JP2621818B2 - Knowledge base verification device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a knowledge base verifying apparatus for verifying the correctness of a knowledge base storing the knowledge of experts on such problems as classification, diagnosis, and examination.

[0002]

2. Description of the Related Art Conventionally, the following are known as a knowledge base verification device.

(1) As described in the literature of Ishikawa, Tanaka, Yamao et al. (“Knowledge Verification Support System Using Hypothesis Inference”, Journal of the Japanese Society for Artificial Intelligence, 7: 1, 1992, pp. 105-116) Device that prepares test data in advance and compares the inference result using the knowledge base with the data to verify the correctness of the knowledge base.

(2) M. Suwa, A .; C. Scott
and E. H. Shortlife et al. (“And approach to Verifyyi”)
ng Completeness and Consi
stability in a Rule-Based Ex
part System ", AI Magazine,
Fall, 1982, pp. As described in 16-21), an apparatus that identifies all possible cases using only the structure of a knowledge expression and checks whether knowledge exists for each case.

(3) As described in Japanese Patent Application Laid-Open No. 2-113338, knowledge is described in a format of IF to THEN, and is converted into a positional cube format at the time of verification, and logical knowledge of the positional cube is obtained. A device that detects inconsistencies between newly added knowledge and existing knowledge and detects lack of knowledge by simple operations.

[0006]

However, the conventional knowledge base verification apparatus (1) using test data has a problem that sufficient data must be collected in advance before verification. It was difficult to collect test data to check the data evenly.

In the device (2) for comprehensively examining all possible cases using the structure of the knowledge expression, the number of possible cases increases as the knowledge base increases, and the verification cost increases. There was a problem.

On the other hand, the knowledge base verification device of (3) does not require special test data and does not exhaustively check all possible cases as in (2), so that verification is not possible. There are few problems in terms of cost. However, the knowledge base is described in a format such as IF-THEN-
Since the process of converting to the cube format is required, there remains a problem that the verification cost increases accordingly. Furthermore, when new knowledge is added, it is configured to verify that the added knowledge does not contradict existing knowledge. Therefore, the flow of verification is to add one knowledge and verify. It becomes repetition. Therefore, it is difficult to collect and store a certain number of knowledge at a time and collectively verify it.

[0009] In any of the prior arts, the knowledge base creator who has determined that there is an error must perform all the work of correcting the knowledge base. There is a problem that correction is not easy.

The present invention has been proposed in view of such circumstances, and its purpose is to express knowledge to be verified in a table format in advance, and to contradict knowledge by a logical operation in the table format. By verifying the shortage, knowledge stored in the knowledge base storage means can be collectively verified at high speed without requiring special test data. Is to make it easy for the user to make corrections.

[0011]

SUMMARY OF THE INVENTION A knowledge base verification apparatus according to the present invention is a knowledge base verification apparatus for verifying the correctness of a knowledge base describing expert's knowledge. A positive case storage unit that stores a positive case that expresses the conditions in the case where is an answer in a tabular format, and a table that describes the candidates for the answer in the target problem and the conditions when the candidate should not be the answer in a table format. A knowledge base storage unit having a negative case storage unit for storing the described negative cases, and a knowledge inconsistency in the knowledge base stored in the knowledge base storage unit. A consistency verification unit for verifying, and a lack of knowledge in the knowledge base stored in the knowledge base storage means, in a table format for positive cases and negative cases. A knowledge base verifying unit having an integrity verifying unit for verifying by a logical operation, and outputting information on inconsistency or lack of knowledge found by the knowledge base verifying unit to an input / output device, and according to an input from the input / output device, Knowledge base modifying means for modifying the knowledge base. Then, the integrity verification unit includes:
Values of other attributes except one attribute in negative cases
Means for integrating negative cases, all of which are the same,
After answering, all answer candidates in the knowledge base
Means for converting the expression into an expression based on
In the fixed case, except for one attribute, the values of other attributes
Means for integrating positive cases that are all the same, and the integration
From the universe
Means and the knowledge base if the result is null.
Judge that there is no lack of knowledge in the
Judgment that there is lack of knowledge and the condition of the lacking case
Means for seeking .

