JP3123836B2 - Text type database device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text type database device for searching for information including a search target sentence described in a natural language or a word string, and more particularly to a search called a role in a semantic attribute of a word. The present invention relates to a text-type database device based on a semantic attribute with a role, which enables attribute information to be defined, thereby enabling a flexible response to a text field and enabling a high-precision search.

[0002]

2. Description of the Related Art In a conventional text database apparatus for performing a search in consideration of the semantic similarity between an input sentence and a registered text (a sentence to be searched), Japanese Patent Application Laid-Open No. 2-287876 is disclosed.
As shown in the "text-type database device" of Japanese Patent Application No. Hei 11-111626, word information such as word notation and semantic attributes (referred to as semantic categories in the invention of Japanese Patent Application Laid-Open No. 2-287876). A specified word dictionary, a morphological analysis unit that divides the input sentence into words, and extracts the semantic attributes of each word, and a search target sentence that includes the semantic attributes extracted by the word notation and morphological analysis unit The registration information of the word dictionary, which is composed of a similarity calculation unit that calculates the degree and defines the semantic attributes of each word, directly affected the search results.

[0003]

Therefore, in order to improve the search accuracy, it is necessary to prepare a word dictionary corresponding to the text to be registered, and a lot of labor is required for the construction of the word dictionary.

For example, although not shown, the following 2
Consider two text databases DB1 and DB2. DB1 is a text-based database of "minutes" in which search target sentences a1 "sales of mobile phones" and search target sentences a2 "development of mobile phones" are registered. DB2 is a search target sentence b1 " It is assumed that it is a text-type database of "product introduction articles" in which the company A has sold a new mobile phone "and the search target sentence b2" mobile phone of Company B "are registered.

[0005] DB1: "minutes" text-type database Search target sentence a1 = "sales of mobile phone" Search target sentence a2 = "development of mobile phone" ... DB2: "Product introduction" Text-based database of "Article" Search target sentence b1 = "Company A sold a new mobile phone" Search target sentence b2 = "Company B mobile phone" ... In DB1, a search instruction sentence It is desirable that the search target sentence a1 be searched in a higher order in “telephone sales”. DB
In No. 2, it is desirable that the search target sentence b2 be searched in a higher order using the search instruction sentence "telephone sold by Company B".

[0006] In order to realize the above search with a conventional device,
In DB1, the words "sales" and "sales" are given a semantic attribute [sales]. In DB2, the words "sales" are either excluded from the search target by providing a special flag, or an unnecessary word dictionary is provided. It is necessary to register the word "sales" and exclude the word registered in the unnecessary word dictionary from the search. As described above, when it is necessary to set a flag for each word and register it in an unnecessary word dictionary, and to target texts in different fields,
These tasks require a lot of effort.

Further, the word dictionary of the conventional device is a dictionary specialized for searching in the field of the text to be registered, and cannot be shared with other natural language processing word dictionaries.

The present invention solves the above-mentioned conventional disadvantages.
For each semantic attribute, information (hereinafter, referred to as a role) indicating which of the semantic attribute and the word notation of the word to which the semantic attribute is assigned is used as a search key or not is used as a search key. By providing a semantic attribute / role relation table specified for semantic attributes and performing searches based on the rules of roles assigned to semantic attributes, individual texts in different fields can be searched. Without changing the word dictionary for each word
An object of the present invention is to provide a text type database device which can be adjusted so that a search suitable for a target text can be performed with a small effort for changing a semantic attribute / role relation table.

