JP3374946B2 - Search device - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to the collation of a feature amount of an input pattern for recognizing, for example, a figure or a character with template information stored in a pattern dictionary, and a database. The present invention relates to a search device used for information matching and information search, such as searching for similar records, searching for similar concepts / similar rules in analogous inference of knowledge processing, and the like. 2. Description of the Related Art In information collation and information retrieval such as pattern recognition, database retrieval, and knowledge base retrieval, input patterns and retrieval requests are applied to previously stored pattern templates, database records, and knowledge descriptions. Even if they do not match exactly, there are many demands to find the most favorable ones according to certain selection criteria. Such information collation and information retrieval are most realized by defining an evaluation function that can express the degree of preference as a scalar quantity, and using the evaluation function that maximizes the evaluation function as the collation result or search result. General. In this case, the evaluation function is an input pattern or a search request, and converts three variables of the stored template or record and a parameter (hereinafter referred to as a parameter) for a user to specify a selection criterion into one scalar quantity. It becomes a function to map. Here, these three variables are generally structured quantities, that is, quantities composed of a plurality of names and a plurality of quantities. Finding an element that maximizes the evaluation function for a given search request and parameter from a plurality of stored elements is referred to as optimal element search. [0003] A more specific description will be given below by taking a search for a similar record in a database as an example. As shown in FIG.
Record 201 which is one element is heading 2
02 and a sequence of one or a plurality of attribute names 203 and weights 204 (this pair is also simply referred to as an attribute), and a search request and a user request having the same format as the record of the database are provided. It is considered that a record having the highest similarity to the parameter representing the selection criterion is searched from the database. Here, the weight 204 takes a real value. [0004] The similarity is a real scalar quantity calculated from the sequence of attributes of the search request, the sequence of attributes of database records, and the parameters. Calculated based on their corresponding weights. As an example of the similarity, a value obtained by multiplying the search request and the weight of the database record for each attribute and summing up the result of the multiplication for all common attributes with respect to the attribute common to these three parties can be considered. . Of course, the similarity shown here is only an example, and if it is a function that can calculate a real scalar quantity from the three variables of the attribute list of the search request, the attribute list of the record of the database, and the attribute list of the parameter, May be any. In order to perform such a search, a search device as shown in FIG. 4 has been conventionally realized. The element storage unit 101 stores records of the database, and the evaluation value calculation unit 103 calculates the similarity from the search request, the parameter, and the record given by the user. The search control unit 104 is a part that controls the search operation, and stores a maximum evaluation value storage variable 105 that holds the maximum value of similarity during the search and an optimal element storage variable 10 that holds a record that gives the maximum value of the similarity.
6 and operates as follows. When the user gives a search request and parameters, the search device first retrieves one record from the database, calculates the similarity, stores the evaluation value in the maximum evaluation value storage variable 105, and stores the record in the maximum evaluation value storage variable 105. It is stored in the optimum element storage variable 106. Subsequently, another record that has not been retrieved is retrieved from the database, the similarity is calculated using the evaluation function, and compared with the maximum value of the similarity stored in the maximum evaluation value storage variable 105. If the calculated similarity is larger, the calculated similarity is stored in the maximum evaluation value storage variable 105, and the extracted record is stored in the optimum element storage variable 106. After fetching records that have not been fetched and calculating similarity, comparing the magnitudes and storing variables based on them, were repeated for all records, the records were stored in the optimal element storage variables 106. Output records as search results. Such a retrieval device that retrieves all records and calculates an evaluation function is called a full-scan retrieval device. The full-scan search device calculates the similarity for all the requests stored in the element storage unit 101 when the search request is generated.