[0012]

[Function] Expresses in a tabular format the answer candidates for the target question, the attributes characterizing each answer candidate, the attribute conditions when each candidate is an answer, and the attribute conditions when each candidate should not be an answer. The obtained knowledge base is stored in the knowledge base storage means.

When this knowledge base is supplied to the knowledge base verification means, the consistency verification unit performs a logical operation in the form of a table on positive cases and negative cases constituting the knowledge base, and outputs the answer described in the knowledge base. It detects whether or not there is a "knowledge inconsistency" in which a plurality of candidates are answered simultaneously. If there is a contradiction, an attribute condition when the contradiction occurs is obtained. on the other hand,
The completeness verification unit detects a “lack of knowledge” in which none of the candidates is satisfied as a solution by a logical operation in a tabular form for the positive case and the negative case. Then, if there is a lack of knowledge, what kind of attribute condition the knowledge is lacking is determined.

If the knowledge base has inconsistency or lack, the knowledge base correcting means is supplied with the attribute conditions for the inconsistency or lack of knowledge output by the knowledge base verification means.
The knowledge base modifying means outputs the attribute condition in which the inconsistency or lack of knowledge has occurred to the input / output device and presents it to the user, and when a correct answer candidate in these attribute conditions is input from the user, A positive case is added to the positive case storage unit so that the answer candidate input by the user is established, and a wrong case is added to the negative case storage unit.

In this way, it is not necessary to prepare special test data or comprehensively verify all possible cases, and it is possible to perform verification of the knowledge base in a lump. Further, since the knowledge to be verified is stored in a table format in advance, it is not necessary to convert the knowledge into a positional cube format or the like again, and accordingly, high-speed verification can be performed.
Further, the user can easily modify the knowledge base simply by inputting a correct answer.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention.
Referring to FIG. 1, the knowledge-based verification apparatus according to one embodiment of the present invention includes a positive case storage unit 11 storing, in a tabular form, candidates for an answer in a target problem and conditions for the candidate to be an answer. A knowledge base storage unit 1 having a negative case storage unit 12 storing, in a table form, answer candidates for a problem to be solved and conditions when the candidate should not be an answer.
A knowledge base verification unit 2 having a consistency verification unit 21 for verifying inconsistency of knowledge in the knowledge base stored in the knowledge base storage unit 1 and an integrity verification unit 22 for verifying lack of knowledge; It comprises a knowledge base modifying means 3 for the user to modify the knowledge base when the inconsistency or lack of knowledge is found by means 2. Reference numeral 4 denotes an input / output device connected to the knowledge base modifying means 3, which is composed of, for example, a CRT 41 and a keyboard 42.

FIG. 2 shows an example of a knowledge base stored in the knowledge base storage means 1. "Answer candidate 1" written on the left of the table is an answer candidate name. Above the table,
An attribute name such as "attribute 1" and a possible attribute value of each attribute such as "attribute value 11" are described. One row in the table corresponds to one case. Cases include positive cases and negative cases, and negative cases are distinguished from positive cases by adding a negative case identifier “#” to the beginning of the answer candidate name. In each column of the table, the relationship between the answer candidate name and the attribute value is input for each case. For example, when a certain answer candidate becomes an answer, if a certain attribute value is satisfied, enter “○”. If not, enter “×”. If all the attribute values are x for at least one attribute, such an attribute condition is called null. Null is a condition that does not hold. Also,
If all the attribute values are ○, the condition is called a universe. The universe is a condition that always holds.

The first line in FIG. 2 indicates that the value of attribute 1 is attribute value 11 and the value of attribute 2 is attribute value 22 or attribute value 2
In the case of 3, the answer candidate 1 is satisfied as an answer. " The fourth line means that "if the value of the attribute 1 is the attribute value 11 and the value of the attribute 2 is the attribute value 21, the answer candidate 2 is not an answer."

