JP2001134599A - Method and device for retrieval of database and recording medium stored with recorded retrieving program for database - Google Patents

Abstract

PROBLEM TO BE SOLVED: To generate suitable retrieval parameters by 'the selection of an object' that a user can view. SOLUTION: A retrieval result display part 10 displays the result of primary retrieval carried out according to initial retrieval parameters to a user in display format corresponding to the initial retrieval parameters. The user views the displayed primary retrieval result and changes and inputs selection conditions to a retrieval statement generation part 7 so that an object similar to the feature variable of a requested object is retrieved; and the retrieval statement generation part calculates hypothesis retrieval parameters from the changed selection conditions and outputs them to a retrieval result display part. The retrieval result display part changes the display format of the primary retrieval result according to the hypothesis retrieval parameters and displays it to the user. Those operations are repeated and secondary retrieval of the database 2 is carried out by using a retrieval statement generated according to hypothesis retrieval parameters that the user finally judges to be proper.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a database search system for searching a database, and more particularly, to a method and an apparatus for the database search system to generate appropriate search parameters.

[0002]

2. Description of the Related Art As a conventional database search method, there is a method in which a user adjusts a search key and a search parameter by himself to create a search sentence, and accesses a database. FIG. 1 is a flowchart for explaining the flow of a conventional database search method. In the following description, data (documents, images, music, etc.) stored in the database will be referred to as “objects”.

In a conventional database, an object O
There n feature amounts ^{O = {o 1, o 2} , ... o n} when with, the user vector K = {k ^1, k ² specifying the set of n types of features as a search key, ... Specifies k ^n}, to specify: a search parameter. 1) The number of required answers q required by the user By specifying the number of required answers, it is specified how many items having higher similarity are required. 2) Weight for each feature amount W = {w ¹ , w ² ,..., W ⁿ } By specifying the weight for each feature amount, which feature amount and how much importance is specified. 3) Similarity evaluation function F = ｛f ¹ , f ² ,.
As the f ⁿ ｝ similarity evaluation function F, for example, a Manhattan distance function or a Euclidean distance function is used.

In FIG. 1, a user specifies a search parameter and creates a search sentence (step S1). The database search unit obtains the similarity of the search key K for each object in the database, and ranks the top q
A case is obtained (step S2). The q objects having a high degree of similarity are presented to the user as similar objects (step S3). The user determines whether there is an object he wants (step S4). If not, the user returns to the process of step S1 in which the user adjusts the search key and the search parameters to create a search sentence. The processing from step S1 to step S4 is repeated until an object desired by the user is displayed.

In step S2, the similarity d between the object O stored in the database and the search key K
Is calculated from the following equation (1).

[0006]

(Equation 2)

Here, f ⁱ (k ⁱ , o ⁱ ) is a similarity evaluation function between the feature quantity k ⁱ and the feature quantity o ⁱ . In the following description, the above conventional database search method is referred to as a first conventional technique.

[0008] An image search method that simplifies the adjustment of a search key and a search parameter performed by a user with respect to the first prior art is described in Japanese Patent Laid-Open No. 6-96134. Hereinafter, the image search method described in this publication will be referred to as a second conventional technology.

Also in the second prior art, a search key is defined to represent an attribute value of a target (several partial images characterizing an image). That is, an object characterized by a plurality of attribute values is expressed in a 1 or 0 vector format indicating the presence or absence of each attribute, and a search key is similarly expressed in a 1 or 0 attribute value vector format.

In the second prior art, an appropriate search is performed by inputting a primary search result searched using initial search parameters and a selection condition of “similar” or “dissimilar”. Adjust keys and search parameters. FIG. 2 is a flowchart for explaining the flow of such a database search method.

According to the flow chart of FIG. 2, a user first searches for a search key having the same attribute as a desired attribute among attributes forming an attribute value vector of a target image. Then, with respect to all the images stored in the database using the other search key having a feature amount of 0) as a search condition, the difference between the attribute value vector of each image and the search condition is calculated, and the difference is small. The database is primarily searched in order from the image (step S1). The search system displays the primary search result (Step S2). The user inputs selection conditions to the search system and selects a similar image and a dissimilar image from the displayed images (step S3). Next, the search system extracts an attribute value vector from the selected similar images and dissimilar images, calculates the dissimilarity between similar images and the dissimilarity between dissimilar images, and calculates a display distance defined as the sum of them. Calculate and generate a new search sentence for secondary search of the database in ascending order of display distance (step S
4). The search system performs a secondary search (step S
5), and display the search result (step S6). The user determines whether there is a target image from the secondary search result (step S7). If there is no target image, the search system removes the current display object from the display target images, and repeats steps S3 to S7 until the target image is searched or until there are no more display target images.
The process up to is repeated.

[0012]

The advantage that the system automatically generates a search key and a search parameter in the above-described conventional database search method is that, for example, a common feature of two images (AND of two images) is used. When performing a search as a search key, the following adverse effects may be caused.

FIG. 3 is a block diagram for explaining the problems of the conventional database search method, and FIG. 4 is a diagram showing an example of the contents of the search unit in FIG. The example object storage unit 31 of FIG. 3 is a memory that stores example objects A, B, C, D, and E prepared by the search system for the initial search. The search unit 32 is an information processing unit that executes a search program. The display unit 33 is a display control device and a display device that display search results. Search unit 32
This part is a black box for the user, because the user cannot see the configuration of and the operation.

In the search system shown in FIG.
And B as request objects (an object having a feature of AND between the features of A and B as a request object), a search key and the like are automatically generated and a search is performed. As C,
Assume that objects are output in the order of E, D, A, and B. However, for the user, there are cases where the required objects A and B and the object E or D closer to the required object are more necessary objects than the highest-ranked C.