Requires a time proportional to the number of elements stored in the. [0007] In order to shorten the search time, a search apparatus has been considered in which a table in which all similarities are calculated and sorted in advance is stored in a storage area. However, since the degree of similarity takes three elements as variables, the size of the table created by calculating the degrees of similarity for all variables is proportional to the cube of the elements stored in the element storage unit. [0008] The full-scan search device is
Since the evaluation function is calculated for all the elements stored in the element storage unit, the search time increases in proportion to the number of elements stored in the element storage unit. Therefore, there is a problem that the search time is long when there are many elements stored in the element storage unit. Further, in a search device that stores a table in which all similarities are calculated and sorted in advance, the search time can be reduced, but a storage area having a scale proportional to the cube of the number of elements stored in the element storage unit is required. Becomes As described above, the conventional device has a problem that a long search time is required. Also,
To solve this problem, a very large storage area is required. [0009] The present invention has been made in view of the above,
An object of the present invention is to provide a search device capable of performing a high-speed search with a small storage area. [0010] In order to achieve the above object, a retrieval apparatus according to the present invention comprises an element storage unit storing a plurality of elements each having at least one attribute which is a set of an attribute name and a degree of importance. When,
For an attribute whose attribute name matches between any two elements and a parameter having the same structure as the element, an evaluation function whose evaluation value is determined based on the importance of the attribute is defined. An evaluation value calculation unit that calculates an evaluation value, and when one element is given as a search request and a parameter is given as a selection criterion, the evaluation value is maximized from all elements other than the element given as the search request. A search control unit for searching for an element, and for all attribute names included in the element stored in the element storage unit, information on the correspondence between each attribute name and the element including the attribute name When the element as a search request and the parameter as a selection criterion are given, the search control unit relates to an attribute name commonly included in the element and the parameter. An element including the attribute name is extracted from the element storage unit with reference to the attribute index storage unit, an evaluation value is calculated for each of the extracted elements, and an element having the maximum evaluation value is selected. And According to the search device of the present invention, with respect to all the attribute names included in the elements stored in the element storage unit, information on the correspondence between each attribute name and the element including the attribute name is stored in the attribute search storage unit. When a search request and a parameter are given, the search control unit refers to the attribute index storage unit to limit the search target to only the element including the attribute name commonly included in the search request and the parameter. Refine and take out the element from the element storage unit,
The evaluation value is calculated for each of the elements, and the element having the highest evaluation value is selected. Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a search device according to one embodiment of the present invention. In the figure, an element storage unit 101 is a part for storing elements, and stores elements having a structure as shown in FIG. 2 so that an arbitrary element can be extracted according to a request from the outside. The evaluation value calculation unit 103 calculates an evaluation value according to an evaluation function and outputs the evaluation value to the outside when the element of the search request, the element to be searched, and the parameter are given from outside. As the evaluation function, the similarity function described as an example in the section of the related art is typically considered. In this embodiment, the similarity function will be described as an evaluation function. The attribute index storage unit 102 stores, for each attribute name, information in which elements including the attribute name are listed.
Specifically, as shown in FIG. 3, a method of storing a list of attribute names and element numbers is conceivable. However, the structure of the storage content shown here is not essentially defined in the present invention, and any structure may be used as long as the correspondence information between the attribute name and the element including the attribute name can be stored. Absent.
The search control unit 110 is a part that controls the search operation, and incorporates a maximum evaluation value storage variable 105, an optimum element storage variable 106, and a common attribute list variable 107. The element that can obtain the maximum evaluation value under the given parameters is searched from the element storage unit 101. Hereinafter, the operation of the search control unit 110 will be described in detail with reference to a processing flowchart shown in FIG. In FIG. 5, first, in a search request storage processing step S601, elements and parameters of a search request from a user are received and held as variables in the control unit. Next, in a storage variable initialization processing step S602, the maximum evaluation value storage variable 105 is initialized to be equal to or less than the minimum value that the evaluation value can take, and the optimum element storage variable 106 is initialized to “empty”. Since the evaluation function in the present embodiment takes a value of 0 or more, the maximum evaluation value storage variable 105 is initialized to 0 in FIG. Subsequently, common attribute extraction processing step S6
In step 03, an attribute name that the element and the parameter of the search request have in common is extracted and stored in the common attribute list variable 107. The above processes are performed only once at the beginning of the search process immediately after the search request and the parameter are given.
