JPH0212336A - Knowledge base system and knowledge base conversion system - Google Patents

Abstract

PURPOSE:To quickly start a knowledge base system by converting a knowledge base with which the development is through into the knowledge structure data described in a procedure type language via a knowledge base conversion module, connecting said structure data to an inference function library via a compiler and producing a program exclusive for application of the knowledge base system. CONSTITUTION:A knowledge base system consists of a knowledge base 1, a diagnosis type inference module 2, a knowledge base editor 3, a knowledge base conversion module 4, a structure data memory 5, a compiler 6, and an inference function library 7. The knowledge base 1 produced by the editor 3 is converted into the structure data described in a procedure type language by the module 4. Then the structure data is translated by the compiler 6 and connected to the library 7 to produce a program exclusive for application which includes the base 1 as a part of an exclusive program. In such a constitution, the knowledge base system is quickly started.

Description

DETAILED DESCRIPTION OF THE INVENTION A) Field of Industrial Application The present invention relates to a knowledge base system having a knowledge base described in a diagnostic machine knowledge expression format.

1) Conventional technology Recently, with the development of knowledge engineering, the application of artificial intelligence technology to industry has expanded significantly, and in particular, knowledge-based systems represented by expert systems are a typical example. A knowledge base system consists of a knowledge base that stores declaratively described problem domain-specific expertise and know-how, and an inference module that is separated from the knowledge base and performs interpretation and inference of the knowledge within the knowledge base. It consists of Because the knowledge base and inference module are configured separately, traditional procedural languages (e.g., PASCAL,
C, FORTI SEAN, etc.), program development efficiency, maintainability, readability, and extensibility are significantly improved.

Now, the knowledge base system (expert system) of a diagnosis rIfrm typically consists of a knowledge base described by a diagnostic knowledge representation, a diagnostic history inference module, and a knowledge base editor for constructing the knowledge base. ,
A knowledge base system is developed by repeatedly constructing a knowledge base using a knowledge base editor and verifying the knowledge within the knowledge base using a diagnostic reasoning module. Hereinafter, the knowledge base editor diagnostic inference module for constructing a knowledge base will be collectively referred to as a diagnostic shell.

9. The knowledge expression format for diagnostic problems is IF T.
There is a Taliteria board knowledge expression that summarizes the rules of the HI format and the criteria for reaching a conclusion in a tabular format called a Taliteria frame. In a rule-based system where rules are used as knowledge expressions, the modularity of individual rules (knowledge) is good, but as the number of rules increases to deal with complex problems, the relationships between individual rules and inferences become difficult. The history of the time system! The problem is that it becomes difficult to understand the information, making it difficult to develop and maintain the knowledge base. On the other hand, Taliteria-type knowledge representation is a knowledge representation that describes a group of related knowledge in a structured manner, and it is easy to understand the reasoning process for diagnostic problems, develop efficiency and maintainability of large-scale knowledge bases. Since it has the advantage of improving knowledge representation, a diagnostic expert system construction tool (diagnostic shell) using this knowledge representation has been proposed (see "Introduction to Knowledge Engineering" by Haruki Ueno, published by Ohmsha).

C) Problems that the invention aims to solve Now, conventionally, when operating and distributing a knowledge base constructed using a diagnostic shell based on this Taliteria type knowledge representation, the development of the knowledge base is completed. If you want to distribute it as an application system to end users, you can either use a form that consists of a knowledge base, a knowledge base editor for building the knowledge base, and a cutting board module, just like in the development stage, or a knowledge base and a knowledge base module. The execution of the application system was carried out either in the form of a diagnostic inference module and a knowledge pace loader module to load rounding.

However, when an application system, which is a knowledge-based system developed in this manner, is difficult to execute, there are the following problems.

That is, when the system starts up, after starting the knowledge base editor or inference module for building the knowledge base,
It is necessary to load the knowledge base stored in external storage means (such as a floppy disk) into the main unit, and when the capacity (size) of the knowledge base is large, it takes a long time to load, and the application system starts up slowly.

