JPS63241638A - Inference device facilitating input and verification of knowledge - Google Patents

Abstract

PURPOSE:To obtain a reliable inference device by executing the verification of inputted knowledge at the time of inputting necessary various knowledge in the inference and immediately correcting at that time if errors and insufficiency are found. CONSTITUTION:The knowledge inputted and corrected through a 'knowledge input device' 101 is collated with the expression of knowledge handled by a 'inference mechanism' 102 so as to check the error of a syntax. If there is the error of syntax a system side points it out and urges correction to a person who inputs the knowledge. The knowledge whose error of syntax is removed is stored in a 'knowledge base' 105 anyhow. A 'knowledge verification part' 103 compares the newly inputted and corrected knowledge with other knowledge already stored in the 'knowledge base' 105 and checks whether or not contradiction and redundancy, etc., exist between the new knowledge and other knowledge. If contradiction or redundancy exists, it informs the effect to the person who inputs the knowledge.

Description

[Detailed Description of the Invention] [Objective of the Invention] (Field of Industrial Application) The present invention supports human intellectual activities by making inferences using human knowledge that has been input into a computer in advance. The present invention relates to an inference device with a function that facilitates knowledge input and verification in such systems.

(Conventional technology) In conventional inference devices based on production systems, frame systems, etc., when inputting or correcting knowledge, humans have to manually check for errors and consistency with other knowledge. Even if checks were carried out, or even if inference devices were carried out, they were only rudimentary, such as simply checking the syntax of the conditional and conclusion parts of production rules and pointing out errors. Furthermore, in most cases, the analysis of the effects of changes in various parameters such as certainty factors used in knowledge on other knowledge was carried out on a trial-and-error basis.

(Problem that the invention aims to solve) The problem is that it becomes difficult to manually perform knowledge consistency and parameter analysis as the scale of knowledge increases, making it difficult to construct a useful and highly secure system. There is.

The present invention has been made to overcome the drawbacks of the prior art described above, and an object of the present invention is to provide an inference device that can easily and efficiently acquire large-scale or complex knowledge.

[Structure of the invention]

(Means for Solving the Problem) In order to achieve the above object, the present invention provides a knowledge input device that can input human knowledge as knowledge data, a knowledge base that stores the knowledge data, and a knowledge base that stores the input knowledge data and the knowledge data. In an inference device equipped with an inference mechanism that performs inference using base knowledge data, consistency between input knowledge data and modified knowledge data and knowledge data stored in the knowledge base is checked, and if there is an error, We have a knowledge verification section that prompts the input of new knowledge, and a knowledge analysis section that sets parameters in the knowledge base and analyzes the influence of these parameters on inference. This is an inference device configured to be easy to perform.

(Operation) The knowledge input and modified through the "knowledge input device" 101 is checked for syntactic errors by comparing it with the knowledge expressions that can be handled by the inference mechanism box 02.

If there is a syntactical error, it is pointed out and the knowledge inputter is prompted to correct it. The knowledge from which syntactical errors have been removed is temporarily stored in the "knowledge base" 105. ``Knowledge Verification Department j1
03 compares the newly input/modified knowledge with other knowledge already stored in the "knowledge base" 105 and checks whether there are any inconsistencies or redundancies between the new knowledge and other knowledge. Check if. make him use the spear. The result of the simulated inference is presented to the person inputting knowledge, and the person inquiring whether the simulated inference result is as intended by the person inputting knowledge. If the information is as intended, the input/corrected knowledge is formally incorporated into the "knowledge base" 105 as part of inference knowledge and stored. When the result of the simulated inference is not as intended by the knowledge inputter, have the knowledge inputter point out the difference between the intended result and the result of the simulated inference, and collect the knowledge that is the cause of the difference. The system constructs the information horizontally and presents it to the person inputting the knowledge to encourage corrections, etc. After repeating this process until the result of the simulated inference is as intended by the person inputting the knowledge, the inputted, corrected, and corrected knowledge is stored in the knowledge base envelope 05. “Knowledge Analysis Department” 103
teeth.

Simulated inference is performed by variously changing various parameters used in the knowledge, and the results are presented to the knowledge inputter to assist the knowledge inputter in determining parameter values.

(Embodiment) FIG. 2 shows the configuration of an embodiment of the present invention. This embodiment includes a "processing device J201" that processes input reception, input checking, inference, knowledge verification, knowledge analysis, etc., an "input device" 202 that receives user input, and a "display device" that presents various information to the user. J 203, and a "storage device" 204 that stores various data.

In this example, the rules regarding inference are IF~THE.
It is expressed in the form of N~, and observed events, hypotheses, etc. are expressed in the form of (event name value confidence).