[0009]

As shown in FIG. 1, a text type database device based on semantic attributes with rolls according to the present invention comprises a word dictionary 1 which defines word expressions, semantic attributes, and parts of speech for each word. And a text-type database storage unit 6 for storing a search target sentence and its related information;
For each semantic attribute, a semantic attribute / role relation table 3 defining a role indicating which of a semantic attribute and a word notation of the word to which the semantic attribute is assigned is used as a search key or not. A morphological analysis unit 2 that divides a search instruction sentence into words with reference to a word dictionary 1 and extracts word expressions, semantic attributes and parts of speech, and a morphological analysis unit 2
The role determination unit 4 extracts the role of the semantic attribute extracted from the semantic attribute / role relation table 3 and the search key based on the role of each semantic attribute extracted by the role determination unit 4 and the part of speech of the word. A search key determining unit 5 for determining semantic attributes and word notations, a sentence searching unit 7 for searching a search target sentence stored in the text type database storage unit 6 with the search key determined by the search key determining unit 5; A search key determining unit for calculating a similarity between the search target sentence searched by the sentence search unit and the search instruction sentence, based on the semantic attribute and the word notation used as the search key; .

Further, in the present invention, the similarity calculating section 8
Is a semantic attribute and word notation that did not become a search key in the search key determination unit 5, but if the role extracted by the role determination unit 4 is a specific role, A word notation having the semantic attribute,
Alternatively, the similarity is calculated by using both the semantic attribute having the role and the word notation having the semantic attribute in the similarity calculation.

[0011]

In the text-based database device based on the semantic attribute with roll according to the present invention, when a search instruction is input, the morphological analysis unit 2 divides the word into words while referring to the word dictionary 1, and displays the notation of each word. The semantic attribute and part of speech are extracted, the role determining unit 4 refers to the semantic attribute / role relation table 3, obtains the role of the semantic attribute extracted by the morphological analysis unit 2, and determines the search key based on the definition of the role. In part 5,
A word notation and a semantic attribute to be used as a search key are determined, and a search target sentence is searched from the text database storage unit 6 using the search key determined by the sentence search unit 7. Therefore, by changing the role of the semantic attribute in the semantic attribute / role relation table 3, it is possible to easily construct a text database for performing a search suitable for the field of the text to be searched.

[0012]

FIG. 1 is a block diagram of an embodiment of the present invention, in which 1 is a word dictionary, 2 is a morphological analysis unit, and 3 is a semantic attribute.
A role relation table, 4 is a role determination unit, 5 is a search key determination unit, 6 is a text type database storage unit, 7 is a sentence search unit, 8 is a similarity calculation unit, and 10 is a processing device including a CPU and a memory. .

In the embodiment of FIG. 1, FIG. 2 shows an example of execution when a search instruction sentence "telephone sold by company B" is input in a text-type database of "product introduction articles".
FIG. 3 shows an example of a rule definition, and FIG. 4 shows an example of execution when a search instruction sentence "telephone sales" is input in a text database of "minutes".

As shown in FIG. 5, the word dictionary 1 describes word notations, internal notations, semantic attributes, parts of speech, and the like.
In the internal notation, the same character string is described for words having the same meaning, and in the case of a word having the internal notation, the internal notation is treated in the same manner as the word notation in a search process described later. For example, “computer” and “computer” are given the same internal notation “computer”. In addition, in the declinable word, it is absorbed by this internal notation. For example, "Pain" and "Pain" are given the internal notation "Pain".

A plurality of semantic attributes can be assigned to one word. Each semantic attribute represents a concept, and the set of the plurality of semantic attributes can approximate the meaning of the word. For example, a semantic attribute [head] and a semantic attribute [pain] are assigned to the word “headache”. Further, a superordinate concept of the meaning of the word may be given as a semantic attribute. For example, a semantic attribute [disease] is assigned to the word “headache”.

The morphological analysis unit 2 refers to the word dictionary 1 and divides an input instruction sentence into words, and extracts word expressions, semantic attributes, parts of speech, and the like of each word. In the example of FIG. 2, the word notations “Company B”, “GA”, “Sale”, “Done”, and “Telephone” are divided. For the word notation “Company B”, “Sale”, and “Telephone”, , Semantic attributes [company], [sales], [telephone] respectively
However, “proper noun”, “noun”, and “noun” are extracted as parts of speech. The semantic attribute is not extracted for the particle “ga” and the semantic verb “wa”, so no semantic attribute is extracted. Here, the case where one word is assigned one meaning attribute is shown. However, when a plurality of meaning attributes are given in the word dictionary 1, a plurality of meaning attributes are extracted.