Since the tt1 application lν system is separated into a diagnostic shell and a knowledge base, the contents of the knowledge base can be
Even if you want to protect the knowledge base from end users (i.e., you want to provide only the problem-solving support function provided by the application system to the second user and keep the contents of the knowledge base secret), the end user may build the knowledge base. If you have a knowledge base editor or a utility with equivalent functionality, your knowledge base can easily be read, and the end user's knowledge will be lost.
base confidentiality cannot be maintained.

=3! Ninth Means for Solving Problem I The present invention is based on facts that indicate facts and observations necessary for rounding up inferences, and intermediate hypotheses and facts that indicate intermediate facts and hypotheses. The criteria are essentially summarized in a table format, and the knowledge for solving the problem is expressed and accumulated through the nine conclusion hypotheses and the knowledge base is created, and the problem is solved or supported based on the knowledge within the knowledge base. a diagnostic micro-inference module that performs inference for the purpose of the invention, a knowledge base editor for constructing the ninth knowledge base, and a knowledge base editor for constructing the ninth knowledge base, and converting the knowledge accumulated in the knowledge base into structured data using a procedural language. A knowledge base conversion means to convert, a PilI scribe type language 1 layer inference 131 library that performs the same inference as the punishment type inference module, and a '@ml inference IA number that translates the structure data. It is a coin-based system that includes a library and a compiler that combines it.

e) The knowledge base constructed by the knowledge base editor for constructing the operational knowledge base is converted into n14 structure data written in procedural agnostic language by the knowledge base conversion means, and furthermore, this structure data is sent to the compiler. The knowledge base is translated and combined with the inference function library to generate a problem-specific dedicated program (application system) in which the knowledge base is incorporated as part of the dedicated program.

f) Embodiment FIG. 1 is a schematic diagram showing one example of a part of the present invention. (
1) is knowledge-based and rare? The knowledge, tricks, and methods for solving Type 1 are divided into facts that indicate the facts and findings necessary for making inferences, intermediate hypotheses that indicate intermediate facts and hypotheses, and conclusions based on the facts and intermediate hypotheses. Knowledge described in a diagnostic type knowledge expression (Taraiteria knowledge expression) format consisting of a conclusion hypothesis with a table format Kt as a judgment criterion for deriving the results is accumulated.

(2) is the ninth diagnostic inference module that interprets and executes the knowledge base (11 (knowledge) and makes inferences; (3) is the knowledge base). This is a knowledge base editor.Usually, this knowledge base (11), a diagnostic inference module (2), and a knowledge base editor (3) for constructing a knowledge base are used to create a diagnostic knowledge base system (diagnosis expert system).
However, one embodiment of the present invention further includes the following configuration.

(4) is the amount of knowledge base conversion joules required to convert knowledge base +11 into structure data that is a program written in a procedural language (hereinafter referred to as a procedural base program), (5
) is a structure data memory that holds structure data C (hereinafter referred to as knowledge structure data) of the procedure board program converted in the knowledge base conversion module.

! 61 is a procedural deaf language compiler used for procedural programs converted by the knowledge base conversion module (4), and is equipped with a translation section &) (compiler in a narrow sense), an assembler, and a linker, and converts knowledge structure data into '4vL. It converts to a tasteful format that can be linked with the sparrow library.

(7) uses the aforementioned procedural language inference function library to interpret and interpret the knowledge base 111 using the diagnostic inference module (2).
It consists of knowledge structure access functions, inference functions, etc. for interpreting and executing knowledge structure data in the same way as for execution.

Now, the basic knowledge base system constructs a knowledge base (13) using a knowledge base editor (3) for constructing a knowledge base, and verifies the knowledge in the knowledge base llj using an inference module (21). The knowledge base il+ developed by the diagnostic shell (diagnostic inference module (2) and knowledge base editor for building a knowledge base (3)) is a knowledge base conversion module. The knowledge structure data is converted to VcyR by (4).

