appearance frequency are calculated for all the characters. Since the probability setting means replaces the average value of the tail probability and the average value of the head probability respectively, before and after the division.
The probability of low accuracy is not set in the character dictionary .

[Procedure Amendment 15]

[Document name to be amended] Statement

[Correction target item name] 0061

[Correction method] Change

[Correction content]

[0061] Further, provided the divided character dictionary string that can become the division point of the compound word is stored in advance, a compound word dividing means condition generating unit 6, a compound word several Referring to split character dictionary It is also possible to divide it into constituent words. For example, if the character string "indication" is set in the divided character dictionary,
Since the “display device” which is a compound word of one character type is divided into the constituent words of “display” and “device”, the document data can be searched more favorably.

[Procedure Amendment 16]

[Document name to be amended] Statement

[Correction target item name] 0068

[Correction method] Change

[Correction content]

However, in such a constituent word dictionary, it is inevitably necessary to appropriately set a huge number of words, and it is expected that the creation thereof will be difficult. Therefore, in order to solve such a problem, as shown in Table 2 below, a pre / post-division character dictionary in which a character immediately before a division point of a compound word and a character immediately after the division point are individually stored in advance is provided. Provided, condition generation unit 6
The compound word dividing means can also divide the compound word into a plurality of constituent words by referring to the character dictionary before and after the division .

[Procedure Amendment 17]

[Document name to be amended] Statement

[Name of item to be corrected] 0070

[Correction method] Change

[Correction content]

That is, since the compound word is divided into a plurality of constituent words, the character at the end of the constituent word is located immediately before this dividing point, and the character at the beginning of the constituent word is located immediately after the dividing point. . Since it is possible to predict a character that is likely to be located at the beginning or end of a word, storing this in the before- and- after character dictionary allows the compound word to be properly divided into constituent words.

[Procedure 18]

[Document name to be amended] Statement

[Correction target item name] 0071

[Correction method] Change

[Correction content]

More specifically, as shown in Table 2 above, "input, output, end, ..." Is set as the character immediately before the division point, and "target, character, number, force" is set as the character immediately after. ... "is set, the above-mentioned" signal input / output terminal "is" No.
"Signal""Input / output" divided at the input "force / end" position
It becomes a "terminal". In other words, character letters before and after splitting as described above
By referring to the book by the compound word dividing means, it is possible to favorably divide the compound word having one character type into a plurality of constituent words without setting a huge number of words.

[Procedure Amendment 19]

[Document name to be amended] Statement

[Name of item to be corrected] 0072

[Correction method] Change

[Correction content]

However, such a character dictionary before and after division also
It is necessary to properly predict and set the letters at the beginning and the end of the constituent words, and to create them requires specialized ability.
In order to solve such a problem, it is desirable that the character dictionary before and after division is automatically created by a dictionary creating device.

[Procedure amendment 20]

[Document name to be amended] Statement

[Correction target item name] 0073

[Correction method] Change

[Correction content]

An embodiment of the dictionary creating apparatus of the present invention will be described below with reference to FIG. First, the dictionary creating apparatus 11 of the present embodiment stores characters online or offline in the before-and-after character dictionary 12 composed of a storage device such as an HD. Therefore, the dictionary creating apparatus 11 of the present embodiment has a document input unit 13 as a document input unit, a word acquisition unit 14 as a word acquisition unit, a character detection unit 15 as a character classification unit, and a frequency calculation unit as a frequency calculation unit. The character setting unit 18 includes a unit 16, a probability calculating unit 17 that is a probability calculating unit , and a character setting unit 18 that is a dividing property determining unit.
Is connected to the before / after character dictionary 12.

[Procedure correction 21]

[Document name to be amended] Statement

[Correction target item name] 0074

[Correction method] Change

[Correction content]

The document input unit 13 is composed of, for example, an electronic file system, and inputs document data of a text file expressed in natural language. The word acquisition unit 14 acquires a word from document data by language analysis such as morphological analysis or syntactic analysis, and the character detection unit 15
Separates the letters that make up the retrieved word. The frequency calculation unit 16 calculates an appearance frequency that appears in a word, a tail frequency that is located at the end of the word, and a head frequency that is located at the beginning of each of the classified characters, and the probability calculation unit 17 A tail probability, which is the ratio of the tail frequency to the appearance frequency, and a head probability, which is the ratio of the head frequency to the appearance frequency, are calculated for each of the separated characters. The character setting unit 18 sets a character having a high end probability and a character having a high start probability in the before-and-after- division character dictionary 12 as a character positioned immediately before and a character positioned immediately after a division point of a compound word. .

