JPH0276033A - Knowledge arrangement supporting device - Google Patents

Abstract

PURPOSE:To reduce the burden of knowledge arrangement work for a user by giving the whole relation of each knowledge as spatial expression to the user as preprocessing when the user arranges knowledge. CONSTITUTION:The preprocessing of the knowledge arrangement, which is executed by the user, is executed an the relation between the respective knowledges is displayed as spatial distribution. Namely, a classification processing part 12 obtains the relation between the respective knowledges 34 and classifies the knowledge based on similarity concerning the respective knowledges 34 to be decided in a similarity decision processing part 11 and a display processing part 51 displays a result to be obtained in the classification processing part 12. A classifying editor 52 has editing function to execute the classification by the user to observe the space distribution in the secondary space of the respective knowledges 34 to be displayed in the display processing part 51. Namely, the user observes the temporary classified result, which is displayed in the display processing part 51, by the classification processing part 12 and executes the confirmation, correction and classification of this result.

Description

[Detailed Description of the Invention] [Summary] The present invention relates to a knowledge organization support device that assists a user in organizing knowledge in a knowledge base such as an expert system.

The purpose is to pre-process the knowledge organization performed by the user and display the relationships between each piece of knowledge as a spatial distribution.

a similarity determination processing unit that determines the similarity between each piece of knowledge; a classification processing unit that determines and classifies the relationship between each piece of knowledge based on the similarity; and the relationship between each piece of knowledge that has been found. The present invention is configured to include a display processing unit that displays the above-mentioned knowledge as a spatial distribution, and a classification editor that performs classification of the respective knowledge displayed as the spatial distribution.

[Industrial application field]

The present invention relates to a knowledge organization support device, and more particularly to a knowledge organization support device that supports a user in organizing knowledge in a knowledge base such as an expert system.

With the recent increase in the scale of knowledge bases in expert systems, user interfaces are required to be easy to organize knowledge. Here, knowledge organization refers to the work of classifying knowledge.

[Conventional technology]

In order to make it easier to organize knowledge even in large-scale knowledge bases, it has conventionally been done to provide users with knowledge bases expressed in table format.

i.e. by the editor.

It provides tables as a framework for expressing knowledge, making it easier to organize knowledge.

FIG. 8 is an explanatory diagram of the prior art.

3 is a knowledge base (table), 31 is an event, 32 is a hypothesis, 33 is an area expressing a causal relationship, and 34 is one piece of knowledge.

The knowledge base 3 includes an event 31 representing an observed situation;
A causal relationship between the event 31 and a hypothesis 32 representing a candidate conclusion drawn from the event 31 is expressed. For this purpose, in each region 33, the presence or absence of a causal relationship between the corresponding event 31 and hypothesis 32 is expressed by the presence or absence of a circle mark.

The user understands that the cause-and-effect relationship has been expressed (knowledge has been input).
The knowledge base 3 is looked at to see if there is any relationship between the pieces of knowledge 34. Then, the knowledge is organized based on the relationship between each piece of knowledge 34. At this time, knowledge base 3
Since the information is expressed in a table, it is easy to examine the relationship between each piece of knowledge 34.

In addition, the basic unit when organizing knowledge, that is, knowledge 34, is 1
This is a set of causal relationships (hypothesis set) between each of the hypotheses 32 regarding the two events 31.

[Problem to be solved by the invention]

According to the prior art described above, a table (knowledge base 3 expressed in a table) as a framework for organizing knowledge is provided to the user.

However, this only provides an expression format that makes it easier to examine and organize the relationships between each piece of knowledge 34, and the actual organization (classification work) of the knowledge has to be carried out by the user.

Therefore, organizing the knowledge required when constructing the knowledge base 3 or adding knowledge to the knowledge base 3 has been a heavy burden on the user.

SUMMARY OF THE INVENTION An object of the present invention is to provide a knowledge organization support device that performs preprocessing for knowledge organization performed by a user and displays relationships between pieces of knowledge as a spatial distribution.