Knowledge Base 111 In the knowledge base 111, the knowledge for diagnostic problems is described using the Criterion Knowledge Expression as described above, and is divided into facts, intermediate hypotheses, and conclusion hypotheses as a knowledge database. Figure 2 Facts for each person 5BsC,
Examples of descriptions of intermediate and conclusion hypotheses are shown. As shown in Figure 2, knowledge is divided into units of $Knowledge data type Knowledge data name? Attribute name 1 Attribute value 1? It is written as Attribute name 2 Attribute value 2 $.

The knowledge base conversion module (4) registers knowledge data types (three types here, but there may be more types) that can be interpreted and executed by the diagnostic shell, and each knowledge data type It has an attribute registration table in which corresponding attribute names are registered. FIG. 3 shows an example of an attribute registration table for facts.

In Fig. 5, the structure is the structure when the knowledge data type (fact) is converted into a structure of a procedural language (in this example, the procedural language will be explained using Ct language as an example). 'I) @def is the macro name of the horizontal body declaration (in this example, the C language typd6f declaration rtyp
ed6f 5truct fact+sa).

The attribute name definition field is defined by the knowledge data 臘!
i-san teru J! The attribute name is written, and in the field ``Member Definition'', the data location and name of the structure are written corresponding to the attribute name. For example, the member definition corresponding to the attribute name "value" is 1ntvalue.

Default> is a field in which a default value is written when a corresponding attribute name and attribute value pair is not described in the knowledge description. Multi-replacement rule〉Du, attribute value to structure member? This field is used to write rules for performing continuous replacement when substituting the value of >t Real type a -> 2, so if the attribute value of the attribute "type" is written as "pool type", for example, rOJ is initialized as the value of type of the structure member. Ru.

In addition, a declaration is made to express knowledge as a procedure I5 type [JA structure knowledge structure in δ language (C language)] according to the data described in the fields of Attribute Name Definition and Member Definition. The sentences have been generated, and the declaration sentences of the generated 9#I structures are compiled in advance as a header file, and later included in the knowledge structure data program to be generated and compiled. In the same way that the Nesei Bonryoku table is defined for each knowledge data type, the d language of the knowledge structure is also generated for each data type. FIG. 4 shows an example of a declaration statement (in C language) for a fact structure.

FIG. 5 shows an outline of the R-data processing in the knowledge base conversion module (4). When the developed knowledge base (1) is input to the knowledge base conversion module (4), a knowledge database is identified for each knowledge unit.

When knowledge data overflow is determined, a structure generation table to be described later is created with reference to the attribute registration table corresponding to the knowledge data type. Next, default and replacement processing is performed based on the structure generation table t and the attribute registration table i, and the knowledge structure data is output. These processes are repeated nine times in the knowledge base until there is no unprocessed knowledge unit of one yen.

To explain in more detail, in order to generate knowledge structure data for each of the input knowledge bases (13 knowledges), Vζ, 1$ knowledge data #IIIt- is identified. Correspondingly, the table for generating nine structures is set according to the electric attribute registration table.For example, if the identified knowledge data dust is a categorization (i.e. 1$catekide + 1), the attribute registration of the 5th kiiu is set. According to the table [FIG. 6], a table is set in which the JI4 value field is nitride.Then, knowledge data name t- is read and written to the corresponding field in the table of FIG.

If the next character read is 1$1, this knowledge (1
This means that we have finished writing about knowledge rate).
Move on to the next process of generating knowledge structure data. '$1
If not, it means that there is an attribute value to be written to the currently set structure generation table (Figure 6), so '? After detecting ", the attribute name and attribute value pair is read and jtI is entered into the corresponding attribute value column of the structure generation table.

This process is repeated until "1$" is detected, which indicates the end of the knowledge description.Then, a structure generation table as shown in Figure 6 is created.Figure 7 shows the creation of a structure generation table. shows a flowchart for.

