JP3422396B2 - Similarity search method based on viewpoint - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a similarity retrieval method for retrieving similar words based on the similarity of words using a database of word semantic attributes, and more specifically, a feature concept to be emphasized in retrieval. The present invention relates to a similarity search method based on the viewpoint of searching for similar words based on.

[0002]

2. Description of the Related Art In recent years, various databases have been created and used due to the dramatic improvement in the processing capacity of a computer typified by WS and the advent of a large-capacity and inexpensive storage medium. . For such a database,
Elements are searched and classified by determining the similarity between individual data (hereinafter, referred to as elements) forming a database or between data and an element input from the outside. For example, in the case of a document database accumulating information on various documents, it is searched whether there is a document similar to the designated document. In addition, regarding a set of text data such as a patent specification, those having similar contents are often collected and classified. Furthermore, when performing a similar search of the image database,
A plurality of characteristic words of an image are given to each element, and a similarity search using this characteristic word is performed.

When performing such a similarity search, it is necessary to determine the similarity between elements, and as a typical example, there is a method using a characteristic concept, which will be described. This method gives each element in the database a characteristic concept,
This is a method of discriminating similar elements by making a comparison about this. The feature concept here is also called a keyword or an attribute, and is expressed by a word representing the feature of an element. In the case of a database on animals, the element "horse"
About, "hoof", "mane",
"Livestock", "run" and other words can be characteristic concepts. In a technique called full-text search, similarity between texts is determined by using each word included in the text as a characteristic concept of the text.

There are various methods for determining the similarity between elements to which a plurality of such characteristic concepts are added. For example, a method of giving a similarity of 1 when a common feature concept exists in the elements to be compared and a similarity of 0 when there is no common feature concept, or a similarity in proportion to the number of matching feature concepts in the elements to be compared. There are other ways to give.

[0005] A similar element is searched by using the element similarity determination based on the characteristic concept. The general method is to calculate the degree of similarity between an element or all the elements in the database when searching the database for an element or an element similar to the data to which the characteristic concept input from the outside is added. , There is a method of outputting an element having a high degree of similarity as a search result.

[0006]

The above-described conventional similarity search method can only give a fixed search result even though the similar relationship between elements varies depending on the situation. There is a problem in that a similar similarity search cannot be performed according to the situation. The words "apple", "nanten", "sea bream",
This problem will be explained by using a database in which "monkey", "tangerine", and "chili pepper" have characteristic concepts.

FIG. 6 shows an example of a word database. When searching the database for an element similar to "Ringo" with element number 1, it is common to compare the characteristic concepts to find the similarity between the elements and output the search result based on the similarity. is there. For example, if the number of common elements between the compared elements is the similarity, the similarity to "apple" is 2 for "nanten", 1 for "monkey", 3 for "tangerine", and 1 for "chili pepper". Then, "Tangerine" with the highest similarity is output as the search result.

However, when the user expects a search result in which the feature concept "red" of "apple" is emphasized, "tangerine" which does not include "red" in the feature concept is obtained by the above search method, and "red" is obtained. It is impossible to obtain "nanten" and "chili pepper" that are regarded as similar words by comparing the characteristic concepts.

Such a characteristic concept that should be emphasized for retrieval is called a viewpoint. Since the search is performed in various situations or conditions, the viewpoint of the search changes depending on the situation or the like. Even when the similar search is performed for the same element, it is necessary to obtain the similar search result according to the viewpoint of the search.

On the other hand, even if a simple search is performed by regarding the viewpoint of the search as a search keyword, in the similar search in which the characteristic concept of the element to be searched is not taken into consideration, an element completely unrelated to the comparison element is searched. For example, in the keyword search which is a conventional method in which a word having the viewpoint “red” is given as a search result, an unexpected “monkey” is obtained at the same time in addition to “nanten”. Further, the characteristic concept "red" of "chili pepper" is synonymous with "red" and "chili pepper" should be included in the search result, but it is not included in the search result by the above keyword search method.

The present invention has been made in view of the above,
It is an object of the present invention to provide a similarity search method based on the viewpoint of searching for similar words in order to perform flexible similarity search according to the search situation.