-4= First, the function of the "knowledge base editor" will be explained. [Knowledge base editor] guides the knowledge inputter to the items to be input, and the input knowledge is used as an inference mechanism.
``Knowledge syntax check process'' 301 in Figure 3
It has processing functions.

For example, when inputting an event that indicates "the magnitude of the current is IOA", enter the event name? : It will ask you the name of the event you are about to enter. On the other hand, the event name? : When you enter current, value?・ Prompts you to enter the next item, and asks ``Value?'' :10A and further confidence? : Against.

Confidence? : By answering 1.0, you can input the event. Furthermore, even if the knowledge inputter makes an incorrect input, the ``knowledge base editor'' automatically checks it and points out the error. For example, where confidence level should be entered, confidence level? : If you enter voltage, the confidence must be a number between -1.0 and 1.0.

Confidence? ： It will point out the error and ask you to enter it again.

Next, the operations performed between the "knowledge base editor" and the "knowledge verification mechanism" will be explained. "Knowledge Verification Organization"
compares and cross-checks the knowledge from which syntactical errors in the knowledge expression have been removed by the [Knowledge Base Editor] with the knowledge already stored in the [Knowledge Base], and compares and verifies the knowledge that has been newly inputted and modified. IF (current 10A) THEN (voltage height 1.0) in the knowledge "knowledge base" 105
Suppose that an inference rule called ...■ is registered. Here, the knowledge input person newly added IF (current 10A) THEN (voltage low 1.0)・
Suppose we input the inference rule ・・・■. There is no syntactical error in the rule ■ itself. However, when compared with the rule ■, these two inference rules both show that the current is l0
This is a rule that is applied at the time of AJ, and the content of the conclusion is that ``the voltage is high'' in the case of ■, whereas the content of the conclusion in ■ is that ``the voltage is low.'' this is,
This indicates that the newly inputted knowledge ■ and the knowledge ■ existing in the "knowledge base" 105 are inconsistent.

When knowledge such as ■ is input, the "knowledge verification unit" 103 informs the knowledge inputter that the knowledge is inconsistent with the knowledge of ■, and prompts the knowledge inputter to correct one of the knowledges.

Also, suppose that the "knowledge base" 105 contains the following inference rule.

IF (current 10A) OR (warning light 0N) THEN (voltage height 1.0)・・・・・・・・・・・・
・・・・・・■On the other hand, IF (current 10A) THEN (voltage height 1.0)・
Suppose we try to add a new inference rule called ...■. The inference rule for ■ has the same conclusion part as the inference rule for ■, and the conditional part is included in the inference rule for ■. In this case, the rule ■ is said to be redundant.

When such redundant knowledge is input, the "knowledge verification unit" 103 notifies the person inputting the knowledge and urges the person to correct it.

The above “Knowledge Verification Section” 103 and “Knowledge Base Editor J”
The major difference from conventional technology in terms of the interaction between the two is that it does not simply present contradictions and redundancies, but it also allows knowledge to be corrected interactively on the spot.

Next, “Knowledge Base Editor”, “Knowledge Verification Department” 103,
The actions performed between each component of the "reasoning mechanism" 102 will be described. ``Verification of knowledge contradictions and redundancy'' mentioned above
is a verification that can be done mainly by checking the representation of knowledge. On the other hand, checking whether the result of inference using accumulated knowledge is as intended by the person inputting the knowledge is
Checking the representation of knowledge is insufficient; it is necessary to make inferences in a simulated manner and check the results.

This kind of check will be referred to as "checking the validity of knowledge" and will be referred to as "knowledge validity checking process" 30 in Figure 3.
It has the processing functions shown in 3. Here, this "validity check" is performed by "Knowledge Base Editor", "Knowledge Verification Department", "
Explain how this is done between the "reasoning mechanisms".

For example, suppose that the following two inference rules are registered in the "knowledge base" 105.

IF (X can swim) Tl (EN (X fish 1, 0)...
■IF (X is a viviparous) THEN (x is a mammal 1, 0) ...
■If we add the fact that (mineral deposits are 1, 0) and perform a simulated inference, we get the inference result (1, 0 that is cetacean fish). What should be the result if we inform the "knowledge verification section" 103 that the result is not satisfactory to the person who entered the knowledge? −> The inference device asks a question. On the other hand, if we enter -> (1, 0, which is a cetacean mammal), in order to draw the conclusion (1, 0, which is a cetacean complement), we need the fact that (1, 0, which is a cetacean viviparous). is required. Would you like to add this fact to your "knowledge base"? −〉, the reason why the conclusion (1, 0, which is a whale mammal) was not drawn, and the information necessary to draw it are urged to be added. Here, if the person inputting the knowledge answers ->Yes, the knowledge (1, 0 which is viviparous whale) is automatically added to the "knowledge base". After that, we perform another simulated inference, resulting in the conclusions (1, 0, which are cetaceans) and (1, 0, which are cetacean mammals). If you only want the conclusion that it is a whale mammal (1, 0), you need to correct one of the following two pieces of knowledge.