[Means to solve the problem]

FIG. 1 is a diagram illustrating the principle of the present invention and shows a knowledge organizing support device according to the present invention.

In FIG. 1, 11 is a similarity determination processing section.

12 is a classification processing unit, and 3 is a knowledge base (table).

31 is one event, 32 is one hypothesis, 33 is an area expressing a causal relationship, 34 is one piece of knowledge, 51 is a display processing unit, and 52 is a classification editor.

The knowledge base 3 includes an event 31 representing an observed situation;
It consists of a table expressing a causal relationship between an event 31 and a hypothesis (conclusion hypothesis) 32 representing a candidate conclusion. For this purpose, the presence or absence of a causal relationship between the corresponding event 31 and hypothesis 32 is expressed by the presence or absence of an O mark, for example, in each of the 95 areas 33.

The unit of knowledge when organizing knowledge, that is, one piece of knowledge 34, is
A set of hypotheses 32 about one phenomenon 31 (hypothesis set)
and includes one event 31 and all the regions 33 corresponding to it. Therefore, there are as many pieces of knowledge 34 as there are events 31.

The similarity determination processing unit 11 determines the relationship between each piece of knowledge 34! ! 134. This similarity is the “hypothesis to be narrowed down 3” between any two pieces of knowledge 34.
It is defined as ``the commonality of two types.'' That is, any 2
Hypothesis 32 that common knowledge 34 has a causal relationship
It is assumed that it depends on the number of

The classification processing unit 12 determines (classifies) the relationship between each knowledge 34 based on the similarity of each knowledge 34 determined by the similarity determination processing unit 11.
4, the distance between each piece of knowledge 34 is determined. The higher the similarity, the closer the distance is by 1. As a result, the relative positional relationship of each piece of knowledge 34 is determined. This relative positional relationship is expressed as a spatial distribution of each piece of knowledge 34 in a 12-dimensional space. That is, - this relative positional relationship is converted into an expression as a two-dimensional spatial distribution by applying a statistical method 1, for example, multivariate analysis, to the similarity. Therefore, the two-classification processing unit 12 determines the relationship between each piece of knowledge 34 in a two-dimensional space and classifies it into three.

The display processing section 51 displays the results obtained in the 1-classification processing section 12. That is, the relationship between each piece of knowledge 34 is displayed as a spatial distribution in a two-dimensional space. Therefore, knowledge 34
are distributed in a two-dimensional space with distances maintained according to the results of multivariate analysis.

The classification editor 52 has an editing function that allows a user who views the spatial distribution of each piece of knowledge 34 displayed on the 1-display processing unit 51 to classify it in a two-dimensional space. That is, the user selects the classification processing section 1 displayed on the 1 display processing section 51.
Look at the results of the tentative classification according to 2, confirm them, correct them, and classify them into 1.

[Effect]

FIG. 2 is an explanatory diagram of the operation of the present invention, and shows the flow of knowledge organization processing according to the present invention.

■ A knowledge base 3 represented as a table as shown in FIG. 1 is prepared. [(The Iu nature determination processing unit 11.

The similarity between each piece of knowledge 34 in the knowledge base 3 is determined according to the above definition (obtained by calculation).

For example, the similarity S between knowledge iJA and B. Search for.

Similarly, Sllc+ and S1+e are determined.

(6) For example, the similarity S in which the classification processing unit 12 performs multivariate analysis on the knowledge 34 based on the similarity thereof and converts it into an expression as a spatial distribution of each knowledge 34 in a two-dimensional space. , S, c and SkC, knowledge A,
The distance #do in the two-dimensional space between B is determined, and dac and abc are similarly determined.

(2) Display processing 51 displays the classification results in the two-classification processing section 12. At this time, each knowledge 34 is displayed on the display screen in the two-dimensional space using the distance between each knowledge in the two-dimensional space as is.For example, as shown in FIG. 1, knowledge A.

B and C are distances Hd-b, d-c in two-dimensional space
and dbc.