[0012]

In order to achieve the above-mentioned object, the present invention according to claim 1 is a database having a feature concept in which each of a plurality of elements is a word representing a feature of the element, in the database. When a similar element similar to the element to be searched specified from is searched from the database, the element to be searched and a viewpoint that is a characteristic concept to be emphasized in the search are specified, and a characteristic concept equal to the viewpoint is held. Elements that are searched for from the database as similar element candidates, the characteristic concept of the element designated as the search target is compared with the characteristic concept of the similar element candidates, and the higher the number of common characteristic concepts, the higher the similarity. Degree is calculated, and a plurality of similar element candidates with high similarity are output as similar elements, or a degree of similarity equal to or higher than a predetermined lower limit is given. Like elements candidate Te is summarized in that output as similar elements.

According to the first aspect of the present invention, an element to be searched and a viewpoint to be emphasized in the search are designated, and an element having a characteristic concept equal to this viewpoint is searched from the database to be a similar element candidate. The feature concept of the element designated as the search target is compared with the feature concept of the similar element candidate, the higher the similarity is calculated as the number of common feature concepts is increased, and the plurality of similar element candidates with high similarity are calculated. Is output as a similar element, or all similar element candidates that give a degree of similarity equal to or greater than a predetermined lower limit are output as similar elements, and thus it is possible to obtain a similar search result according to a viewpoint representing the search situation. At the same time, the search processing time can be shortened.

The present invention according to claim 2 is the same as claim 1.
In the invention described above, when searching an element having the viewpoint, an element list including the characteristic concept is created for each characteristic concept in the database, and the element is searched using the element list. Is the gist.

According to the second aspect of the present invention, an element list including the characteristic concept is created for each characteristic concept in the database, and the element is searched using the element list.

Further, the present invention according to claim 3 is similar to an element to be searched, which is designated from the database, in a database having a characteristic concept in which each of a plurality of elements is a word representing the characteristic of the element. When retrieving similar elements from the database, all characteristic concepts in the database are replaced with semantic classifications that are meaning-based classification names, and the element to be searched and the viewpoint that is a characteristic concept to be emphasized in the search are specified. , Searching the database for an element having a semantic classification equal to the semantic classification of the viewpoint as a similar element candidate, comparing the semantic classification of the element designated as the search target with the semantic classification of the similar element candidate, The higher the number of common semantic classifications, the higher the similarity is calculated, and the top multiple similar element candidates with high similarity are output as similar elements. Either, or a gist that output all similar elements candidate giving pre-determined lower limit or more similarity as similarity factors.

According to the present invention as set forth in claim 3, all the characteristic concepts in the database are replaced with semantic classifications based on meanings, the element to be searched and a viewpoint to be emphasized in the search are specified, and a semantic classification equal to this viewpoint is set. The retained element is searched from the database as a similar element candidate, the semantic classification of the element designated as the search target is compared with the semantic classification of the similar element candidate, and the higher the number of common semantic classifications, the higher the similarity. By calculating the degree of similarity and outputting a plurality of similar element candidates having a high degree of similarity as similar elements, or outputting all the similar element candidates that give a degree of similarity equal to or higher than a predetermined lower limit as similar elements, It is possible to obtain the similar search result according to the viewpoint representing the situation and reduce the search processing time. In addition, it is possible to detect a feature concept that is synonymous or close in meaning to the viewpoint in meaning, and it is possible to perform a flexible similarity search that is close to human sense.

According to a fourth aspect of the present invention, in the invention according to the third aspect, when searching for an element that holds the semantic classification of the viewpoint, the semantic classification is performed for each semantic classification of each element in the database. The gist is to create an element list including the elements and search for the element using the element list.

According to the fourth aspect of the present invention, an element list including the semantic classification is created for each semantic classification of each element in the database, and the element is searched using the element list.

[0020]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a flowchart showing the operation of the similarity search method based on the viewpoint according to the first embodiment of the present invention. The similar search method based on the viewpoint shown in the figure introduces a characteristic concept a which is a viewpoint to be emphasized in the search when searching the database for an element similar to the element A in the database.