A) IF (X can swim) THEN (X fish 1, 0) B) (mineral deposit 1, 0) Which would you correct? -〉 and asks for correction. In this case, if the inference rule in A) is insufficient and I want to add the condition (x egg size 1.0), if I answer ->yes, how should I correct it? Please choose from the following.

1) Rewriting the conditional part 2) Adding a term in the conditional part 3) Deleting a term in the conditional part 4) Others -> It asks how to make the correction. In this example, the problem can be solved by adding the term (x oviparous) to the conditional part of the inference rule in A), so if you select ->2, enter the conditional term to be added.

-> It will ask you to input. On the other hand, if you enter -〉(× oviparous), is the combination of conditional terms AND or OR? −〉, so enter −)AND and the corrected result is as follows.

IF (X can swim) AND (X is oviparous) THEN (X is fish 1, 0) Is this okay? You will be asked to confirm IZ-1〉, so if you answer ``-〉yes, this inference knowledge will be added to the "knowledge base" 105.

Next, the operation between the "knowledge analysis unit" 104 and the "reasoning mechanism" 102 will be described. The “knowledge analysis department” 104 is “
performs a simulated inference using the "inference mechanism" 102, analyzes how various parameters used in the knowledge affect the inference, and assists the knowledge inputter in determining parameter values. It has the processing functions shown in the flowchart of FIG. For example, the following inference rule is ``
Assume that it is registered in the "Knowledge Base".

IF (fever over 38 degrees) AND (throat pain) AND (feeling sluggish) AND (cough) WITH0,8 THEN (disease name cold) Here, rtlITH0,8J is the event of each term in the conditional part. The confidence level of the entire conditional part is calculated from each confidence level, and if it exceeds 0.8, this inference rule is applied. The event of each term is identified using some means, but the size of the confidence and the threshold value for the event of each term will greatly affect the applicability of this inference rule. In such a case, it is necessary to analyze the effect that changes in the event certainty of each term of the threshold have on the application of this inference rule and, ultimately, on the entire inference. By executing simulated inference by varying various parameters that affect inference within the range of values that the parameters can take, and presenting the results to the knowledge inputter, the knowledge inputter can Assists in the process of determining parameter values.

As seen in this example, when inputting various knowledge necessary for inference, the inputted knowledge can be verified on the spot, and if there is an error or inadequacy, it can be corrected on the spot. It is also possible to analyze the parameters used in the knowledge, so you can input, verify, and verify the knowledge necessary for inference.
Analysis etc. can be easily performed.

〔Effect of the invention〕

By using the inference device according to the first aspect of the present invention, it becomes easy to input, correct, verify, analyze, etc. knowledge used for inference, and various inference systems can be efficiently developed.

It is particularly effective in developing inference systems that require large-scale or complex knowledge.

[Brief explanation of drawings]

FIG. 1 is a block diagram of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, and FIGS. 3 and 4 are flowcharts regarding knowledge acquisition and parameter analysis of functions of the embodiment of the present invention. 101...Knowledge input device 102...Inference mechanism 1
03...Knowledge verification section 104...Knowledge analysis section 1
05... Knowledge base 201... Processing device 20
2... Input device 203... Display device 204
...Storage device 301...Knowledge syntax check process 302...Knowledge consistency check process 303...Knowledge validity check process agent Patent attorney Nori Chika Ken Yutong Wang Hou Hongwen 1tl- 104/θ2 Figure 1 Fig. 2 1N Kaisho 63-241638 (6) Fig. 3 Fig. 4

Claims (1)

[Claims] An inference device that includes a knowledge input device that inputs knowledge data, a knowledge base that stores knowledge data, and an inference mechanism that performs inference using the input knowledge data and the knowledge data of the knowledge base, input knowledge data. and a knowledge verification unit that checks the consistency between the corrected knowledge data and the knowledge data stored in the knowledge base and prompts for input of new knowledge if there is an error, and a knowledge verification unit that sets parameters in the knowledge base and An inference device that facilitates the input and verification of knowledge, characterized in that it is provided with a knowledge analysis section that analyzes the influence of information on inference, thereby making it possible to easily input, verify, and analyze the knowledge necessary for inference.