JP2591751B2 - Similarity judgment method between objects - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a similarity determination method between objects, and particularly to a similarity determination method suitable for an expert system or the like which is one of artificial intelligence applications.

[Conventional technology]

As a method of determining the similarity between objects, if the object is a pattern such as a character, it has been conventionally performed to determine the similarity using various pattern matching methods. However, there are few methods suitable for determining the similarity between conceptual objects, such as various objects targeted by an artificial intelligence application system such as an expert system.

[Problems to be solved by the invention]

In recent years, various expert systems applying various results obtained from research on artificial intelligence have been put into practical use.
In such an expert system, it is often necessary to execute inference for solving a problem while determining the similarity between many objects belonging to a problem area. For example, assuming an apparel product planning expert system in the fashion industry, the following objects can be considered as objects belonging to the problem area.

(1) Objects belonging to the concept group of the customer class (cluster) Carrier elegance, sensitive carrier,
Elegance Traditional, etc. (2) Objects belonging to the styling theme concept group Asphalt, Cafe Society, Conflict, etc. And, for example, when the customer class is inferred as carrier elegance, styling most similar to the characteristics of the carrier elegance When inferring a theme, it is necessary to determine the similarity between the carrier elegance and asphalt belonging to the styling theme.

Realizing the similarity determination between objects as described above by an artificial intelligence application system such as an expert system involves various difficult problems. One is what data is used to determine similarity, and the other is how those data are retained. It is also important how to determine the similarity using the data. Furthermore, since the object is conceptual, how to easily notify the user of the object determined to be most similar by the similarity determination and the next and subsequent candidates, etc. It is a difficult problem.

The present invention has been made in view of such circumstances, and an object of the present invention is to provide a method suitable for determining the similarity between conceptual objects handled by an artificial intelligence application system.

[Means for solving the problem]

In order to achieve the above object, the present invention provides a method for determining a similarity between a plurality of objects, comprising: image storage means for storing image information representing each of the objects; and determining the similarity between the plurality of objects. The evaluation score for each object for each of a plurality of items determined in advance as a comparison item is set as a characteristic value sequence in a characteristic slot of a frame corresponding to each object, and stored in the image storage means. A frame storage means for storing a plurality of frames in which the image number of the own object is set in the image slot; and a comparison purpose based on similarity of shape and similarity of area when the characteristic value sequence is virtually graphed. Rule storage means for storing a rule for determining the degree of similarity between the object and the object to be compared. By executing a rule stored in the rule storage unit, a comparison target object most similar to the comparison target object is determined from among a plurality of comparison target objects, and image information of a frame of the frame of the object is determined as information regarding the determination result object. Is configured to output an image corresponding to the image information stored in the image storage means to the display device in accordance with the image number set in the display device.

In a preferred embodiment of the present invention, the rule is:
The content of determining the similarity between objects is determined by the ratio between the intersection area and the union area of the graph when the characteristic value sequence is virtually graphed, and at least one of the comparison target objects When the characteristic value differs from the value of the same characteristic of the object to be compared by a predetermined value or more, a condition part is provided that does not make the object to be compared a comparison candidate for similarity determination using the virtual graph.

[Action]

When a certain object is set as a comparison target object and some other objects are set as comparison target objects, a comparison is made between a characteristic value sequence previously stored in a characteristic slot in a frame of the comparison target object stored in the frame storage means, Based on the characteristic series previously stored in the characteristic slot in the frame of the target object, the similarity determination rule stored in the rule storage unit is executed, and the graph shape when both characteristic value columns are virtually graphed is displayed. The similarity between objects is determined based on the similarity and the similarity of the area.

〔Example〕

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram of an expert system for implementing the similarity determination method of the present invention, and shows an example in which the present invention is applied to an apparel product planning expert system. The expert system includes a knowledge base 1, a knowledge base management unit 2, an inference unit 3, an image information storage unit 4,
It comprises a commentary storage unit 5, a judgment result display control means 6, a user interface 7, and an input / output device 8 having a display device 8a such as a graphic display and an input device 8b such as a mouse and a keyboard.