The operation of the similarity search method will be described with reference to the flowchart shown in FIG. In FIG. 1, first, the element A to be searched and the viewpoint a of the search are specified, and
A parameter i for counting the number of elements is set to 1 (step S11). The target of the search may be specified from the database, or may be any data as long as it has the characteristic concept like the elements in the database. Further, from the viewpoint a, the searcher inputs a characteristic concept that is important when performing a search. At that time, the viewpoint may be selected from the characteristic concept in the element A.

Next, it is checked whether the parameter i is smaller than the number of elements (step S12). If the parameter i is smaller than the number of elements, the process proceeds to step S13 to search for all the elements, and it is checked whether or not the viewpoint a is included in the characteristic concept of the element Ai. Returns to step S12 to increment the parameter i in step S15 and perform the same process for the next element. However, if the viewpoint a is included in the characteristic concept of the element Ai, the element Ai is set as the search candidate α. Add to the inside (step S14). Then, the parameter i is incremented (step S15), and step S1
Return to 2.

As described above, it is determined whether or not the feature concept equal to the viewpoint a is included for all the elements in the database, and the search candidate α consisting of the included elements is determined.
To generate.

When the above determination is made for all the elements in the database and search candidates are generated, the parameter i becomes larger than the number of elements in step S12.
It proceeds to step S16.

In step S16, the element A and the search candidate α
Calculate the similarity with the elements inside. Any calculation method may be used as the calculation method of the similarity degree as long as the common number of the characteristic concepts in the elements to be compared increases, the similarity degree increases. Finally, the element having a high degree of similarity in the search candidates α is output as a similarity search result (step S17). As a method of determining a high similarity element, a method of outputting a plurality of high similarity similarity elements as a result, or a lower limit of similarity is determined in advance, and an element giving a similarity equal to or higher than the lower limit of similarity is given. Any method such as a method of outputting everything may be used.

Next, the similarity search method of FIG. 1 will be described more specifically with reference to FIG. An element similar to the element A “apple” in the database of FIG. 6 is searched and acquired for the viewpoint a “red”.

First, the viewpoint "red" from the database
When the element including is searched, "nanten" and "monkey" are obtained, and these are set as similar search candidates α. Next, the degree of similarity with the element A “apple” and each element in the similar search candidate α is calculated. Here, the number of feature concepts common to the elements to be compared is the similarity. In this case, the degree of similarity is 2 for "South Ten" and "Apple",
"Monkey" and "Apple" become 1. When outputting the element with the highest degree of similarity as a search result, "nanten" is output as an element similar to "apple" in terms of "red".

In the similarity search method of the above embodiment, when it is determined whether or not all the elements in the database include the characteristic concept equal to the viewpoint a, it is necessary to determine all the elements in the database each time a search is performed. . Therefore, for all types of feature concepts included in the database,
An element list including a set of element names including the characteristic concepts is created in advance. Then, when searching for a feature concept equal to the viewpoint, the feature concept corresponding to the viewpoint in the element list is referred to, a set of element names including the feature concept is set as a search candidate, and similar search is performed for this search candidate. . By referring to the element list, it is possible to determine whether or not the feature concept equal to the viewpoint a is included only in the element list.

Here, with respect to the database shown in FIG. 6 as a concrete example, the element "apple" from the viewpoint of "red"
A method of searching for a similar element will be described as an example. When an element including the viewpoint “red” is searched from the database of FIG. 6, an element list including the characteristic element is created for each characteristic element included in the database, as shown in FIG. By referring to this element list, it is possible to determine that the element including the characteristic concept "red" is "southern" or "monkey" without checking all the elements in the database.

Next, a second embodiment will be described.
In this embodiment, when comparing a viewpoint and a characteristic concept in order to search an element in a database including a viewpoint of a search, each viewpoint and the characteristic concept are converted into meaning-based semantic classification names, and character strings are different. Is a feature concept that has almost the same viewpoint and meaning, and performs a similarity search considering the meaning of words.

Before searching, all the characteristic concepts of each element in the database are converted into semantic classification names. The meaning classification name represents a name given to the cluster by clustering (semantic classification) of features having similar meanings. As a method of semantic classification, a method of determining by referring to the classification vocabulary, a cluster analysis based on the similarity between all types of characteristic concepts, and a method of assigning a semantic classification name from the obtained cluster, or an existing method Any method may be used, such as a thesaurus or a thesaurus method.