FIG. 4 shows an example of the contents of the search unit 32 (black box part) which is automatically processed by the system in order to clarify the reason why such a contradiction occurs. The search unit 32A in FIG. 4 expresses two feature amounts of the searched objects A, B, C, D, and E in a two-dimensional manner.

In the figure, if a user designates A and B as request objects, the generated search key is indicated by a cross. Since the similarity is higher as the distance from the search key is shorter, the similarity calculation result in this case indicates that the similarity is higher in the order of C, E, D, A, and B as described above.

This is due to the fact that the request object or search parameter originally given to the system by the user was not appropriate. That is, the positions of the request object A and the request object B in the xy space of FIG. 4 are separated in both the x direction and the y direction. This means
The request object A and the request object B do not have a common feature as far as the feature amounts x and y are concerned. As described above, the characteristic of the request object A and the characteristic of the request object B which do not have the common characteristic
To search for an object featuring ND, the feature vector of the request object A and the request object B
Even if the average κ with the feature amount vector is used as a search key, the search key κ has characteristics that are not similar to the request object A and the request object B. Therefore,
Even if you search for an object that is close to the search key κ,
The object obtained by the search is an object having neither the characteristics of the request object A nor the characteristics of the request object B. For this reason, in the search shown in FIGS. 3 and 4, the object C that is not similar to any of the request objects A and B is searched in a higher order than the objects E and D.

Although the conventional database search method has the above-mentioned problem, the search unit 32 is a black box, and the user can avoid the above-mentioned problem since the user cannot know the automatic generation process in the black box. It is difficult to implement a database search method.

An object of the present invention is to solve the above problems,
An object of the present invention is to provide a database search method and an apparatus therefor that enable a search system to generate appropriate search parameters so that an object requested by a user can be obtained easily and quickly.

[0020]

In order to achieve the above object, a database search method performs a primary search of a database using a search sentence generated with initial search parameters.
The primary search result obtained by the primary search is displayed to the user, the user inputs a selection condition for the displayed primary search result, and based on the input selection condition, A database search method for calculating a second search parameter and performing a second search using a search sentence created based on the second search parameter, wherein the first search result is determined according to the second search parameter. Is changed and displayed to the user, and the user looks at the primary search result whose display format has been changed, and determines whether the second search parameter is appropriate for obtaining the desired object. Is determined, the selection condition is corrected and re-entered, a third search parameter is calculated based on the input selection condition, and the display format of the primary search result is changed according to the third search parameter. Change it and show it to the user, The user looks at the primary search result whose display format has been changed, determines whether or not the third search parameter is appropriate for obtaining the desired object, and repeats this to finally determine the user Performs a secondary search of the database with a search sentence created using search parameters determined to be appropriate to obtain the desired object.

In this database search method, one coordinate axis on the display surface is associated with one feature amount of each object, and the value of the one feature amount is associated with coordinates on the one coordinate axis axis. Are associated with different coordinate axes of the display surface, and the p feature amounts of the object are associated with the coordinates of the vector space defined by the p coordinate axes of the display surface.
The feature of the object with respect to the individual feature amounts is expressed on a display surface, and further, the correspondence between the value of the one feature amount and the coordinates on the one coordinate axis is determined with respect to the set feature amount. The similarity of one feature quantity is defined so as to correspond to the coordinates on the one coordinate axis.

Since the display surface is two-dimensional, the number of features displayed on the display surface is two or three. Therefore, when there are four or more feature quantities, the feature quantity displayed on the display surface is selected as follows. When the vector space defined on the display surface is called a feature amount space, when the object searched by the primary search is characterized by n feature amounts, the object is p-dimensional (p <n). When displaying in the feature amount space, a p-dimensional feature amount space composed of the most important p feature amounts of the n feature amounts is displayed on the display surface. Thus, the most important p of the objects retrieved by the primary search
Are displayed as the coordinates of the object in the p-dimensional feature amount space, and the user looks at the arrangement of the object displayed on the display surface and sets the search key based on the current search parameters. It is determined whether or not the position in the feature amount space is within a desired distance range on the display surface with respect to the position of the object designated by the user as the request object, and is not within the desired distance range. Then, the user determines that the search parameter is not appropriate for obtaining a desired object, and further changes the selection condition to a new selection condition including a logical operation.

At present, which feature value is important for an intended search is determined as follows. In other words, among the objects searched by the primary search, the p feature amounts are assigned to the most important p items in ascending order of the variation in the values of the feature amounts that characterize the plurality of required objects specified by the user as selection conditions. It is determined as individual feature quantity. Further, the above-mentioned variation in the value of each feature can be evaluated by the standard deviation of the value of each feature for a plurality of required objects.

Whether the search parameters are appropriate or not is determined by
The determination can be made based on whether or not the search key generated based on the search parameter is similar to the request object. For this purpose, the number of features that characterize an object is n, and the similarity evaluation function F for each feature is ｛f
¹ , f ² ,..., F ⁿ }, and weights W = {w ¹ , w ² ,
... when established in proportion to w ^n} on the standard deviation, the similarity d between each object O and the search key K set by the user are evaluated by the following equation.

[0025]

(Equation 3)

Here, k ⁱ and o ⁱ (1 ≦ i ≦ n) are the i-th feature amounts of the search key K and the object O. F ⁱ (k ⁱ , o ⁱ ) is a similarity evaluation function between the feature quantity k ⁱ and the feature quantity o ⁱ .