The knowledge base 1 includes a frame storage unit 10, a rule storage unit 11, and a knowledge storage unit 12 for storing other knowledge. The frame storage unit 10 includes a frame group 10a belonging to a certain concept group such as a customer group, and a frame group belonging to another concept group such as a styling theme group.
10n etc. are stored. The rule storage unit 11 stores a plurality of rules required for solving a problem, and includes a similarity determination rule 11a.

The knowledge base management means 2 is the inference means 3 and the knowledge base 1
The inference means 3 uses the frames, rules, and the like obtained through the knowledge base management means 2 to solve the problem by referring to the frames, rules and other knowledge. Perform inference.

The commentary storage unit 5 stores the concept of the object determined by the similarity determination as a sentence in the display device 8a of the input / output device 8.
Stores the commentary data 5a to 5m required to be displayed in.
In addition, simply displaying the concept of the object determined by the similarity determination to the user using only the commentary statement is not enough for the user to understand. Therefore, an image that best represents the concept of the object is also displayed on the display device 8a. I do. Such images are stored in the image information storage unit 4 as image information 4a to 4m.

The judgment result display control means 6 relates to a certain object which is the result of the similarity judgment obtained by the inference means 3 by inference,
The commentary data and the image information are extracted from the commentary storage unit 5 and the image information storage unit 4, respectively, and control is performed when the commentary data and the image are displayed on the display device 8a via the user interface 7.

The user interface 7 includes an input / output device 8 and inference means.
3. An interface with the determination result display control means 6 is provided.

FIG. 2 shows an example of the contents of a frame stored in the frame storage unit 10. The frame shown in FIG. 2A is a carrier elegance frame (eg, frame 10a1 in FIG. 1) belonging to the customer group, and the frame shown in FIG. 2B is an asphalt frame (eg, FIG. 1) belonging to the styling theme group. 10n1). Each frame has an instance of that frame (INSTANC
E), a class (CLASS) type, and a plurality of slots in which instance values are set. Then, FIG.
In the sensibility slot of the frame, the characteristic value sequence (52111234) of carrier elegance is stored, (image ID 104) is stored in the image slot, and (description ID
205) is stored. In addition, the sensitivity slot of the frame in FIG.
4) is stored, (image ID 150) is stored in the image slot, and (explanatory text ID 246) is stored in the comment slot. Here, the image IDs 104 and 150 are numbers indicating the image information corresponding to carrier elegance and asphalt among the plurality of image information in the image information storage unit 4,
The commentary IDs 205 and 246 are numbers indicating the commentary corresponding to carrier elegance and asphalt among the plurality of commentary data in the commentary storage unit 5. Furthermore, the characteristic value column (52
(111234) and (42111324) are evaluation scores given to the carrier erengance and asphalt for each of eight predetermined items as comparison items when determining the similarity between a plurality of objects.

FIG. 3 is a diagram in which a characteristic value sequence stored in a feeling slot of each frame of an object is virtually graphed.
In this embodiment, a total of eight common characteristics as described below are given as evaluation items to each object belonging to different concepts such as a customer group and a styling theme group.

Sophisticated characteristics (characteristic of wanting to be an urban woman) Elegance character (characteristic of wanting to be an elegant woman) Romantic type (characteristic of wanting to be a cute woman) Ethnic character (characteristic of wanting to be a simple woman) Country Type (character that wants to be a free woman) Active type (character that wants to be a bright woman) Mannish type (character that wants to be an independent woman) Modern type (character that wants to be an intelligent woman) Figure 3 Is an octagonal graph, which indicates that the characteristics are stronger as the distance from the center point increases, and the degree of the strength is represented by a numerical value. In the example shown in the figure, the characteristic value sequence is (13545223).