At the time of search, first, the element A to be searched and the semantic classification a from the viewpoint of search are specified. The target of the search may be designated from the database, or may be any data as long as it has the characteristic concept like the elements in the database. When searching data outside the database, the characteristic concepts in the data are converted into semantic categories. Further, from the viewpoint a, the searcher inputs a characteristic concept that is important when performing a search. At that time, the viewpoint may be selected from the semantic classification obtained by converting the characteristic concept in the element A. When the searcher inputs the viewpoint a, the viewpoint a is converted into the semantic classification. Next, for all elements in the database, it is determined whether or not the semantic classification equal to the viewpoint a is included, and the search candidate α consisting of the included elements is determined.
To generate.

Then, the degree of similarity between the element A and the element in the search candidate α is calculated. Any calculation method may be used as the similarity calculation method as long as the larger the common number of semantic classifications among the compared elements, the higher the similarity.
Finally, the element with a high degree of similarity in the search candidates α is output as a similarity search result. As a method of determining a high similarity element, a method of outputting a plurality of high similarity similarity elements as a result, or a lower limit of similarity is determined in advance, and an element giving a similarity equal to or higher than the lower limit of similarity is given. Any method such as a method of outputting everything may be used.

Here, the present embodiment will be described more specifically using the database shown in FIG. First, all the characteristic concepts of each element in the database are converted into the semantic classification using the semantic classification table of FIG. For example, when converting the characteristic concept "red" of the element "chili pepper", search for a meaning classification that includes "red" in the meaning classification table, and use the meaning classification name "red" to identify "red" in the database. replace. In this way, the database of FIG. 6 is converted as shown in FIG.

Next, the point of view "red" is replaced with the meaning classification "red" using the meaning classification table of FIG. Then, in the database of FIG. 4, it is checked whether or not each element includes the meaning classification “red”, and the included elements “apple”, “nanten”, “monkey”,
A set of "chili peppers" is set as a search candidate α.

Then, the degree of similarity between the element A "apple" and each element in the similar search candidate α is calculated. Here, the number of feature concepts common to the elements to be compared is the similarity. In this case, the degree of similarity is 3 for "nanten" and "apple", 1 for "monkey" and "apple", and 3 for "chili pepper" and "apple". When outputting the element with the highest degree of similarity as a search result, "nanten" and "chili pepper" are output as elements similar to "apple" in terms of "red". In the first embodiment, the viewpoint "red" and the meaning "red" are similar in the element "chili pepper".
Despite being included, it was not included in the search candidates and search results. By using the semantic classification table as shown in FIG. 3, in the second embodiment, “chili pepper” similar to “apple” in terms of “red” can be output together with “nanten”.

In the second embodiment, when it is determined whether or not all elements in the database include the semantic classification equal to the viewpoint a, when every element in the database is checked each time a search is performed, the determination is made every time. Take time for. Therefore, for all types of semantic classifications included in the database, an element list including a set of element names including the semantic classifications is created in advance. Then, when searching a semantic classification that is equal to a viewpoint, the semantic classification corresponding to the viewpoint in the element list is referred to, a set of element names including the semantic classification is set as a search candidate, and similar search is performed for this search candidate. .

Now, the second embodiment will be described more specifically. An example of a method of searching for similar elements of the element "apple" from the viewpoint of "red" in the database of FIG. 4 will be described. FIG. 5 shows a list of elements including the meaning classification for each meaning classification included in the database when checking the elements including the viewpoint “red” from the database of FIG. By referring to this element list, it is possible to determine that the element including the characteristic concept "red" is "nanten", "monkey", "chili pepper" without checking all the elements in the database. .

[0040]

As described above, according to the present invention as set forth in claim 1, a viewpoint to be emphasized in a search is designated, an element having a characteristic concept equal to the viewpoint is searched from a database, and similar element candidates are searched. Since similar elements are searched from the similar element candidates, the similar search according to the viewpoint of the search situation is made in comparison with the similar search result in which only the output of the search result which has been fixed in the past is obtained. The result can be obtained and the search processing time can be shortened.

According to the second aspect of the present invention, an element list including the characteristic concept is created for each characteristic concept in the database, and an element is searched using the element list. Only the table is referenced, and the search processing time can be shortened.