The database search apparatus of the present invention has a database, initial search condition input means, selection condition input means, search sentence creation means, database search means, and search result display means. The database stores objects to be searched. In the initial search condition input means, a user sets and inputs an initial search condition. The selection condition input means allows a user to input an object selection condition for an object retrieved as a result of the primary search.
The search sentence creating means inputs an initial search condition and generates a search sentence based on the initial search parameter during the primary search,
While searching for appropriate search parameters for secondary search,
Generate hypothesis search parameters based on the selection criteria,
When the user determines that an appropriate search parameter has been generated, a search sentence based on the formal search parameter is generated. The database search means executes the search statement to search the database and outputs object information of the searched object. The search result display means inputs the search result of the primary search as object information and displays the object information according to the search parameters.

When searching for an object, the database search means performs a primary search of the database using a search sentence generated based on the initial search parameters, and the search result display means displays the result of the primary search in the initial search. It is displayed to the user in a display format corresponding to the parameter, and the user selects the sorting condition so that an object similar to the feature amount of the requested object is searched by visually checking the displayed primary search result. Change and input to the search sentence creating means via the selection condition input means, the search sentence creation means calculates the hypothesis search parameter from the changed selection condition and outputs it to the search result display means, and displays the search result The means changes the display format of the primary search result in accordance with the hypothesis search parameter and displays the primary search result to the user. Judge whether the object is appropriate to search for the object to be searched.If not, correct the selection condition from the selection condition input means and re-enter it. The database is secondarily searched using a search sentence created based on the hypothesis search parameter determined to be present.

The search sentence creation means inputs the feature value of the primary searched object from the search result display means to determine the standard deviation for each feature quantity, and for each feature quantity, increases in descending order of the standard deviation. It has the function of giving weight. The search result display unit evaluates the similarity of each feature amount between the object searched by the primary search and the search key set based on the hypothesis search parameter input from the search sentence creation unit. A similarity evaluation function of a scalar quantity is calculated by a predefined operation of a feature quantity described in a vector format of the object and a feature quantity described in a vector format of a search key, and the calculation result and the From the weights of the respective feature amounts, the similarity for each feature amount between the object and the search key is calculated,
The similarity is used as a one-dimensional display of the feature amount of the object, and each of a predetermined number of feature amounts having a large weight is associated with a coordinate axis on the display surface, and the similarity degree for each feature amount is displayed on the display surface. It is displayed as coordinates on a coordinate axis corresponding to the feature amount.

The search result display means has a cache memory for storing the object information searched by the database means. With this cache memory, it is possible to secure the real-time property necessary for the user to specify the search condition through trial and error.

[0031]

According to the present invention, it is not necessary for the user to set difficult search parameters by using the objects searched by the initial search, and the object can be intuitively understood by "selection of objects". It is intended to generate search parameters for a secondary search.
Here, “easily understood by humans intuitively” means “displayed on the display surface”.

For this purpose, first, the features of the object searched by the initial search are displayed on the display surface. However, since the display surface is a two-dimensional space, the number of feature quantities that can be easily displayed on the display surface is two. Alternatively, the number of feature quantities that can be expressed is three at most even when expressed three-dimensionally. (Hereinafter, the number of feature amounts displayed on the display surface is referred to as p.) Therefore, when the features of an object characterized by more than p feature amounts are displayed on the display surface, the features characterizing the object are displayed. An important feature amount is selected from the amounts, and the feature of each object is displayed on the display surface for the selected feature amount.

In this case, which feature value is important is
The determination is made as follows. Among the objects searched by the initial search (primary search), a plurality of objects similar to the object requested by the user are selected as the requested objects. Then, for the plurality of required objects, the variation (or the standard deviation) of the feature value is calculated for each feature value. In this way, the larger importance (weight) is given in the order of smaller variation.

In this way, a p-dimensional feature quantity space is constructed by associating the p feature quantities with the largest weights with the p coordinate axes on the display surface. Can be displayed on the display surface.

Normally, the content of the feature value (feature value) is described by a vector, and this vector is called a feature vector. However, in the present invention, one feature amount is associated with one coordinate axis on the display surface (that is, since each feature amount value is displayed as coordinates on one coordinate axis), each feature amount is represented. It is necessary to represent not a multidimensional vector format but a one-dimensional vector, that is, a scalar format. In the present invention, the feature amount vector corresponds to a feature amount criterion corresponding to the coordinate origin (first embodiment) or a similarity to a search key (second embodiment) set based on a search parameter. Let me. Therefore, the smaller the distance between two points in the feature space is, the more similar important features of the object displayed by the two points are. In this way,
Important features of the object searched by the initial search can be intuitively displayed on the display surface.

The user changes search parameters by inputting various selection conditions while looking at the display screen, and a search key automatically generated on the display screen based on each search parameter, Whether the current search parameter is appropriate or not is determined depending on whether or not the distance (similarity) with the request object specified above is within a predetermined range. Then, when the distance between the search key and the request object specified by the user on the display surface is within a predetermined range, the search parameter at that time is set to 2 as an appropriate search parameter.
Used to create the next search sentence.

[0037]

Embodiments of the present invention will be described with reference to the drawings.

FIG. 5 is a block diagram of an embodiment of the database search apparatus according to the present invention. FIG. 6 is a flowchart illustrating the operation of the present invention. FIG. 7 is a diagram illustrating a display example based on the initial search parameters. FIG. 8 is a diagram showing a display example based on hypothesis search parameters. FIG.
FIG. 14 is a diagram illustrating an example of a next search result.

The database search device of the present invention includes an object search device 1, an initial search condition setting device 2, a display device 3, and a selection condition setting device 4. The object search device 1 searches for an object stored in the database for a search key specified by a user or an initial search key presented as a reference when the system is started. The object search device 1 also searches for an appropriate search parameter for the secondary search according to the selection condition input by the user, and performs a secondary search based on the search parameter determined by the user to be appropriate. The initial search condition setting device 2 sets a search key specified by the user or an initial search key presented as a reference by the search system. The display device 3 displays a search result searched by the object search device 1.