[Procedure correction 22]

[Document name to be amended] Statement

[Correction target item name] 0075

[Correction method] Change

[Correction content]

With such a configuration, the dictionary creating apparatus 11 of the present embodiment automatically detects the characters located at the beginning or end of a word when appropriate document data is input as a sample. The character detected in this manner is used as a character positioned before and after the division point of the compound word, and is used as the character dictionary before and after division 12
Is set to, the pre-division character dictionary 12 of the document search device 1 can be automatically created.

[Procedure amendment 23]

[Document name to be amended] Statement

[Correction target item name] 0081

[Correction method] Change

[Correction content]

The character setting unit 18 regards a character having a high end probability and a character having a high start probability as a character positioned immediately before and a character positioned immediately after a division point of a compound word, and is used as a character before and after division.
Each is set in Book 12. For example, the threshold for this setting is "0.
If it is set to 50 ”, the character with a higher end probability than“ 0.50 ”is divided into the character before and after the compound word.
The character set to 2 and having a higher start probability than "0.50" is set in the divided before-and-after character dictionary 12 as a character located immediately after the compound word. In this case, as shown in Table 4 above,
"Bin, In, Out, End, ..." is set as the character positioned immediately after the division point, and "Go, Power, Child, ..." is set as the character positioned immediately before the compound word.

[Procedure correction 24]

[Document name to be amended] Statement

[Correction target item name] 0082

[Correction method] Change

[Correction content]

That is, the dictionary creating apparatus 11 of the present embodiment automatically detects the characters before and after the dividing point of the compound word from the document data input as a sample, and the character words before and after the dividing.
Since it can be set in the document 12, the user does not need to perform the complicated work that requires specialized ability. In particular, if the user inputs the document data having the same content as the document data used for the search by the document search device 1 to the dictionary creation device 11 as a sample, the pre / post-division character dictionary 12 is created in the optimum form for the user's work. can do.

[Procedure correction 25]

[Document name to be amended] Statement

[Name of item to be corrected] 0083

[Correction method] Change

[Correction content]

The dictionary creating device 11 is used as described above.
Optionally creating a split longitudinal character dictionary 12, as document data to be input as a sample is a large amount, divided before and after
Although the setting contents of the character dictionary 12 are accurate, this increases the work burden and time. In this regard, the burden and time of work if the document data is a small amount is decreased, this case is the minute
The accuracy of the before-and-after character dictionary 12 cannot be expected.

[Procedure Amendment 26]

[Document name to be amended] Statement

[Correction target item name] 0087

[Correction method] Change

[Correction content]

In this case, the appearance frequency of all characters is "4.
Since "0", end frequency is "14", and start frequency is "14", the average value of the end probabilities of all characters is "14/40 = 0.35", and the average value of the start probabilities is also "0.35". By replacing the probability of a character having a low appearance probability with the average value as described above, it is possible to prevent the setting of an inappropriate probability and improve the accuracy of the character dictionary 12 before and after division. .

[Procedure Amendment 27]

[Document name to be amended] Statement

[Correction target item name] 0088

[Correction method] Change

[Correction content]

As described above, by setting the characters located immediately before and after the division point of the compound word in the before-and-after character dictionary 12, the compound word dividing means of the document retrieval apparatus 1 forms a plurality of compound words. Can be well divided into words,
As described above, the trailing probability and the leading probability calculated by the dictionary creating device 11 are set in the character dictionary 12 before and after the division together with the characters, and the compound word dividing means of the document search device 1 refers to these probabilities and the compound word dividing means. It is also possible to divide a compound word into a plurality of constituent words.

[Procedure correction 28]

[Document name to be amended] Statement

[Correction target item name] 0089

[Correction method] Change

[Correction content]

More specifically, since the leading probability and the ending probability are set for each character in the before-and-after character dictionary 12 as shown in Table 4, the document retrieval apparatus 1 uses the compound word. The probability of characters before and after the division point is multiplied to obtain a division evaluation value, and this division evaluation value is compared with a threshold to determine whether or not division is possible. For example, when dividing a compound word “signal input / output terminal”, the divided evaluation value is calculated as follows, so the signal / number → 0.00 × 0.00 = 0.00 number / input → 0.88 × 0.70 = 0.61 input / output → 0.30 × 0.67 = 0.20 output / force → 0.17 × 0.06 = 0.01 force / end → 0.50 × 0.75 = 0.38 end / child → 0.25 × 0.00 = 0.00 These divisions When the evaluation value is compared with the threshold value "0.30", the positions of "go / enter" and "force / end" are divided points, which are divided into "signal", "input / output" and "terminal".