■ The user who viewed each displayed knowledge is classified into one category editor 5.
0 to perform classification work (organizing) such as classification and correction using 2.
For example, since knowledge 11iA and C are located close to each other, they are classified as one group.

Furthermore, if it is determined as a result of referring to the knowledge base 3 that it is preferable to classify knowledge A and knowledge B as one group, they will be classified in this way.

Here, if we focus on the former classification task, the user will know everything! Classification can be performed simply by looking at the relationship between It34 displayed as a spatial distribution. That is, each piece of knowledge 34 is classified in advance by the eight classification processing unit 12 as a preprocessing for knowledge organization performed by the user. Therefore, in this classification work, the only work that the user actually performs is to instruct the classification editor 52 on the knowledge to be classified, and this burden is light.

On the other hand, the latter classification work is a user's correction of the classification performed by the two-classification processing unit 12.

As described above, according to the third aspect of the present invention, it is possible to reduce the burden of knowledge classification (organization) work on the user.

〔Example〕

FIG. 3 is a block diagram of an embodiment, showing an expert system.

In FIG. 3, 1 is a knowledge organization support processing section, 2 is a knowledge editor, 4 is an inference mechanism, and 5 is an interface.

+a+ The configuration knowledge editor 2 is for editing the knowledge base 3, and has a function of representing the knowledge base 3 in a table.

The knowledge organization support processing section 1 includes a similarity determination processing section 11 and a classification processing section 12, and performs pre-processing of knowledge organization when editing the knowledge base 3, that is, processing for expressing the distribution of each piece of knowledge 34 in a two-dimensional space. I do.

The inference mechanism 4 uses the edited knowledge base 3 to perform inference in response to the event 31 input by the user, and selects a conclusion from a set of hypotheses 32 having five causal relationships.

The interface 5 is for smooth interaction between the user and the expert system. interface 5
includes a 1 display processing unit 51 and a classification editor 52, and when editing the knowledge base 3, performs 1 classification work by expressing the distribution of each knowledge 34 in a two-dimensional space, and at the time of inference, the ink face 5 The event 31 is taken in and the conclusion (hypothesis 32) is output.

Cb) An example of determining the similarity between knowledge The determination of the similarity between the respective pieces of knowledge 34 performed by the similarity determination processing unit 11 will be explained.

The similarity between any two pieces of knowledge A and B is determined by a linear equation.

H(XnY) H(XLIY) X: Hypothesis set of knowledge A Y: Hypothesis set of knowledge B H(Z) : Number of hypotheses in hypothesis set Z S (X, Y)
: Similarity H(XnY) between knowledge A and knowledge B is a hypothesis 32 that there is a common causal relationship between the event 31 of knowledge A and the event 31 of knowledge B in the knowledge base 3 expressed as a table.
is the number of H(XLIY) is the sum of the number of hypotheses 32 that have a causal relationship with the event 31 of knowledge A and the number of hypotheses 32 that have a causal relationship with the event 31 of knowledge B.

Note that the causal relationship between event 31 and hypothesis 32 is expressed as a percentage (
%) for example 1.

It is sufficient to consider that there is a causal relationship in 50% or more cases.

Further, it is also possible to set several values of % that are considered to have a causal relationship, obtain information about a plurality of similarities, and apply multivariate analysis to each of them to obtain a plurality of different spatial distribution representations.

(Example of C1 Multivariate Analysis Multidimensional scaling (MDS) will be described as an example of multivariate analysis performed by the classification processing unit 12.

(a) Overview According to MDS, the similarity relationship of the entire 5 can be known from the information of the tUQl relationship between each element (each knowledge 34 in this case).

Mapping elements onto coordinates (in this case, two-dimensional space) is called "configuration," and by configuring them while correctly reflecting the order relationship of similarity (in this case, similarity), the results obtained can be It will reflect the real world. Therefore, "the greater the similarity, the closer the mapped distance"
It is necessary to guarantee the monotonicity of this relationship.