The selection condition setting device 4 inputs the selection conditions of the user for the object obtained as a result of the search. As the selection condition, a logical operation condition such as AND / OR between a plurality of selection conditions can be used. Further, it is possible to instruct or input a search result object displayed on the display by using a pointing device (cursor key, mouse, etc.). For example, with respect to an object displayed on a display, an object search apparatus can easily and intuitively understand a selection condition such as "similar" or "dissimilar" to an object requested by a user. 1 can be entered. Also, an AND or OR condition can be specified between each selection condition. As a result, more flexible search condition specification is possible.

The object search device 1 includes an initial search condition input unit 5, a selection condition input unit 6, a search sentence creation unit 7, a database search unit 8, a database 9, a search result display unit 10, and a cache memory 11. .

The initial search condition input unit 5 functions as an interface for inputting the initial search conditions output from the initial search condition setting device 2 and inputting the conditions to the search sentence creation unit 4.

The selection condition input unit 6 functions as an interface for inputting selection conditions output from the selection condition setting device 4 and inputting the conditions to the search sentence creation unit 4.

At the time of the initial search, the search sentence creating unit 7 applies an average initial search parameter to the search key specified by the user or the initial search key presented as a reference by the system (at the initial stage, the feature amount A primary search sentence is generated (to set the weight of each to an average value), and after the initial search, a search parameter and a search sentence are automatically generated from the input selection condition. As will be described in detail in connection with the operation of the present embodiment described later, the search sentence creating unit 7 of the present invention provides a hypothesis search parameter and a Formal search parameters and search sentences for secondary search are automatically generated. The search sentence creating unit 7 also calculates the average of the standard deviations of the feature amounts of all the objects stored in the database at the time of the initial search (see the second embodiment).

The database search unit 8 includes a search sentence creation unit 7
The database is accessed to search for an object based on the search sentence generated in step (1). The database 9 stores objects to be searched.

The search result display unit 10 stores the object information searched by the database search unit 8 in the cache memory 11 and displays the object information on the display device 3 based on the hypothesis search parameters generated by the search sentence creation unit 7. To display to the user. At this time, for example, two or three types of feature amounts that can optimally express the search parameters on the display device are selected and displayed on the display device 3 in a two-dimensional display.

As a result of the primary search, the feature amount of the selected object is displayed on the planar display surface by arranging the objects (positions of the objects or relative positions of the objects).
When displaying as, the feature amount space displayed by the display surface is not selected so as to optimally display the search condition specified by the search parameter, and the user is required to display the object on the display surface. By looking at the arrangement, it is not possible to intuitively understand how much each object satisfies the search condition.

For example, if an object has a color, shape, size,
Consider a case in which the user is characterized by the feature amount of price and intends to search with emphasis on shape and size. In this case, when displaying the result of the primary search on the display device, if the feature amount of each object is displayed on the display surface using the color and price on the xy coordinate axes of the display surface, The arrangement only indicates the similarity of the objects evaluated from the attributes of color and price, and does not indicate the similarity of the shape and size requested by the user with the search parameters.

In contrast, if the features of each object are displayed on the display surface using the shape and size as the coordinate axes of the display surface, the arrangement of the objects on the display surface is referred to as shape and size. Since the similarity of the object evaluated from the attribute is represented, the similarity of the shape and the size requested by the user by the search parameter is displayed. Thus, the display surface is a two-dimensional space,
The number of independent features that can be displayed on the display surface is two, and at most three. Therefore, when the object is characterized by four or more features, a feature space representing the most important search condition among the search conditions requested by the user by the search parameter is displayed on the display surface. Is necessary to determine the optimal search parameter for secondary search.

As described above, when there are a plurality of feature values,
It is necessary to convert each feature amount into a one-dimensional numerical value (an amount represented by coordinates on one coordinate axis) and select a feature amount to be displayed so as to optimally express the search parameter.
(Refer to the operation of the first and second embodiments). Also, Kohon
en T. , "The Self-Organizi
ng Map ", Proc. IEEE, 1990, Vo.
l. 78, no. 9, pp. The reduced dimension can be displayed using the self-organizing map shown in 1464-1480.

The search result display section 10 operates as follows. First, the object information searched by the database search unit 5 is stored in a cache memory 9 (described later), and the hypothesis search parameters created by the search sentence creation unit 4 are received in real time. The hypothesis search parameters are created based on the selection conditions input via the selection condition input unit 9. Search result display section 10
Then, when a secondary search is performed using the hypothesis search parameter, the user determines what the tendency of the secondary search is, whether the AND condition or the OR condition is appropriate to specify, and the like. Thus, information necessary for creating a new search sentence for the secondary search can be appropriately presented to the user on the display device 3.

The cache memory 11 holds primary search object information. The search result display unit 1 uses the primary search object information stored in the cache memory 9 when presenting the above search parameter information. As a result, it is possible to secure the real-time property necessary for the user to specify the search condition through trial and error.

As a first embodiment of the present invention, the present invention will be described in the case of accessing a database storing objects characterized by three feature values x, y, and z. Here, the three feature quantities x, y, and z are, for example, three attributes such as a color, a shape, and a dimension that characterize a certain object. Normally, each feature amount is expressed as a feature amount vector by expressing a plurality of independent contents (feature amount values) in a vector format. For example, the feature amount color is expressed in a vector format by evaluating the red, green, and blue components included in the color of the object, such as (a numerical value of red, a numerical value of green, a numerical value of blue), and digitized. You. The shape is expressed in a vector format, for example, (symmetry, ratio of sides not specified by symmetry, angle not specified by symmetry). The dimension is expressed in a vector format, for example, (the length, direction, and weight of a specific side).