When there are a plurality of positive cases having the same answer candidate name in the knowledge base, at least one
If one is established, the candidate is the answer. When there are a plurality of negative cases having the same answer candidate name and at least one of the plurality of cases is established, the candidate is not an answer. Further, when there is a positive case and a negative case having the same answer candidate name, the part obtained by removing the condition for forming the negative case from the condition for forming the positive case is the condition for forming the answer candidate. That is, if only positive cases are satisfied and negative cases are not satisfied, the candidate is an answer. On the other hand, when a negative case is established, the candidate is not an answer regardless of whether a positive case is established or not.

For example, in FIG. 2, "answer candidate 2"
Is "when the value of the attribute 1 is the attribute value 11 or the attribute value 12 and the value of the attribute 2 is the attribute value 21" or "the value of the attribute 1 is the attribute value 11 and the value of the attribute 2 Is the attribute value 23
In this case, the answer candidate 2 becomes the answer. However, "Attribute 1
Is the attribute value 11 and the value of the attribute 2 is the attribute value 21 ”is excluded.”

Although positive cases and negative cases are collectively displayed in FIG. 2, positive cases in the knowledge base are stored in the positive case storage unit 11, and negative cases are stored in the negative case storage unit 12.
Is stored in

The knowledge base verifying means 2 detects inconsistency or lack of knowledge with respect to the knowledge base stored in the table format as described above.

First, the consistency verification unit 21 of the knowledge base verification unit 2 will be described. The non-contradiction verifying unit 21 checks whether there is a case where a plurality of candidates are answered simultaneously (verification of contradiction). FIG. 3 shows the consistency checking unit 21.
2 shows a flowchart of an example of the processing of FIG. First, for all the answer candidates in the knowledge base, the condition for forming a negative case is removed from the condition for forming a positive case, and the result is rewritten into an expression using only the positive case (P1). A calculation method for removing negative cases from positive cases uses a sharp operation. Next, all the combinations of two different candidates are generated for the positive cases obtained in this way (P2), and the attribute condition for the two candidates to simultaneously answer for each combination is calculated (P3). If the two candidates do not simultaneously answer in all combinations (yes in P4), the knowledge base is determined to be consistent (P5), and the combination of the two candidates that can simultaneously answer is obtained. If there is (P
No. 4), an attribute condition in which the two candidates are the answer,
A candidate that includes this attribute condition as a fulfillment condition is determined, and the candidate is notified to the knowledge base modifying means 3 (P6).

Here, the process of judging whether or not the two candidates at the same time as the answer in step P3 will be described in detail. For convenience of explanation, the two candidate C _i, a positive case of C _i generated by C _j Step P1 d _k
(K = 1,..., N _i ) A positive example of C _j generated in step P1 is defined as e _l (l = 1,..., N _j ). d
_{k (k = 1, ...,} n i) and _{e l (l = 1, ...} , n j) and the attribute condition which satisfies at the same time, obtaining the logical product of the condition for realizing establishment condition of d _k and e _l Calculate by That is, for each attribute value defined in the knowledge base, d
_When both the value of _{k and} the value of e _l are ○, a circle is added,
In other cases, by assigning x, it is possible to obtain the condition that d _k and e _l are simultaneously satisfied. If the logical product of any of the attributes is null in all combinations of d _k and e _l , it is determined that the candidates C _i and C _j will not be answered simultaneously. Also, if the logical product of all the attributes is not null, the candidate C _i and C _j will be answered at the same time when the attribute value marked with “積” is satisfied by the logical product. Is determined to be included.

FIG. 4 shows a specific example of consistency verification. Since the knowledge base before verification (FIG. 4 (1)) includes positive cases and negative cases, it is converted to a description of only positive cases (FIG. 4 (1)).
(2)). Based on this knowledge base, the logical product of each candidate is calculated (FIG. 4 (3)). As a result, candidate A and candidate B
Are the answers at the same time when the attribute value 1b of the attribute 1 and the attribute value 2a of the attribute 2 are satisfied, and it is determined that the knowledge base contains inconsistency.