The relationship between the semantic attribute and the role is described in the semantic attribute / role relation table 3. Here, for example, five types of rolls as shown in FIG. 3 are defined. The processing method for calculating the search key and the similarity differs depending on each role. By changing the role assigned to the semantic attribute in the semantic attribute / role relation table 3, it can be changed to perform the search according to the field. It is also assumed that a default role is defined for semantic attributes not specified in this table. Here, the default role is a role that uses a semantic attribute having the role and a word notation of a word having the semantic attribute as a search key.

The role determining unit 4 extracts the role of the semantic attribute extracted by the morphological analysis unit 2 from the semantic attribute / role relation table 3. In the example of FIG. 2, for each of the semantic attributes [company], [sales], and [telephone], the role r
1, r0, r2 are obtained.

The search key determining unit 5 determines the semantic attribute and word notation as search keys based on the role of each semantic attribute extracted by the role determining unit 4 and the word class of the word. For example, with regard to part of speech, it is determined that adjuncts such as particles and variances such as "do" are not used as search keys. Rolls are based on the rules in FIG. For example, in this example, the search key is determined as follows.

(1) Since the role of the semantic attribute [company] is r1, this semantic attribute is not used as a search key, and the word notation "B
Company "as a search key. (2) Since the part of speech of "ga" is a particle, it is not used as a search key.

(3) Since the semantic attribute [sales] has the role r0, both the word notation “sales” and the semantic attribute [sales]
Not a search key. (4) The word "was" is not used as a search key because it is a verb.

(5) Since the semantic attribute [telephone] has the role r2, the word notation "telephone" is not used as a search key, and the semantic attribute [telephone] is used as a search key. As described above, the word notation "Company B" and the semantic attribute [telephone] are the search keys.

As shown in FIG. 2, the text-type database storage unit 6 includes information including a search target sentence of a sentence 21 “Company A sold a new telephone” and a sentence 22 “portable telephone of Company B”. Has been accumulated. These search target sentences may be analyzed at the time of search, or may be those analyzed in advance as follows. That is, various sentence to be searched are analyzed using the word dictionary 1 by the same processing as the morphological analysis unit 2, and an index table that can be searched using the word notation and the semantic attribute as a key is generated in advance. By accumulating the index table in the text-type database storage unit 6, the search can be speeded up. In the description below, it is assumed that the search target sentence has already been analyzed.

The sentence search unit 7 searches the search target sentence stored in the text type database storage unit 6 using the search key determined by the search key determination unit 5. The sentence 22 containing the word “telephone” and the sentence 22 containing the word “telephone” with the semantic attribute [telephone] are searched for in the word notation “company B”.

The similarity calculation unit 8 calculates the similarity between the search target sentence searched by the sentence search unit 7 and the search instruction sentence on the basis of the semantic attributes and word expressions used as search keys by the search key determination unit 5. calculate. Various methods are known as a similarity calculation method, and any of them may be used. Here, for simplicity, 10 points are given to one search key.
The score obtained by adding the score for each search key included in the search target sentence is defined as the similarity. As a result, 20 for sentence 22
Points and sentences 21 are given 10 points.

As described above, in the case of the text type database of "product introduction articles", the sentence 22 "mobile phone of company B" is searched with the highest similarity in the search instruction sentence "telephone sold by company B". Is done.

Next, the case of a text database of "minutes" will be described with reference to FIG. “Minutes” including search target sentences such as search target sentence 11 “sales of mobile phones” and search target sentence 12 “mobile phone development”
Is assumed to be registered in the text database storage unit 6.

Here, the semantic attribute / role relation table 3
Then, the role of the semantic attribute [sales] is changed to r2.
When the search instruction sentence "sales of telephone" is searched, the semantic attribute [sales] is assigned to "sales" and its role is r2. 20 points and 10 points for sentence 12,
Sentence 11 “sale of mobile phone” is searched with the highest similarity.