However, in the present invention, one coordinate axis of the display surface is made to correspond to one characteristic amount of each object in order to display the feature amount of the primary searched object in a two-dimensional space called a display surface. Therefore, each feature value is set to 1
Expressed as a numerical value of the dimension. Then, the numerical value of each feature is
Correspond to the coordinates on the corresponding coordinate axis.

Since the method of converting the vector display of each feature value into a scalar (one-dimensional numerical value) display is described in detail in the second embodiment, in this embodiment, each feature value is represented by one. The case where the dimension is represented by a numerical value will be described.

The first embodiment is implemented as follows using the database search device of FIG. A search key specified by the user via the initial search condition setting device 2, or
The search sentence creating unit 7 applies an average initial search parameter to the initial search key presented as a reference by the system to generate a primary search sentence.
Next search is performed (step S1 in FIG. 6).

The search result display unit 10 displays five objects A among the objects obtained by the primary search.
(9, 9, 4), B (1, 3, 8), C (8, 6,
5), D (2, 8, 4), and E (3, 4, 7). Where the number in parentheses for each object is
Each feature value of the object is a scalar (or 1
(Dimensional numerical value).

The obtained primary search result is displayed on the display device 3 by the search result display unit 10 based on the initial search parameters (step S2). In the initial search parameters of the present embodiment, x and y are emphasized as the feature amount vectors, and the display to the user is as shown in FIG.

Next, the user inputs a selection condition such as "similar" or "dissimilar" to the request object using the mouse in the displayed primary search result using the mouse. To enter. Further, the user designates an AND or OR condition between a plurality of selection conditions (step S).
3). In the present embodiment, the user selects A and B as “similar” objects, and inputs an AND condition for both in the sense that “I want an object that is similar to both”.

This input information is passed to the search sentence creation unit 7 via the selection condition input unit 6, and the search sentence creation unit 7 determines which feature amount of the object is most important to the user. Then, a search parameter corresponding to the input information is generated and passed to the search result display unit 10 as a hypothesis search parameter (step S4).

Comparing the feature values of the objects A and B with the display of FIG. 7, the variation of the feature value x of the objects A and B is 9-1 = 8. The variation in the value of the feature y of the objects A and B is 9-3 = 6, and the variation in the value of the feature z is | 4-8 | = 4. That is, it can be seen that z, y, and x are in ascending order of variation (variance). This means the following.

When the user specifies a common feature for the objects A and B as the feature of the request object, usually, the object A, B
We are not asking for an object similar to B. For example, if the user is requesting an AND of objects A and B, where the object is characterized by x = color, y = shape, z = size, the user may, for example, have a shape and size Require that objects be similar to objects A and B, but do not care about color. In such a case, as a result of the primary search, the user selects one of the objects displayed on the display device.
The fact that the objects A and B are selected means that the shape and size of the objects A and B are similar to the request object from the viewpoint of the user. Therefore, the feature value of the shape of the object A and the feature value of the shape of the object B have substantially the same value. As a result, the variation in the feature value of the shape should be small. The feature value of the size of the object A and the feature value of the size of the object B have substantially the same value.
The variation in the magnitude feature value should be small.

In other words, when the variation of each feature value of a plurality of objects is calculated, a feature value with a small variation in feature value (for example, a shape) is calculated.
Means that the objects are similar. As described above, the variation in the feature amounts of the plurality of objects can be a measure for measuring the similarity of the plurality of objects with respect to the feature amounts.

When this is applied to the objects A and B selected by the user as the required objects, the user obtains the objects A and B with respect to the feature amounts y and z.
Request for an object similar to
Is not an important feature for the user.

As described above, the variation in the characteristic values of the objects A and B increases in the order of z, y, and x.
It can be said that the evaluation of the similarity between the objects A and B specified by the user indicates that the similarities of the feature amounts y and z are more important than the similarity of the feature amount x. Therefore,
Hypothesis search parameters are more y and z than initial search parameters.
An important one is generated (step S4).

The display of the primary search result is changed in the search result display section 10 based on the hypothesis search parameters (step S5). In the hypothesis search parameter of the present embodiment,
In the feature amount vector, y and z are emphasized, and the display to the user is as shown in the upper part of FIG.

With this display, the user can judge the quality of the hypothesis search parameter (step S6), and can correct and re-input the selection condition and the AND / OR condition (step S3).

Whether the hypothesis search parameter is good or bad is determined by
It is performed as follows. At this stage, since the AND search is instructed, the search key is obtained by averaging A and B, and is indicated by K. If you search in this state,
The search sentence is as shown in the middle part of FIG. 8 and the similarity is in the order close to K, so the search result is as shown in the lower part of FIG. Even if y and z, which have relatively small variations in the feature vector values, are used, the distance between A and B is large on the display surface, and even if the search is performed as it is, the similarity of C is the highest. You can see that.

Here, the user does not perform an AND search, but
By switching to the R search, it can be understood that it is better to perform a search that makes the similarity between D and E higher than C. As a result of repeating re-input, the hypothesis search parameter finally obtained is used as a formal search parameter, a secondary search sentence is generated in the search sentence creating unit 7 (step S7), and a secondary search is performed in the database search unit 8. (Step S8).

The obtained secondary search result is displayed on the search result display section based on the formal search parameters (step S9). As a result of switching to the OR search, the display to the user is as shown in the upper part of FIG. (In the figure, the dotted line extending obliquely upward is a perpendicular bisector between the position of the object A and the position of the object B, and represents a similar area in the OR search.) The search sentence is as shown in the middle part of FIG. The degree is A
Alternatively, since the search order is closer to B, the search result is as shown in the lower part of FIG.