Next, the integrity verification section 22 of the knowledge base verification means 2 will be described. The integrity verification unit 22 verifies that none of the candidates can be answered (insufficient verification). FIG. 5 shows a flowchart of an example of the processing of the integrity verification unit 22. First, for all the answer candidates in the knowledge base, the conditions for forming negative cases are removed from the conditions for forming positive cases, and the answer is rewritten as an expression using only positive cases (Q1).
Next, the conditions for establishing all positive cases obtained in step Q1 are removed from the universe (Q2). For this removal, a sharp operation is used. If the result is null,
The knowledge base is judged to be complete without lack (Q
3) If the result is not null, it is determined that the case under the attribute condition is insufficient, and the result is notified to the knowledge base correcting means 3 as a lack of knowledge (Q4).

FIG. 6 shows a specific example of integrity verification. Since the knowledge base before verification (FIG. 6 (1)) includes positive cases and negative cases, it is converted to a description of only positive cases (FIG. 6).
(2)). The condition of the case included in this knowledge base is removed from the universe (FIG. 6 (3)). As a result, attribute 1
The case when the attribute value 1b of the attribute 2 and the attribute value 2c of the attribute 2 are insufficient is insufficient, and it is determined that the knowledge base is incomplete.

As described above, the knowledge base verifying means 2
If the inconsistency or lack of knowledge is found, the verification result is input to the knowledge base correcting means 3, and the knowledge is corrected.

FIG. 7 shows a flowchart of an example of the processing of the knowledge base modifying means 3. For the sake of explanation, the candidates detected by the consistency verification unit 21 are _denoted by C _i1 and C _i1.
i2, …, C _it C _i1, C _i2, …, C _it
1 The correct answer in S1 is A1. The attribute condition with insufficient knowledge detected by the integrity verification unit 22 is A2, and the correct answer in S2S2 is A2.

If what is detected by the knowledge base verification means 2 is "knowledge inconsistency" (R1), the knowledge base correction means 3 displays the candidate C _i1, C
_i2, ..., _Cit and the attribute S1 are displayed (R2), and the user is asked for a correct answer when S1 holds (R3). When the user inputs the answer A1 from the keyboard 42, the input A1 is read. (R4). And A1 is C _i1, C
If it is equal to one C _{ik of i2,} ..., C _it , then S1
The _{C i1, ..., C i (} k-1), C i (k + 1), ..., and stores the negative case storage part 12 as a negative example of C _it (R5). On the other hand,
A1 is C _i1, C _i2, ..., in the case also differ from any of C _it is stored in the affirmative case storage unit 11 as a positive case of S1 is A1, S1 is C _i1, C _i2, ..., negative C _it The case is stored in the negative case storage unit 12 (R6).

For example, as shown in FIG.
And attribute conditions (attribute values 1b, 2
a) is detected by the consistency verification unit 21 and
If these are presented to the user and the user inputs, for example, A as a correct answer, the knowledge base modifying means 3 sets (attribute values 1b, 2a) as a negative case of the answer B, that is, as an answer. Attribute value 1 with B and attribute 1
b, a negative case in which the attribute value 2a of the attribute 2 is ○ and all other attribute values are x is stored in the negative case storage unit 12.
On the other hand, if the user inputs, for example, E as a correct answer, the knowledge base modifying means 3 determines (attribute values 1b, 2
a) is an affirmative case of the answer E, that is, an affirmative case in which the answer has E and the attribute value 1b of the attribute 1 and the attribute value 2a of the attribute 2 are ○, and all other attribute values are x. In the storage unit 11, (Attribute value 1b, 2a) is a negative case of the answers A and B, that is, the answer has A, and the attribute value 1b of the attribute 1 and the attribute value 2a of the attribute 2 are ○. A negative case in which all the attribute values are x, a solution having B as the answer, and the attribute value 1b of the attribute 1 and the attribute value 2a of the attribute 2,
Negative cases in which all other attribute values are “x” are stored in the negative case storage unit 12.