The order of these processes does not need to be the same as described above, as long as necessary data is available and the processes can be performed.
For example, the common attribute extraction processing step S603 may be performed before the storage variable initialization processing step S602. The subsequent processing is repeated until the search termination condition is satisfied. First, the common attribute list variable 10
7 to extract one attribute name. Here, the previously extracted attribute name is not extracted twice. Subsequently, in the element information extraction processing step S605, the attribute index storage unit 102 is referred to based on the attribute name, and one piece of element information including the attribute is extracted. At this time, the previously extracted information is not extracted twice. Further, as described above, in this embodiment, an element number is used as element information. In the element retrieval processing step S606, one element corresponding to the element number is retrieved from the element storage unit 101 based on the element number retrieved in the element information retrieval processing step S605. When the element is extracted, in evaluation value calculation processing step S607, the extracted element, the element of the search request, and the parameter are given to the evaluation value calculation unit 103 to obtain an evaluation value. The evaluation value obtained here is compared with the evaluation value stored in the maximum evaluation value storage variable 105 (comparison processing step S613). If the obtained evaluation value is large, the storage variable update processing step S613 is performed.
608 is performed. In the storage variable update processing step S608, the obtained evaluation value is stored in the maximum evaluation value storage variable 105.
The extracted element is stored in the optimum element storage variable 106. At this time, the values previously stored in each variable are deleted. The above element information extraction processing step S6
05 to the stored variable update processing step S608 until the element including the attribute for one attribute name cannot be retrieved from the attribute index storage unit 102 (repetition processing step S612). The process is repeated until the attribute name cannot be extracted from the common attribute list variable 107 in step S604 (repetition processing step S611). At the end of the repetition, the element stored in the optimum element storage variable 106 is the optimum element, and the evaluation value at that time is stored in the maximum evaluation value storage variable 105, which is output as a result. As described above, by searching only elements having common attributes necessary for increasing the evaluation value, elements to be searched can be reduced and high-speed search can be realized. As described above, in the search device of the present invention,
Attribute index storage unit that stores, in addition to the constituent elements of the conventional all-scan search device, information on the correspondence between individual attribute names and elements including the attribute names for all attributes included in the stored elements. 102, a search control unit 110
Does not perform a search for all elements as in the conventional method, but when a search request and a parameter are given, refers to the attribute index storage unit to search for the search target and the parameter. It is characterized in that an optimum element search is performed by narrowing down to only elements including an attribute name that is commonly included in. Since the value of the evaluation function is calculated from the attributes and parameters of the search request and the attributes that are commonly included in the three elements to be searched, the attributes that are not commonly included in the three are included in the value of the evaluation function. Has no effect. As described above, the point that the search target is narrowed down to only the element including the attribute common to the three members where the value of the evaluation function becomes large is essentially different from the conventional full-scan search apparatus. Further, in order to narrow down the search target, a configuration difference from the conventional method is that an attribute index storage unit storing correspondence information between attribute names and elements including the attribute names is provided as a component. That is, in the present invention, when searching for the element having the maximum evaluation value, only the element including the attribute which is shared by the element of the search request and the parameter is searched by referring to the attribute index storage unit. And For this reason, elements to be searched can be narrowed down, and the number of evaluation value calculations through the search operation is reduced, thereby speeding up the optimum element search. Further, the attribute index storage unit has a format in which the information in the element storage unit is rearranged based on the attribute names, so that the storage capacity is almost the same as that of the element storage unit. In this embodiment, the search control unit 110 stores the maximum evaluation value storage variable 105 and the optimum element storage variable 106 to store the maximum evaluation value and the element giving the maximum evaluation value.