Thus, the user can easily obtain an object closer to the requested one. The user determines whether or not there is a desired object among the objects displayed on the display surface (step S1).
0). If there is no desired object among the objects displayed on the display surface, the processing from step S3 to step S10 is repeated. If the desired object is found, the database search process ends.

Next, the second method of the database search method of the present invention will be described.
An embodiment will be described. The first embodiment described above is a database search method in the case where each feature is described in a scalar. The database search method in this embodiment is a database search in which each feature is described in a vector. Is the way. The method of the present embodiment is also executed according to the processing flow shown in FIG. 6 using the database search device of FIG.

The database has three feature values.

[0074]

(Equation 4)

An object O having
The database search device accesses this database.

Now, as a primary search result, the following seven objects are displayed on the display device 3.

[0077]

(Equation 5)

Was obtained. Among these, the user visually selects A, B, and C as “similar”, and A, B,
The selection conditions specified for all of C in the AND conditions are input to the selection condition setting device 4.

Based on the input information, the search sentence creating unit 7 sets a set of search key feature vector K =
{K ^x , k ^y , k ^z } and a weight W = {w ^x , w ^y , w ^z } for each feature amount are created.

First, a set of search key feature vector K =
{K ^x , k ^y , k ^z } is the object A,
It is created by averaging the feature amount vectors of B and C.

[0081]

(Equation 6)

Is obtained.

Next, the weight W = {w ^x , w ^y , w ^z } for each feature amount
Compares the degree of variation of the feature amount vectors of the objects A, B, and C with the degree of variation of the feature amount vectors of all the objects stored in the database.
It is created by judging that feature values with small variations are regarded as important. The degree of variation can be compared using the standard deviation.

For example, n in the characteristic amount j = x, y, z
The average of the standard deviation for each feature of objects

[0085]

(Equation 7)

Can be obtained. (Σ in the root of equation (5) is the sum of objects O (that is, objects A, B, and C).) Using this equation (5), the standard deviation of objects A, B, and C And the average of the standard deviations of all the objects stored in the database, and the one with the smallest ratio can be the one with a small degree of variation in the feature amount vector. The calculation of these standard deviations is executed in the search sentence creating unit 7. If the degree of variation is small in the order of the feature amounts x, y, and z, a large value is given to w ^x and a small value is given to w ^z .

The search parameters generated in this way are:
It is passed to the search result display unit 10 as a hypothesis search parameter. When the search result display unit 10 displays the object on the display device 3, the characteristic vectors of the object are displayed on the two-dimensional display surface using some function in order to display the characteristics of the object using the coordinate axes for each characteristic amount. It must be integrated (expressed as a scalar) into a one-dimensional numerical value (a numerical value that can be expressed as coordinates on one coordinate axis). This numerical value is called the similarity of each feature.

The similarity d ^x , d ^y , d ^z between each characteristic amount of the object O and the characteristic amount of the search key K is represented by the Manhattan distance function.

[0089]

(Equation 8)

In the above example,

[0091]

(Equation 9)

Is obtained. The calculation of Expressions (6) and (7) is executed by the search result display unit 10. The symbol of the form Ad ^{x in} the equation (7) indicates the similarity between the feature amount x of the object A and the feature amount x of the search key K. In order to simplify the discussion, in calculating the similarity of Expression (7), all the weights to be given to the respective feature amounts by Expression (5) are calculated as 1.

If the degree of variation of the feature vector is small in the search parameters, that is, if the weight W is large, the feature x and y are displayed as shown in FIG. FIG. 10 is a diagram illustrating the Manhattan distance between the search key K and each object.

In the present embodiment, from the objects displayed on the display surface by the primary search, a search key having the characteristics of the object selected by the user as the requested object is defined according to the search parameters. Next, the variation of the feature value for each feature value of the selected object is determined based on the standard deviation, and a large weight is given to the feature value having a small standard deviation value.

Next, the similarity d between the feature amount of each object and the retrieval key is calculated for each feature amount according to the equation (1). Thereby, each feature amount usually described in a vector format is described by a one-dimensional numerical value. As a result, m
The features of an object characterized by a plurality of feature amounts (for example, color, shape, size, etc. in the case of an image) are m similarities d ¹ , d ² , d ³ ,. described by d ^m .

Now, let p be the dimension number of the feature space defined on the display surface to display the important features of the object searched by the primary search (p = 2 in this embodiment).
In this case, the feature amount space is constituted by the top p feature amounts in ascending order of the standard deviation value. Then, the features of the primary searched object are displayed on the display surface (on the feature space). In the display, for the p most important features of the request object selected by the user, the similarities of the p feature amounts of all the primary searched objects are displayed as coordinates on the display surface. .

In the embodiment shown in FIG. 10, the feature values that the user values (the standard deviation is small) are x and y, and the feature value x of each object and the feature value x of the search key K are similar. degrees d ^x, similarity similarity d ^y pairs (d ^x, d ^y) between the feature y of the feature y and search key as the coordinates, an important feature amount of each object is displayed. In FIG. 10, the objects A, B, and C are located close to each other and close to the search key (coordinate origin). Therefore, it is determined that the search parameters at this time are relatively appropriate. However, whether or not this search parameter can be used for the secondary search depends on whether or not the distance (similarity) between the objects A, B, and C and the search key is within a predetermined range. Determined by If the distances (similarities) between the objects A, B, and C and the search keys are not within a predetermined range, the user further updates the selection condition and sets a new search condition. , The object and the search key are displayed on the display surface. Such processing is repeated until the similarity between the objects A, B, and C and the search key falls within a predetermined range.