Referring again to FIG. 7, when the knowledge base verification means 2 detects "insufficient knowledge" (R1), the knowledge base correction means 3 displays the attribute condition S2 on the screen of the CRT 41. It is displayed (R8), the user is asked for the correct answer when S2 holds (R9), and when the user inputs the correct answer A2 from the keyboard 42, it is read (R10). Then, S2 is stored in the positive case storage unit 11 as a positive case of A2 (R11).

For example, as shown in the example of FIG. 6, when the integrity verification unit 22 detects that the case where the attribute value 1b of the attribute 1 and the attribute value 2c of the attribute 2 are satisfied is insufficient. If they are presented to the user and the user inputs the answer F as a correct answer, the (attribute value 1b, attribute value 2c) is a positive example of the answer F, that is, having F as the answer and The positive case in which the attribute value 1b of the attribute 1 and the attribute value 2c of the attribute 2 are ○ and all the other attribute values are × are stored in the positive case storage unit 11.

Next, a more preferred embodiment of the present invention will be described.

The processing for removing the condition of the negative case from the condition of the positive case and converting it into the expression of only the positive case (step P1 in FIG. 3, step Q1 in FIG. 5) and the condition of each case from the universe are removed. (Step Q in FIG. 5)
In 2), the processing time increases as the number of conditions to be removed increases. Therefore, when there are a plurality of cases to be removed, those that can be integrated among them are collectively reduced to reduce the number of cases, so that the processing time of the removal can be reduced.

For example, when there are a plurality of cases in which the values of all the other attributes except for one attribute are the same among the cases to be removed, the attribute values of the different one attribute are One case in which all attribute values of a case are set to x, the attribute value is set to x, and the other attribute values are set to o (the same attribute is inherited between cases as it is)
The conditions obtained by the above include all the conditions of all the cases before the integration, and there is no expansion or reduction of the logical meaning before and after the integration.

FIG. 8 is a flowchart of an embodiment in which an operation for integrating conditions of positive cases is added to the integrity verification processing shown in FIG. In this embodiment, after all the answer candidates are converted into expressions of only positive cases (T1), a plurality of conditions having the same attribute except for one attribute among the conditions obtained as a result are set to 1 (T
2). Then, after the number of conditions is reduced in this way, removal from the universe is performed (T3). Steps T4 and T5 are the same as steps Q3 and Q4 in FIG.

FIG. 9 shows an example in which the integrity verification of FIG. 8 is applied to the knowledge base shown in FIG. The first and second rows in FIG. 9B have the same other attributes 1 except for one attribute 2, and the third and fourth rows have one attribute 2 except for one attribute 2. And other attributes 1 are the same. Therefore, FIG. 9C is obtained by combining the first and second rows and the third and fourth rows. As a result, the condition must be removed from the universe only twice, and the obtained result (FIG. 9)
(4)) is the same as the result when the conditions are not integrated (FIG. 6 (3)).

A process for removing the condition of the negative case from the condition of the positive case and converting it into the expression of only the positive case (FIG. 3)
In step P1 of FIG. 5 and step Q1) of FIG. 5, the calculation time can be reduced by previously integrating the negative cases to be removed in the same manner as described above.

[0041]

As described above, in the present invention, the knowledge base is stored in the form of a table. Therefore, since the logical operation can be applied to the case stored in the knowledge base as it is, it is possible to speed up the verification of the inconsistency or lack of knowledge. Of course, there is no need to prepare special test data. Further, the knowledge base can be easily modified using the verification result. Since the user only needs to input correct results, it is easy to modify a large and complex knowledge base.

In particular, according to the configuration in which cases other than one attribute having the same attribute value except for one attribute are integrated and the processing is advanced, further higher speed can be achieved.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an embodiment of the present invention.

FIG. 2 is a diagram illustrating an example of a knowledge base.

FIG. 3 is a flowchart illustrating a processing example of a consistency verification unit;

FIG. 4 is a diagram showing a specific example of consistency verification according to the present invention.

FIG. 5 is a flowchart illustrating a processing example of an integrity verification unit.

FIG. 6 is a diagram showing a specific example of integrity verification according to the present invention.

FIG. 7 is a flowchart illustrating a processing example of a knowledge base modifying unit;

FIG. 8 is a flowchart illustrating another example of the processing of the integrity verification unit.