However, in reality, monotonicity is rarely completely guaranteed and errors are often included, so the portions that deviate from monotonicity are corrected. The order relationship regarding the distance between each element obtained by this correction is called disparity, and a configuration that minimizes the degree of misfit between this and the original order relationship is determined.

The configuration with the smallest degree of incompatibility within the range that maintains monotonicity between similarity and distance is determined as the final configuration.

(B) The similarity Sjk between the configuration method elements a, b, c, and d is shown in Figure 4 (A).
As a result of mapping the relative positional relationship between these elements a to d into a two-dimensional space, the fourth
Assume that it is placed as shown in Figure (B).

In this case, the inter-point distance djk between each element is shown as shown in Fig. 4 (C), and the correlation between the three-point distance dJk and the similarity Sjk is shown as shown in Fig. 4 (D). .

As can be seen from FIG. 4(D), there is a relationship between the l'I similarity Sjk and the point-to-point distance djk: ``If the similarity Sjk is large, the point-to-point distance djk is small.''

From this.

Sca>Sab>S, b>Sag>Sdc>Scb
(If it is 11.

all(<ah,<a.<a,,<dc4<db
It can be said that the positions (arrangements) of the four elements are determined so that c (21).

(1) Let the number of formulation elements be n, the number of dimensions of the configuration be m, and the p-th dimension coordinates of a point (point j) representing a certain element j be Xjp.

At this time, let dJ& be the distance between the two points j that constitute the configuration embedded in the 2m-dimensional Euclidean space.

It is. Using this equation (3), find the path Wrdtk between net points whose magnitude order corresponds to equation (1) as shown in equation (2) for nC, assortatives (distribution 1). That is, the point-to-point distance dJk for any two of the n elements is determined, and this is performed for all combinations.

However, the configuration 1 obtained in this way, ie.

Even if the configuration satisfies equations (2) and (3), - it is not determined arbitrarily.

Therefore, the degree of conformity (degree of unconformity) of cloth 1 is defined together with the configurations that do not satisfy equation (3).

For this purpose, we first introduce disparity.

In the configuration that satisfies the equation (3), as shown in Figure 4 CD).

Each point decreases monotonically (distance between points djk of the configuration is monotonically decreasing with similarity Sjk). Similarity of the determined configuration Sjk
The disparity determines the degree of conformity, that is, the extent to which the point-to-point distance djk of the configuration satisfies the monotonically decreasing relationship with the similarity Sjk.

The disparity is defined for each point-to-point path Ra Jk and is always #-key decreasing with the similarity Sjk.

That is, if the disparity between element j and element j is expressed as d, then the disparity corresponding to equation (2) is.

'fac<dba<d. ＜Ca＜ fca＜♂b
c (satisfies 41 and is defined as linear.

(i) For the point-to-point distance djk that decreases with the similarity S, , the point-to-point distance dj remains the same as the disparity 6j
Let it be k.

(ii) Similarity Sjk and point-to-point distance dj that is not monotonically decreasing
Regarding k, the average value in the range that is not monotonically decreasing is set to a disparity of 4.

Now, suppose that there is a configuration as shown in FIG. 5(A) that does not satisfy equation (3). At this time, the distance between 5 points d ac + d□, d
Regarding CD and dbc, the distance between the 9 points becomes the disparity as it is due to (i) above.

AC-d 11c+ ♂am =d ld+ic
a-d car fbc=d bcr. On the other hand, the point-to-point distances db4 and d. about.

From (11) above, the disparity is:

d ha= 6-b= (d ba+ d-b) /
2 (5).

Figure 5 (A) is corrected using equation (5) and is shown in Figure 5 (
B), that is, the point-to-point distance d4. The difference between and disparity (dotted line in Figure 5 (B)) is.

(3) This is the amount of correction required to convert the point-to-point distance that does not satisfy 2 into a disparity that satisfies equation (4).

Next, the degree of incompatibility with respect to the degree of similarity S and k is determined.

The degree of incompatibility is expressed by the sum of the squares of the point-to-point path 11tt[dJh and the disparity jk. For n elements.