Is not essentially defined in the present invention, and any method may be used as long as the maximum evaluation value during the search and the element giving the maximum evaluation value can be known. For example, a method of giving a flag area to the elements stored in the pre-evaluation value storage unit and the element storage unit and setting a flag to the element giving the maximum evaluation value is also conceivable. Also, the operation of the search control unit 110 is not specified in the flowchart of the present embodiment. Any processing flow may be used if a search is performed. Further, in the configuration diagram of this embodiment, each unit is represented as an independent module, but it is not necessary that these units are physically independent. For example, using a computer including a processor and a memory, the element storage unit 10
1, attribute index storage unit 102, maximum evaluation value storage variable 10
5, optimal element storage variable 106, common attribute list variable 10
7 are allocated to the memory, and the evaluation value calculation unit 103
A method of allocating each function of the search control unit 110 to the processor can be considered. As described above, according to the present invention,
When searching for an element that maximizes the evaluation function from among a large number of elements based on the elements and parameters of the search request, an attribute index storage unit having the same storage capacity as the element storage unit is provided. Based on the information stored in the search request, the search is limited to only the elements having the common attribute to the elements and parameters of the search request. Therefore, even if there are many elements, the search can be performed quickly with a relatively small storage capacity. can do.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a block diagram showing a configuration of a search device according to one embodiment of the present invention. FIG. 2 is a diagram showing an example of an element storage mode. FIG. 3 is a diagram illustrating an example of a storage form of an attribute index storage unit. FIG. 4 is a block diagram showing a configuration of a conventional search device. FIG. 5 is a flowchart showing the operation of the search device of the embodiment shown in FIG. [Description of Signs] 101 element storage unit 102 attribute index storage unit 103 evaluation value calculation unit 105 maximum evaluation value storage variable 106 optimal element storage variable 107 common attribute list variable 110 search control unit

Continuation of the front page (56) References JP-A-7-85068 (JP, A) JP-A-6-162099 (JP, A) Takashi Yukawa, Kaname Kasahara, Kazumitsu Matsuzawa, Tsutomu Ishikawa, About inference: Multi-perspective concept base Acceleration of Similarity Concept Retrieval in IPSJ, The 47th Annual Meeting of the Information Processing Society of Japan (late 1993), National Symposium on Informatics (3), Information Processing Society of Japan, September 27, 1997, pp. 51-58 Takashi Yukawa, Kaname Kasahara, High-speed Search Method for Similar Concepts, IPSJ 49th (Late 1994) National Conference Proceedings (3), September 20, 1994, pp. 75-76 Kazumitsu Matsuzawa, Tsutomu Ishikawa, Takashi Yukawa, Tsukasa Kawaoka, About the Basic Concept of About Inference Method, IEICE Technical Report (AI93-77), January 20, 1994, Vol. 93, No. 93. No. 424, p. 41-48 (58) Field surveyed (Int. Cl. ⁷ , DB name) G06F 17/30 JICST file (JOIS)

Claims (1)

(57) [Claim 1] An element storage unit storing a plurality of elements having at least one attribute which is a set of an attribute name and a degree of importance, and arbitrary two elements and the same structure as the elements An evaluation function whose evaluation value is determined based on the importance of the attribute is defined for an attribute whose attribute name matches among the three parameters, and an evaluation value calculation unit that calculates the evaluation value according to the evaluation function. A search control unit that searches for an element that maximizes an evaluation value among all elements other than the element given as a search request when one element is given as a search request and a parameter is given as a selection criterion. A search device having an attribute index storage unit that stores information on correspondence between each attribute name and an element including the attribute name for all attribute names included in the element stored in the element storage unit . The search control unit includes: When given as an element and a parameter as a selection criterion, regarding an attribute name commonly included in the element and the parameter, referring to the attribute index storage unit, the element including the attribute name is referred to as A retrieval apparatus, wherein the retrieval unit retrieves an element from an element storage unit, calculates an evaluation value for each of the retrieved elements, and selects an element having the maximum evaluation value.