Next, the difference between the change operation in the conventional apparatus and the change operation of the semantic attribute / role relation table 3 in the apparatus of the present invention will be described. For example, in the case of a text-type database such as “product introduction article” including search target sentences such as sentence 21 and sentence 22, “sales”, “sales”,
Words such as "release" are not important words in the search, and it is desirable not to search for these words. On the other hand, "minutes" including sentences to be searched, such as sentences 11 and 12,
In the case of a text database such as the above, it is necessary to search for the above words.

In the conventional device, the individual words "sales",
Although it is necessary to change “sales” and “release”, in the case of the apparatus according to the present invention, a semantic attribute [sales] is added to the above words in the word dictionary 1 and the semantic attribute / role relation table 3 is added. , It is only necessary to change the role of the semantic attribute [sales] from r0 to r2.

As described above, in the conventional apparatus, each word must be changed. However, in the apparatus according to the present invention, only by changing the role of the semantic attribute, a plurality of words including the semantic attribute can be changed. Can be changed at once, making it easy to build a text database suitable for each field. In order to assign the same semantic attribute to words belonging to the same concept, if there are N words having a certain semantic attribute, the conventional device must change the N words. However, in the apparatus according to the present invention, it is only necessary to change one semantic attribute, and it is only necessary to perform 1 / N of the work compared to the conventional technique.

Further, in the case of the text type database device according to the present invention, since the operation in the search processing of each word can be specified by the role assigned to the semantic attribute, the word dictionary 1 is constructed so that it can be used for general purposes. Therefore, a word dictionary common to other natural language processing can be used, and the labor for dictionary maintenance can be reduced.

For example, in a search process of a text-type database such as the “product introduction article” described above,
The information that “company A” and “company B” belong to the concept of “company” is not important, and it is important whether “company A” is included in the search target sentence. On the other hand, in a syntactic and semantic analysis or the like, when a rule that a word belonging to the concept of [company] can be the subject of "sell" is applied, "company A" and "company B" are replaced by "company". ] Is required. Therefore, in the word dictionary 1, a semantic attribute [company] is assigned to "company A" and "company B", and a role r1 is assigned to the semantic attribute [company] in the semantic attribute / role relation table 3. , Make sure that the semantic attribute [company] does not become a search key. As a result, the word dictionary 1 can be used for applying the above rules. FIG. 6 is a more detailed block diagram of the embodiment of the present invention .

Although not used as a search key, by providing a role that can set a semantic attribute to be considered in the similarity calculation, it is possible to perform efficient search.
Specifically, for example, a roll such as r4 shown in FIG. 3 is provided.

For example, suppose that the role of the semantic attribute [telephone] is defined as r4 in the text type database of the above-mentioned “product introduction article”. When there is a search target sentence including a word having the concept [telephone], the semantic attribute [telephone] is not used as a search key, but is searched for a word having another semantic attribute in the search instruction sentence. 8 when calculating the similarity, taking into account the semantic attribute [telephone].

In this case, if the search instruction sentence is “Company A's mobile phone”, the search target sentence containing the word having “Phone” is 2
Even if there are 00 sentences, if there are only 10 sentences to be searched for "Company A" or "mobile phone", similarity calculation only needs to be performed for those 10 sentences. be able to.

Also, by using various rolls,
Fine search control can be performed. FIG. 7 is a diagram illustrating an example of the relationship between word notation and semantic attributes and the relationship between search keys and similarity calculations when a plurality of semantic attributes are assigned to one word and various roles are assigned. . In the semantic attribute assigned to the word "ABC", the role r
Since there is a semantic attribute assigned with No. 3, the semantic attribute [b],
[C] is a search key, and the semantic attribute [a] to which no role is assigned is excluded from the search key. The notation "DE
F "and the semantic attribute [d] do not become search keys, but are considered in the similarity calculation.

[0038]

As described above, according to the present invention,
There is a semantic attribute / role relation table that describes the relationship between roles and semantic attributes, and the search is performed based on the role assigned to the semantic attribute.