Assuming that the degree of incompatibility with respect to the degree of similarity in configuration is 31.

S"=Σ Σ(d, ni-cr, k)2- (6)j>
It is k.

However, if the arrangement is scaled up or down without changing the relative positions of the elements, the degree of unfitness S0 in equation (6) also changes.

Therefore, first-order normalization is performed. Consider the sum of squares T0 of the differences from the average value of nC distances. Here, d is the average value of nCz distances.

Σ Σ d, k J-1k-1 cz It is. The degree of unfitness S2 is It is expressed as

As described above, although the previously obtained configuration was not uniquely determined, one optimal configuration is obtained by setting the configuration that minimizes the degree of incompatibility S2 as the final configuration.

If m=2, the configuration in two dimensions, ie.

A spatial distribution of elements in two-dimensional space is obtained.

At this time, we obtain a tentative configuration using equations (1) or 3),
One configuration that minimizes the degree of incompatibility is obtained using equations (4) to 7).

(X) Knowledge organization (classification) processing flow FIG. 6 shows the knowledge organization processing flow when editing the knowledge base 3 in the expert system shown in FIG.

■ For users to edit Knowledge Base 3.

Start Knowledge Editor 2.

The knowledge editor 2 provides the user with a table framework in order to represent the knowledge base 3 in a table.

The user enters events 31 and hypotheses 32 in each area 33 of the table.
Enter the causal relationship between That is.

Enter knowledge 34.

■ After inputting knowledge, the user must:
The knowledge organization support processing section 1 is activated.

■ Knowledge organization support processing unit 1 ([similarity determination processing unit 11
) calculates the similarity between each piece of knowledge 34 in the knowledge base 3.

■ Knowledge organization support processing unit 1 (classification processing unit 12).

Using MDS (multivariate analysis), the overall relationship (relative positional relationship) of each piece of knowledge 34 is determined from the similarity between pieces of knowledge 34 . At this time, the relative positional relationship of each person 1a34 is 2
It is used to indicate relative positional relationships in dimensional space.

That is, the 114g1 nature of each knowledge 34 is expressed as a spatial distribution in a two-dimensional space.

(2) The interface 5 (display processing unit 51) displays each piece of knowledge 34 (knowledge A to knowledge 0) on a display screen that is a two-dimensional space, as shown in FIG. 7(A).

■ The user uses the interface 5 (classification editor 52)
Classify (organize) knowledge through .

The knowledge 34 is classified into four groups as shown in FIG. 7(B).

■ The user corrects the classification as necessary while referring to the knowledge base 3 or the actual knowledge content.

〔Effect of the invention〕

As explained above, according to the present invention, when the user organizes the knowledge, as a preprocessing, the overall relationship of each piece of knowledge is provided to the user as a spatial representation, so that the user can roughly understand the knowledge. It is possible to automatically organize information, reduce the burden of knowledge organization work on the user, and shorten the time required for one work.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the principle configuration of the present invention. FIG. 2 is an explanatory diagram of the operation of the present invention. FIG. 3 is a configuration diagram of an embodiment. FIG. 4 and FIG. 5 are explanatory diagrams of MDS. Figure 6 shows the knowledge organizing process flow. FIG. 7 is a diagram showing an example of a display. FIG. 8 is an explanatory diagram of the prior art. 11 is a similarity determination processing section, and 12 is a classification processing section. 3 is a knowledge base (table), and 31 is one event. 32 is one hypothesis, 33 is an area expressing causal relationship, 3
4 is one piece of knowledge, 51 is a display processing unit, and 52 is a classification editor.

Claims (1)

[Claims] A similarity determination processing unit (11) that determines the similarity between each piece of knowledge (34), and determines and classifies the relationship between each piece of knowledge (34) based on the similarity. a classification processing unit (12); a display processing unit (51) that displays the relationship between each of the obtained pieces of knowledge (34) as a spatial distribution of each piece of knowledge (34); A knowledge organizing support device comprising a classification editor (52) for classifying each piece of knowledge (34).