Similarity determination rule 11a stored in rule storage unit 11
Is a rule for determining the similarity between the comparison target object and the comparison target object based on the similarity of the shape and the similarity of the area of the virtually generated graph as shown in FIG. As a specific method for realizing this, the similarity between objects is determined based on the ratio between the intersection area and the union area of the graph. Also,
A condition is added that, when a certain characteristic of the comparison target object has a high score and the same characteristic of the comparison target object has a low score, or vice versa, it is not a candidate for similarity calculation. FIG. 4 shows an example of the content of such a similarity determination rule 11a. As shown in the figure, the similarity determination rule 11a includes a condition part and an execution part,
The condition unit determines that a certain characteristic value of the comparison object is “5” and a corresponding characteristic value of the comparison object is “1”, or that a certain characteristic value of the comparison object is “1” and If the corresponding characteristic value of the comparison target object is “5”, a condition is set that the execution unit is not executed. It is also described that the execution unit calculates the ratio of the product set area to the union set area as the similarity between the comparison target object and the comparison target object. Here, the intersection area is, for example, as shown in FIG. 5, the virtualized graph of the comparison object is indicated by a solid line 50 in FIG. 5, and the virtualized graph of the comparison target object is indicated by a broken line 51 in FIG. If it is shown, it corresponds to the area of the shaded portion (the overlapping portion of the portion surrounded by the solid line 50 and the portion surrounded by the broken line 51) in FIG.
In addition, the union area is defined by the dotted part (solid line) in FIG.
(A portion obtained by performing a logical sum of a portion surrounded by 50 and a portion surrounded by a broken line 51).

FIG. 6 shows an example of display contents of the display device 8a. In this example, the object of the styling theme having the characteristic value sequence most similar to the characteristic value sequence assigned to carrier elegance, which is one of the customer demographic groups, and the characteristic value sequence obtained by modifying the characteristic value sequence as necessary is described above. 3 objects (asphalt with a CF of 53%, cafe society with 51% of the same, conflict with 49% of the same) were obtained by the similarity judgment, and a commentary on the asphalt with the highest certainty CF and a description of the asphalt An example is shown in which a style image that best represents a concept is displayed on the same screen.

Hereinafter, the similarity determination operation in the expert system shown in FIG. 1 will be described with reference to the drawings.

In the apparel product planning expert system according to the present embodiment, first, a customer group targeted for product planning is selected by inference by the inference means 3 using various parameters from the input / output device 8 and various knowledge in the knowledge base 1. Is done.
Next, assuming that the selected customer class is, for example, carrier elegance, the characteristic value sequence of the sensibility slot shown in FIG. The styling theme having the most similar characteristic value sequence is determined by inference using the similarity determination rule 11a by the inference means 3.

In this case, the inference means 3 is a rule storage unit of the knowledge base 1.
Similarity determination rules from 11 through the knowledge base management means 2
11a, and then sequentially read the objects belonging to the styling theme group of the frame storage unit 10, the characteristic value sequence assigned to the read object, and the characteristic value sequence of carrier-elegance as the comparison object. To execute a similarity determination process using the similarity determination rule 11a. At this time, as described above, when a certain characteristic of the comparison object and the same characteristic of the comparison object are values that are far apart from each other, the similarity calculation using the virtual graph is not performed, and the process proceeds to the next comparison object. When the condition set in the condition part is satisfied, the similarity between the two objects is obtained from the ratio of the product set area and the union set area of the virtual graph, as described with reference to FIG. Keep it. The similarity determination as described above is performed for all objects in the styling theme group stored in the frame storage unit 10. When the similarity calculation for all the objects is completed, a predetermined number (three in the case of FIG. 6) of objects are obtained in order from the most similar object, and the commentary ID and the image ID for those objects are obtained.
From each frame, and notifies the determination result display control means 6 as a determination result.

Upon receiving the above notification from the inference unit 3, the judgment result display control unit 6 reads out the commentary data corresponding to the notified commentary ID from the commentary storage unit 5, and displays the commentary of the most similar object on the front. Are displayed on the right side of the screen of the display device 8a as shown in FIG. Further, the image information storage unit 4 is searched using the image ID corresponding to the most similar object among the image IDs notified from the inference means 3 as a key, and displayed as shown in FIG. 6 according to the obtained image information. An image is displayed on the left side of the screen of the device 8a.

The user can know the contents of the object of the styling theme determined by the expert system by looking at the commentary and the image displayed on the display screen 8a. Then, the user visually inspects the commentary and image of the styling theme presented as the most similar, and if it is judged inappropriate, switches to the commentary and image of the next candidate styling theme. Are input from the input device 8b, and the styling theme of the next candidate is verified. Further, feedback is performed to reflect the result of the verification to the knowledge base 1 and the like as necessary.

In the above embodiment, the present invention is applied to an apparel product planning expert system. However, the similarity determination method between objects of the present invention is not limited to such an expert system, and can be applied to all artificial intelligence application systems. Of course there are.