Further, according to the present invention as set forth in claim 3, all the characteristic concepts in the database are replaced with the semantic classification based on the meaning, the viewpoint which is the characteristic concept to be emphasized in the search is designated, and is equal to the semantic classification of the viewpoint. Since an element having a semantic classification is searched from a database as a similar element candidate and a similar element is searched from the similar element candidate, it is possible to detect a characteristic concept that is synonymous with or similar in meaning to the viewpoint, and a human sense. It is possible to perform a flexible similarity search close to.

According to the present invention as set forth in claim 4, since an element list including the meaning classification of each element of each element in the database is created and the element list is searched, the element is searched. Since only the list is referenced, the search processing time can be shortened.

[Brief description of drawings]

FIG. 1 is a flowchart showing an operation of a similarity search method based on a viewpoint according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of an element list including a characteristic concept.

FIG. 3 is a diagram showing an example of a word database in which a characteristic concept is converted into a semantic classification.

FIG. 4 is a diagram showing an example of a database represented using semantic classification names.

FIG. 5 is a diagram showing an example of an element list including meaning classifications.

FIG. 6 is a diagram showing an example of a word database.

Continuation of the front page (56) References JP-A-6-162099 (JP, A) JP-A-5-233689 (JP, A) Kasahara Kaname et al. Multi-view concept based self-refining method, IEICE technology Research report, Japan, The Institute of Electronics, Information and Communication Engineers, May 13, 1994, VOL94 No. 33, pp. 1-8 Kaname, et al., Automatic generation method of viewpoint in similar word search, IPSJ research report, Japan, IPSJ, 25 July 1996, 96-F1-42, Page 29 to 36 Kaname, Kaname et al., Similarity search for common sense words using concept base, IEICE Technical Report, Japan, The Institute of Electronics, Information and Communication Engineers, September 28, 1995, AI95-22 ~ 23, pp. 23 to 30, Kaname, et al., Similarity discrimination between concepts based on viewpoint, IPSJ journal, Japan, IPSJ, March 15, 1994, vol35 No. 3, p. 505, p. 509, 509, Kasahara et al., Concept-based construction method based on refinement, IEICE technical report, Japan, The Institute of Electronics, Information and Communication Engineers, May 26, 1995, DE95-1 to 8, pages 49 to 56 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 17/30 JISST file (JOIS)

Claims (4)

(57) [Claims] 1. A database that holds a feature concept, wherein each of a plurality of elements is a word representing the feature of the element ,
Similar to the search target element specified in the database
When searching for similar elements from the database, the elements to be searched and the features to be emphasized in the search
Specifies the concept is in view, the element carrying the same features concept with the viewpoint retrieved from said database and similar elements candidate, the compound wherein the concept of the element specified as the search target
Comparing the characteristic concepts of similar element candidates,
The higher the number, the higher the similarity is calculated, and the top multiple similar element candidates with high similarity are set as similar elements.
Output or give a similarity above a predetermined lower limit
A similar search method based on the point of view that all the similar element candidates are output as similar elements . 2. When retrieving an element having the viewpoint, an element list including the characteristic concept is created for each characteristic concept in the database, and the element is retrieved using the element list. A similarity search method based on the viewpoint of claim 1. 3. A database that holds a feature concept, each of a plurality of elements being a word representing the feature of the element ,
Similar to the search target element specified in the database
When searching for similar elements from the database, all feature concepts in the database are replaced with semantic classification that is a classification name based on meaning, and elements to be searched and features to be emphasized in the search
Specifies the viewpoint is a concept, the elements carrying the same meaning classification and semantic classification of the viewpoint retrieved from said database and similar elements candidate semantic classification and the class of the specified element as the search target
Comparing the semantic classifications of similar element candidates,
The higher the number, the higher the similarity is calculated, and the top multiple similar element candidates with high similarity are set as similar elements.
Output or give a similarity above a predetermined lower limit
A similar search method based on the point of view that all the similar element candidates are output as similar elements . Wherein when searching for elements that possess semantic classification of the viewpoints for each semantic classification of the elements in said database to create an element list containing the semantic classification, by using the element list The similarity search method based on the aspect of claim 3, wherein the element is searched.