In this manner, the user does not need to set difficult search parameters, and the system gives appropriate search parameters by a search instruction by "selection of objects" which is intuitive and easy for the user to understand. be able to.

Also, when the user inputs the selection condition, it is possible to clarify which objects are “similar” and which objects are “dissimilar”,
It is possible to reduce the number of trial and error of the user.

Lastly, the database search device of the present invention operates under the control of an information processing device (not shown). The information processing device controls the database search device according to a control program recorded on a recording medium (not shown).

The control program includes a primary search of the database using the search sentence generated by the initial search parameters, and displays the primary search result obtained by the primary search to the user, Is input to the primary search result in which is displayed, a second search parameter is calculated based on the input selection condition, and a search sentence created based on the second search parameter is input. Describes a procedure for executing a database search process for executing a secondary search using. This program causes the database search device to change the display format of the primary search result according to the second search parameter and display it to the user, and the user can view the primary search result whose display format has been changed. Then, it is determined whether or not the second search parameter is appropriate for obtaining a desired object. If the selection condition is corrected and re-input, the third search is performed based on the input selection condition. The parameters are calculated, the display format of the primary search result is changed according to the third search parameter and displayed to the user, and the user looks at the primary search result whose display format has been changed and looks at the third search result. It is determined whether or not the search parameters are appropriate for obtaining a desired object, and the selection condition is corrected and input. When the selection condition is input, the fourth search parameter is calculated based on the input selection condition. This Repeating final user uses the search parameters judged to be appropriate in order to obtain the desired object to create a search statement describes a procedure for searching a secondary database by the search statement.

[0102]

As described above, in the database search method and apparatus according to the present invention, the features of the objects searched by the primary search are displayed on the display surface using the feature amounts considered important by the user as coordinate axes. The user visually judges the similarity between the objects regarding the important feature amount and the similarity between the object and the search key based on the distance on the display surface, and determines whether the search parameter is appropriate or not. Has the following effects. 1) The user himself / herself does not need to set difficult search parameters, and appropriate search parameters can be provided to the system by a search instruction based on “selection of objects” which is intuitive to humans. 2) When the user inputs the selection condition, it is possible to clarify which object is a "similar" object and which object is a "dissimilar" object. The number of trial and error can be reduced.

The database search device of the present invention uses a cache memory for displaying objects, thereby reducing the number of queries to the database which requires computational cost during the trial and error stage in which the user inputs selection conditions. Can be speeded up.

[Brief description of the drawings]

FIG. 1 is a diagram for explaining a flow of a conventional “database search method in which a user generates a search sentence by himself”.

FIG. 2 is a diagram for explaining a flow of a conventional “database search method in which a system generates a search sentence by a user inputting selection conditions”.

FIG. 3 is a diagram for pointing out a problem of a conventional “database search method in which a system generates a search sentence by a user inputting selection conditions”.

FIG. 4 is a diagram showing what kind of processing is performed in a black box part in FIG. 3 and explaining the cause of the problem of the conventional method.

FIG. 5 is a diagram illustrating an embodiment of the present invention.

FIG. 6 is a flowchart illustrating the operation of the present invention.

FIG. 7 is a diagram illustrating a display example of a primary search result based on an initial search parameter according to the present invention.

FIG. 8 is a diagram illustrating a display example of a primary search result based on a hypothesis search parameter according to the present invention.

FIG. 9 is a diagram illustrating a display example of a secondary search result using a formal search parameter according to the present invention.

FIG. 10 is a diagram showing a Manhattan distance between a search key K and each object.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 object search device 2 initial search condition setting device 3 display device 4 selection condition setting device 5 initial search condition input unit 6 selection condition input unit 7 search sentence creation unit 8 database search unit 9 database 10 search result display unit 11 cache memory 31 exemplary object storage unit 32, 32A search unit 33 display unit

 ────────────────────────────────────────────────── ─── Continuing from the front page (72) Inventor Masashi Yamamuro 2-3-1 Otemachi, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation F-term (reference) 5B075 ND06 NK06 PP02 PP03 PP13 PP23 PP28 PQ02 PQ14 PQ36 PQ46 PR06 QM08 QP10

Claims (10)