(1) By using the created word dictionary, text type databases in different fields can be constructed so that the search can be performed with high accuracy and with high accuracy and with little effort to change the meaning attribute / role relation table.

(2) The word dictionary can be constructed so as to be versatilely used for other natural language processing, and the labor for dictionary maintenance can be reduced. And so on.

In particular, in the present invention, although not a search key,
By providing a role that can set a semantic attribute to be considered in the similarity calculation, more efficient search can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram of one embodiment of the present invention.

FIG. 2 is an explanatory diagram of a processing operation according to an embodiment of the present invention.

FIG. 3 is an explanatory diagram of an example of a rule of a roll according to an embodiment of the present invention.

FIG. 4 is an explanatory diagram of a processing operation according to the embodiment of the present invention.

FIG. 5 is a diagram showing an example of information registered in a word dictionary for explaining an embodiment of the present invention.

FIG. 6 is a configuration diagram of a detailed embodiment of the present invention .

FIG. 7 is a diagram illustrating an example of a relationship between a word notation and a semantic attribute, a search key, and similarity calculation in the embodiment of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Word dictionary 2 Morphological analysis part 3 Semantic attribute / role relation table 4 Role determination part 5 Search key determination part 6 Text type database storage part 7 Sentence search part 8 Similarity calculation part 10 Processing device

────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-4-54564 (JP, A) JP-A-2-287876 (JP, A) JP-A-1-300336 (JP, A) JP-A-2-300 253371 (JP, A) JP-A-2-253372 (JP, A) JP-A-4-84271 (JP, A) JP-A-4-90052 (JP, A) JP-A-4-314171 (JP, A) JP-A-4-1822871 (JP, A) JP-A-3-172966 (JP, A) Hiroshi Matsuo et al., "Text-Based Search Method Based on Semantic Attributes", Transactions of Information Processing Society of Japan, Vol. 32 No. 9, 1991 (Heisei 03-09-15), p. 1172-1179 Haruo Kimoto et al., "Japanese Text Database Retrieval and Classification Technology", NTT R & D Vol. 40 No. 7, Telecommunications Association of Japan, 1991 (Heisei 03-07-10), pp. 915-924, Atsuo Kawai, "Automatic Document Classification Method Based on Semantic Attribute Learning Results," Transactions of the Information Processing Society of Japan, Vol. 33 No. 9, 1992 (Heisei 04-09-15), p. 1114-1122 (58) Field surveyed (Int. Cl. ⁷ , DB name) G06F 17/00 JICST file (JOIS)

Claims (1)

(57) [Claims] 1. A word dictionary (1) defining a word notation, a semantic attribute, and a part of speech for each word, and a text type database storage unit (6) for storing a sentence to be searched and its related information. For each semantic attribute, a semantic attribute and a semantic attribute defining search attribute information indicating which of the semantic attribute and the word notation of the word to which the semantic attribute is assigned are used as a search key or not. A morphological analysis unit (2) that divides a search instruction sentence into words by referring to a relation table (3) of attribute information and the word dictionary (1), and extracts word expressions, semantic attributes, and parts of speech; Search attribute information for the semantic attribute extracted by the analysis unit (2) is extracted by the search attribute information determination unit (4) extracted from the relation table (3) and the search attribute information determination unit (4). Attribute information and words for each semantic attribute A search key determining unit (5) for determining a semantic attribute and a word notation as a search key based on the part of speech, and a search key determined by the search key determining unit (5). ), And a similarity between the search target sentence searched by the sentence search unit (7) and the search instruction sentence is determined by the search key determination unit (5). ) Is calculated based on the semantic attribute and word notation used as the search key .
Meaning of not being a search key in the search key determination unit (5)
Attribute and word notation
The search attribute information extracted in section (4) is
Information, the semantic attribute having the search attribute information
Or the word notation having the semantic attribute, or the search attribute
Of a semantic attribute having information and a word notation having the semantic attribute
A similarity calculating unit (8) for calculating a similarity by using both of them in the similarity calculation.