〔The invention's effect〕

According to the similarity determination method between objects of the present invention described above, the following effects can be obtained.

(1) Since a predetermined number of items are used as comparison items when determining the similarity between a plurality of objects, the similarity determination process is standardized. That is, if the number of items indicating the characteristics of each object is provided by the number corresponding to the concept of each object, the characteristics of the object can be accurately expressed, but the similarity determination processing among a plurality of objects becomes different, In particular, the process of determining the similarity between objects having different concept groups becomes complicated, but by limiting the number of comparison items common to a plurality of objects as in the present invention, it is possible to use a plurality of objects using the same determination rule. It is possible to determine the similarity between different objects.

(2) Since the evaluation score for the comparison item when determining the similarity of each object is stored as a characteristic value sequence in the characteristic slot of the frame of the object, the correspondence between each object and the characteristic value sequence of each object is determined. It can be easily performed, and the management of the characteristic value sequence is also facilitated.

(3) Since the comparison items of each object are shared,
The similarity determination between objects can be determined based on the similarity of the shape and the similarity of the area when the characteristic value sequence is virtually graphed, thereby simplifying the similarity determination process. In particular, when a characteristic value sequence is virtually graphed, a method of determining the similarity between objects based on the ratio of the intersection area and the union area of the graph can be adopted, and the similarity determination can be performed at high speed. Can be.

(4) When notifying the user of the most similar object obtained by the similarity determination or other candidates,
Since an image that well represents the characteristics of the object is displayed on the display device, it is possible to notify the user of the conceptual object easily.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a diagram showing an example of the contents of a frame, FIG. 3 is a diagram showing an example of contents of a virtual graph, and FIG. FIG. 5 is an explanatory diagram of a similarity determination method using a virtual graph, and FIG. 6 is a diagram illustrating an example of display contents of a display device 8a. In the figure, 1 ... knowledge base, 8 ... input / output device 2 ... knowledge base management means, 8a ... display device 3 ... inference means, 8b ... input device 4 ... image information storage unit, 10 ... Frame storage units 4a to 4m: image information, 10a, 10n: frame group 5: commentary storage units, 10a1, 10n1: frames 5a to 5m: commentary data, 11: rule storage unit 6 ... Judgment result display control means, 11a ... Similarity judgment rule 7 ... User interface, 12 ... Other knowledge storage unit

 ──────────────────────────────────────────────────の Continued on the front page (72) Inventor Yoshinori Hayami 5-33-1, Shiba, Minato-ku, Tokyo Within NEC Corporation (72) Inventor Isao Narukawa 1-1-11 Minamihonmachi, Higashi-ku, Osaka Teijin Limited Inside System Technology (72) Inventor Toshihiro Satomi 1-1-11 Minamihonmachi, Higashi-ku, Osaka City Inside Teijin System Technology Co., Ltd.

Claims (3)

(57) [Claims] 1. A method for determining a degree of similarity between a plurality of objects, comprising: image storage means for storing image information representing each object; and a comparison item for determining the degree of similarity between the plurality of objects. The evaluation score for each object for each of the plurality of predetermined items is set as a characteristic value sequence in the characteristic slot of the frame corresponding to each object, and the image number of the own object stored in the image storage means is set. Frame storage means for storing a plurality of frames set in the image slot, a comparison object and a comparison object based on similarity of shape and similarity of area when the characteristic value sequence is virtually graphed And a rule storage unit for storing a rule for determining the similarity of the rule. By executing the rule,
Determining a comparison target object most similar to the comparison target object from among the plurality of comparison target objects,
The image corresponding to the image number set in the image slot of the frame of the object as the information on the object of the determination result is output to the display device an image corresponding to the image information stored in the image storage means. A method for determining the similarity between featured objects. 2. The similarity determination method between objects according to claim 1, wherein the rule is that an intersection area and a union area of the graph when the characteristic value sequence is virtually graphed. The similarity between objects is determined based on the ratio between the two. 3. The similarity determination method between objects according to claim 2, wherein the rule is that at least one characteristic value of the comparison target object differs from a corresponding characteristic value of the comparison target object by a predetermined value or more. A similarity determination method between objects, comprising a condition part that does not make the comparison target object a comparison candidate for similarity determination using the virtual graph.