[Claims] 1. A primary search of a database using a search sentence generated by an initial search parameter, a primary search result obtained by the primary search is displayed to a user, and the user is displayed. A selection condition is input to the primary search result, a second search parameter is calculated based on the input selection condition, and a second search parameter is calculated using a search sentence created based on the second search parameter. In a database search method for executing a next search, a display format of a primary search result is changed and displayed to a user according to the second search parameter, and the user is provided with a primary search result having a changed display format. , It is determined whether the second search parameter is appropriate for obtaining a desired object, and the selection condition is corrected and re-input, and the third selection parameter is determined based on the input selection condition. Search parameters Is calculated, and the display format of the primary search result is changed according to the third search parameter and displayed to the user. The user looks at the primary search result whose display format has been changed and sees the third search result. If the search parameters are correct to get the desired object,
It is characterized by performing a secondary search of the database with a search sentence created using search parameters determined to be appropriate for finally obtaining a desired object by repeatedly determining whether or not the database is a desired object. How to search the database you want. 2. A method according to claim 1, wherein one coordinate axis of the display surface is associated with one feature amount of each object, and the value of said one feature amount is defined as:
Corresponding to the coordinates on the one coordinate axis axis, different feature quantities characterizing the object are associated with different coordinate axes on the display surface, and p feature quantities of the object are
The feature of the object relating to the p feature amounts is represented on the display surface in correspondence with the coordinates of the vector space defined by the p coordinate axes of the display surface, and further, the value of the one feature amount and 2. The method according to claim 1, wherein the correspondence with the coordinates on one coordinate axis is defined such that the similarity of the one feature to the set feature corresponds to the coordinates on the one coordinate axis. 3. The described method. 3. When the vector space defined on the display surface is referred to as a feature value space, when the object searched by the primary search is characterized by n feature values, the object is p-dimensional. (P <n) When displaying in the feature amount space, a p-dimensional feature amount space composed of the most important p feature amounts of the n feature amounts is displayed on the display surface, and a primary search is performed. The most important p feature quantities of the object searched for by are displayed as the coordinates of the object in the p-dimensional feature quantity space, and the user visually observes the arrangement of the object displayed on the display surface. The position of the search key set based on the current search parameter in the feature amount space is determined by the position of the object specified by the user as the request object. It is determined whether or not it is within a desired distance range on the display surface. If not, the user determines that the search parameter is not appropriate to obtain a desired object. 3. The method according to claim 2, further comprising the step of changing the selection condition to a new selection condition including a logical operation. 4. The method according to claim 1, wherein, among the objects searched by the primary search, the p feature amounts are arranged in ascending order of variation in the value of each feature amount characterizing a plurality of request objects specified as a selection condition by a user. 4. The method according to claim 3, wherein the most important p features are determined. 5. The variation in the value of each feature value is a ratio of the standard deviation of the value of each feature value for the plurality of required objects to the standard deviation of the value of each feature value for the entire object stored in the database. The method according to claim 4, wherein the evaluation is performed by: 6. The number of features that characterize an object is represented by n
And the similarity evaluation function F for each feature amount is denoted by ｛f¹, f^Two, ... f
ⁿ、, and weight W = の w for each feature amount¹, w^Two,… WⁿIn front of｝
When set in proportion to the standard deviation,
Between the search key K and each object O set based on
The similarity is d, and the i-th search key K and object O
Kⁱ, Oⁱ(1 ≦ i ≦ n)
Quantity k ⁱAnd feature oⁱLet f be the similarity evaluation function betweenⁱ(Kⁱ,
oⁱ), The degree of similarity is expressed by the following equation:The method according to claim 5, wherein the evaluation is performed by: 7. A database storing objects to be searched, an initial search condition input means for setting and inputting initial search conditions by a user, and a user searching for an object searched as a result of the primary search. Selection condition input means for inputting selection conditions of an object, and at the time of primary search, inputting initial search conditions, generating a search sentence based on the initial search parameters, and searching for appropriate search parameters for secondary search A search sentence generating means for generating a hypothetical search parameter based on a selection condition and generating a search sentence based on a formal search parameter when the user determines that the proper search parameter has been generated; Database search means for executing a statement to search a database and outputting object information of the searched object; Search result display means for inputting a search result of the next search as the object information and displaying the object information in accordance with the search parameter, and when searching for an object, the database search means performs a search based on the initial search parameter. Primary search of the database using the search sentence generated by
The result of the primary search is displayed to the user in a display format according to the initial search parameter, and the user looks at the displayed primary search result and searches for an object similar to the feature amount of the requested object. In such a manner, the selection condition is changed and input to the search sentence creation means via the selection condition input means, and the search sentence creation means calculates the hypothesis search parameter from the changed selection condition and displays the search result. The search result display means changes the display format of the primary search result in accordance with the hypothesis search parameter and displays it to the user. The search condition is judged to be appropriate or not. If not, correct the selection condition from the selection condition input means and re-enter it. User database retrieval apparatus characterized by searching the secondary databases using the search statement generated based on the hypothesis search parameters judged to be appropriate. 8. The search sentence creating means inputs a feature amount value of a primary searched object from a search result display means to obtain a standard deviation for each feature amount, and adds a standard deviation of the standard deviation to each feature amount. The search result display means assigns a large weight in ascending order, and the search result display means sets a feature amount between an object searched by the primary search and a search key set based on a hypothesis search parameter input from the search sentence creation means. In order to evaluate the similarity, a similarity evaluation function of the scalar amount is calculated by a predefined operation of the feature amount described in the vector format of the object and the feature amount described in the vector format of the search key. And calculating the similarity between the object and the search key for each feature amount from the calculation result and the weight of each of the feature amounts, and calculating the similarity degree for the object. In the one-dimensional display of the feature amount of the object, each of a predetermined number of feature amounts having a large weight is made to correspond to a coordinate axis on the display surface, and the similarity for each feature amount is corresponded to the feature amount on the display surface. The apparatus according to claim 7, wherein the coordinates are displayed as coordinates on a coordinate axis. 9. The apparatus according to claim 7, wherein the search result display means has a cache memory for storing the object information searched by the database means. 10. A primary search of a database using a search sentence generated by an initial search parameter, a primary search result obtained by the primary search is displayed to a user, and the user is displayed. A selection condition is input to the primary search result, a second search parameter is calculated based on the input selection condition, and a second search parameter is calculated using a search sentence created based on the second search parameter. A recording medium for recording a program describing a procedure for executing a database search process for executing a next search, wherein the program controls an information processing apparatus to control the database search apparatus, The display format of the primary search result is changed according to the search parameter and displayed to the user, and the user looks at the primary search result whose display format has been changed and sets the second search parameter to the desired object. For proper in order to obtain the-object and to determine whether to re-enter and modify the selection condition,
A third search parameter is calculated based on the input selection condition, the display format of the primary search result is changed according to the third search parameter, and displayed to the user. By checking the changed primary search result to determine whether the third search parameter is appropriate for obtaining the desired object, and correcting and inputting the selection condition, the input selection A fourth search parameter is calculated based on the selection condition, and the fourth search parameter is repeated. Finally, the user is caused to create a search sentence using the search parameter determined to be appropriate for obtaining a desired object, and to perform the search. A recording medium in which a program for causing a secondary search of a database by a sentence is recorded.