FIG. 9 is a diagram showing a specific example of integrity verification according to the processing example of FIG. 8;

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Knowledge base storage unit 11 ... Positive case storage unit 12 ... Negative case storage unit 2 ... Knowledge base verification unit 21 ... Consistency verification unit 22 ... Completeness verification unit 3 ... Knowledge base correction unit 4 ... Input / output device 41 ... CRT 42 ... Keyboard

Claims (4)

(57) [Claims] 1. A knowledge base verification apparatus for verifying the correctness of a knowledge base describing expert knowledge, in which a candidate for an answer in a target problem and a condition when the candidate is an answer are expressed in a table format. And a negative case storage unit that stores a negative case in which a candidate answer for the target problem and a condition when the candidate should not be an answer are stored in a table format. A knowledge base storage unit having the knowledge base storage unit, a consistency verification unit for verifying inconsistency of knowledge in the knowledge base stored in the knowledge base storage unit by a logical operation in a tabular form for positive cases and negative cases, and the knowledge base storage unit And an integrity verification unit for verifying the lack of knowledge in the knowledge base stored in Includes a base verification unit, inconsistencies and lack of information for the knowledge found by the knowledge base verification means outputs the output device, and a knowledge base correction means for correcting said knowledge base in accordance with an input from said input output device, The integrity verification unit removes one attribute from negative cases in the knowledge base.
Negative cases where the values of other attributes are all the same
Means for matching, and affirming all answer candidates in the integrated knowledge base.
Means for converting to an expression using only examples, and excluding one attribute from positive cases obtained by the conversion
Positive cases where all other attribute values are the same.
Means for matching and the conditions for establishing the case obtained by integration are removed from the universe.
The knowledge base if the result is null.
Judge that there is no lack of knowledge, and if not null, lack of knowledge
To judge that there is and find the condition of the missing case
Knowledge base verification apparatus characterized by comprising a stage. 2. The inconsistency verification unit removes one attribute from negative cases in the knowledge base.
Negative cases where the values of other attributes are all the same
Means for combining, a means for converting all answer candidates in the integrated knowledge base into an expression using only positive cases, and generating all combinations of two different candidates from the positive cases obtained by the conversion. Means for calculating the conditions under which the two candidates are simultaneously answered for each of the generated combinations; and when the two candidates are not simultaneously answered in all the combinations, the knowledge base is consistent. Is determined, and
Means for judging that the knowledge base is not inconsistent if there is a combination of two candidates that can simultaneously provide an answer, and calculating a candidate that includes a condition for the two candidates to be an answer as a fulfillment condition. knowledge base verification apparatus according to claim 1 Symbol mounting, characterized in that. 3. The knowledge base modifying means calculates a candidate that includes a condition in which the knowledge base is determined to be non-consistent in the consistency verification unit and at the same time the two candidates are answered as satisfied conditions. At this time, the calculated candidate and its establishment condition are output from the input / output device,
It facilitates the input of a correct answer when the condition is satisfied, and when the answer input from the input / output device is equal to any one of the calculated candidates, the remaining candidates should not be the answers. The negative case in which the satisfaction condition is set as a condition is stored in the negative case storage unit, and if the input answer is different from any of the calculated candidates, the input answer is a condition when the input answer is an answer. A positive case in which a condition is satisfied is stored in the positive case storage unit, and a negative case in which the calculated condition is set as a condition when the calculated candidate should not be an answer is stored in the negative case storage unit. 3. The knowledge base verification device according to claim 2, wherein: 4. The knowledge base modifying means, when the integrity verification unit determines that there is a lack of knowledge and determines a condition of the lacking case, determines whether or not the obtained case is insufficient. The condition is output from the input / output device, and the input of the correct answer when the condition is satisfied is promoted.When the input of the answer from the input / output device is performed, the condition is insufficient when the input answer becomes the answer. The knowledge-based verification apparatus according to claim 3 , wherein a positive case in which a condition of a certain case is set is stored in the positive case storage unit.