[Procedure correction 29]

[Document name to be amended] Statement

[Correction target item name] 0097

[Correction method] Change

[Correction content]

[0097] document search apparatus of the invention of claim 5 wherein is provided with a divided character dictionary string that can become the division point of the compound word is stored in advance, a compound word dividing means with reference to the divided character dictionary By splitting a compound word into multiple constituent words,
Even if the independent word of the search request input by the user is a compound word that is not suitable for data search, it is automatically divided into constituent words suitable for data search, so that document data can be searched well.

[Procedure amendment 30]

[Document name to be amended] Statement

[Correction target item name] 0099

[Correction method] Change

[Correction content]

According to a seventh aspect of the present invention, there is provided a document search device, which comprises a pre- division character dictionary in which a character positioned immediately before a division point of a compound word and a character positioned immediately after the division point are individually stored in advance.
By dividing the compound word dividing means with reference to the divided characters before and after dictionary compound words into a plurality of constituent words, the front and rear divided in anticipation of likely character located at the beginning or end of the configuration word character
If stored in the dictionary , the compound word can be properly divided, and it is not necessary to predict and store a huge number of constituent words, so that the capacity of the dictionary can be reduced.

[Procedure correction 31]

[Document name to be amended] Statement

[Correction target item name] 0100

[Correction method] Change

[Correction content]

[0100] document search apparatus of the present invention according to claim 8, the partial
The probability of being located at the end and the probability of being located at the beginning of the word are set for each of the characters stored in the character dictionary before and after splitting, and the compound word dividing means refers to these probabilities to divide the compound word into a plurality of constituent words. By dividing and predicting the character with the probability of being located at the beginning or end of the constituent word and storing it in the character dictionary before and after the division , the probability of the division point is evaluated by these probabilities and the compound word becomes the constituent word. Since it can be divided, the compound word can be divided better.

[Procedure correction 32]

[Document name to be amended] Statement

[Correction target item name] 0103

[Correction method] Change

[Correction content]

The dictionary creating apparatus according to the invention of claim 11
A word acquisition unit that acquires a word from document data, a character classification unit that classifies the characters that make up the acquired word, an occurrence frequency that appears in a word for each of the classified characters, and a position at the end of the word Frequency calculating means for calculating the tail frequency and the head frequency located at the head, and probability calculation for calculating the tail probability that is the ratio of the tail frequency to the appearance frequency and the head probability that is the ratio of the head frequency for each character. Means and
When set to divide front and rear character dictionary as a character located the head probable characters immediately after the dividing point of the compound word and sets the character position just prior to the dividing point of the compound word trailing probable character breaks judgment By including means , it is possible to automatically detect the characters located before and after the division point of the compound word from the document data and set them in the character dictionary before and after division, so that the user needs to perform such complicated work. Absent.

[Procedure amendment 33]

[Document name to be amended] Statement

[Correction target item name] 0104

[Correction method] Change

[Correction content]

According to the twelfth aspect of the present invention, there is provided a dictionary creating device,
A character determining means for determining a calculated character having a low appearance frequency is provided, and the tail probability and the head probability of a character having a low appearance frequency are respectively replaced with the average value of the tail probabilities of all the characters and the average value of the head probabilities. by providing the probability setting unit, a character located before and after the dividing point of the compound word, split along with the probability
Since it can be set in the preceding and following character dictionary , the document retrieval device can appropriately divide the compound word by these probabilities.

[Procedure amendment 34]

[Document name to be amended] Statement

[Name of item to be corrected] Brief description of the drawing

[Correction method] Change

[Correction content]

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment of a document search device of the present invention.

FIG. 2 is a schematic diagram showing a search request for a natural language subjected to language analysis.

FIG. 3 is a schematic diagram showing search conditions generated from a search request.

FIG. 4 is a schematic diagram showing a search condition in which a synonym is added to one of the independent words in the second embodiment of the document search device.

FIG. 5 is a schematic diagram showing search conditions obtained by dividing one compound word, which is an independent word, into three constituent words in the third embodiment of the document search device.

FIG. 6 is a schematic diagram showing a search condition in which a compound word is divided and a synonym is added in a modified example of the document search device.

FIG. 7 is a schematic diagram showing a search request for a natural language subjected to language analysis in another modification of the document search device.

FIG. 8 is a schematic diagram showing search conditions generated from a search request.

FIG. 9 is a schematic diagram showing search conditions in which one compound word is divided into two constituent words.

FIG. 10 is a block diagram showing an embodiment of a dictionary creating device of the present invention.

[Explanation of Codes] 1 Document Retrieval Device 2 Database 4 Request Input Means 5 Language Analysis Means 6 Condition Generation Means 7 Data Retrieval Means 11 Dictionary Creation Device 12 Pre / Post Character Dictionary 14 Word Acquisition Means 15 Character Classification Means 16 Frequency Calculation Means 17 Probability Calculation means 18 Dividability determination means

[Procedure amendment 35]

[Document name to be corrected] Drawing

[Name of item to be corrected] Fig. 10

[Correction method] Change

[Correction content]

[Figure 10]

Claims (12)

[Claims] 1. A database in which a plurality of pieces of document data are stored in advance, a request input means for inputting a search request for document data in natural language, an independent word and an adjunct word by language analysis from the search request input in natural language. And a condition analyzing means for individually converting each of the extracted auxiliary words into one of a plurality of preset operators and combining them with the corresponding independent word to generate a search condition. And a data search means for searching document data from the database according to the generated search condition. 2. The document search device according to claim 1, wherein the operator of the condition generation means is set to AND, OR, and AND NOT. 3. A compound word detecting means for detecting a compound word from an independent word extracted from a search request, and a compound word dividing means for dividing the detected compound word into a plurality of constituent words. The document search device according to claim 1 or 2. 4. A constituent word dictionary in which constituent words of a compound word are stored in advance, and the compound word dividing means divides the compound word into a plurality of constituent words with reference to this constituent word dictionary. Item 3. The document search device according to item 3. 5. A divided character dictionary in which a character string located at a division point of a compound word is stored in advance, and the compound word dividing means divides the compound word into a plurality of constituent words with reference to this divided character dictionary. The document search device according to claim 3, wherein 6. A division permission / prohibition table in which permission / prohibition of a combination of character types located before and after the division point of the compound word is preset is provided, and the compound word division means refers to the division permission / prohibition table to form a plurality of compound words. The document retrieval device according to claim 3, wherein the document retrieval device is divided into words. 7. A before-and-after-division dictionary in which a character immediately before a dividing point of a compound word and a character immediately after the dividing point of the compound word are individually stored in advance, and the compound-word dividing means refers to the compound-before-and-after dictionary to combine the compound words. 4. The document retrieval device according to claim 3, wherein the word is divided into a plurality of constituent words. 8. The probability of being located at the end and the probability of being located at the beginning of a word are set for each of the characters stored in the pre / post-division dictionary, and the compound word dividing means refers to these probabilities to generate a plurality of compound words. 8. The method is characterized by dividing into the constituent words of
Document retrieval device described. 9. The document retrieval device according to claim 7, wherein the compound word dividing means divides only a compound word consisting of a preset specific character type into a plurality of constituent words. 10. A synonym dictionary in which a combination of words that are synonyms is stored in advance is provided, and a synonym detection means for detecting a word stored in this synonym dictionary from an independent word extracted from a search request is provided. 10. The document retrieval device according to claim 1, further comprising synonym expansion means for expanding the detected independent word into another synonym. . 11. A word acquisition unit for acquiring a word from document data, a character classification unit for classifying characters constituting the acquired word, an appearance frequency and a word that appear in a word for each of the classified characters. A frequency calculating means for calculating a tail frequency located at the end and a head frequency located at the beginning, a tail probability that is the ratio of the tail frequency to the appearance frequency for each character, and a head probability that is the ratio of the head frequency; And a character having a high end probability is set in the dictionary before and after the division as a character positioned immediately before the division point of the compound word, and a character having a high head probability is positioned immediately after the division point of the compound word. And a character setting unit for setting the dictionary as a dictionary. 12. A character determination means for determining a calculated character having a low appearance frequency is provided, and an average value of the end probabilities and an average value of the start probabilities of the end probabilities and the head probabilities of the characters having a low appearance frequency are set. 12. The dictionary creating apparatus according to claim 11, further comprising: probability setting means for replacing with and.