Once a structure generation table is created for each announcement m (each knowledge unit), a knowledge structure is generated by performing attribute value default processing and replacement processing.This processing will be explained with reference to the flowchart in FIG. do.

First, a pair of attribute name and attribute value is read into the item number from the structure generation table. At this time, if the attribute value is undefined, the attribute registration table of the corresponding knowledge data type is referred to and the default value described in the field of the corresponding attribute is written (default processing).

If the attribute value has already been defined, the attribute value in the structure generation table is changed by -fi4 by referring to the ``replacement rule'' field in the attribute registration table, and if a 1-replacement rule is written, it is changed accordingly. If there is no replacement rule, it is set to 1, and the attribute value is set to 11. The above processing is performed for all attributes described in the structure generation table.

Figure 6 in Figure 9! , 4 body generation table is shown after performing default processing and replacement processing based on the attribute registration table of FIG. In the dlS9 diagram, the blank character r, which is the default value, is added to the "description text" whose attribute value is undefined.
” J wrote n.

Also, according to the replacement rule, the attribute value "real number type" of the attribute "type" is replaced with "2", and the attribute value "determined" of the "evaluation flag" is replaced with "1".

Then, according to the structure generation table that has been subjected to default processing and replacement processing, knowledge structure data in procedural language H (C language) is output, and structure data memo IJ (5
Stored in 1. FIG. 10 shows a fact structure in which the facts shown to the person in FIG. 2 as knowledge in the knowledge base 11+ are converted according to the structure generation table of FIG. 9 created corresponding to the facts. Here, rFAcTJ is the macro name of the 7-act structure, then the pointer to the fact "house FIJ", the assignment symbol "castle", then the initial values of the members written in the fact structure, 1 value 1 degree value 2. . . . value N), and ";" indicating a sentence break is output at the end.

Looking at the correspondence between the knowledge (facts) described in Figure 2 and the knowledge structure data written in a procedural language in Figure 10, the knowledge data type Fact is in the structure FACT, and the knowledge data name F1 is in the structure. In the pointer to the structure FACT, each attribute name corresponds to each member definition of the structure FACT in FIG. 10, and each attribute value corresponds to the initial value of each member of the structure FACT in FIG.

Although the knowledge structure data shown in FIG. 10 is based on the initial value setting of the structure, it is also possible to explicitly assign attribute values to each member of the structure as shown in FIG. 11.

In this way, the knowledge in the Taliteria knowledge representation is converted into structure data in a procedural language (C language in this embodiment). Intermediate hypotheses and conclusion hypotheses that are not exemplified are also converted into knowledge structure data in a procedural language through the same processing as the exemplified facts.

Once the knowledge base (1) developed using the Shinken 11Ta shell is once converted into knowledge structure data in a procedural language using the knowledge base conversion module (4), it can be converted into knowledge structure data using a procedural language. A program for application cages is created by a process similar to that used in program development using a typed language (conversion process from a source degram to an object program). That is, knowledge structure data written in a procedural language is processed by a compiler (6) using a translation tautology and an assembler for the corresponding procedural language (C language).
Reasoning I! It is converted into a relocatable format that can be linked with the A number library (7), and further linked with the inference function library (7) by a linker, and created as a program exclusively for the application game.

By the way, the inference mechanism (inference module) in a knowledge base system is basically based on the function of accessing each knowledge within the knowledge base (referencing and assigning attribute values), no matter what the inference mechanism is. , The inference function provided in the inference function library (71) is a knowledge structure of a procedural program that converts the knowledge pace function part of the diagnosis function of the diagnosis function library (71) into the knowledge structure of the procedural program. This corresponds to the replaced data access function.

g) Effects of the Invention As is clear from the above description, the present invention converts a knowledge base that has been developed into knowledge structure data in a procedural language using a knowledge base conversion module, and combines it with an inference function library using a compiler. It is possible to create dekugrams specifically for the application cage of the knowledge base system. For this reason, compared to the conventional case where a completed diagnostician expert system is operated and distributed in the form of a pair of diagnostician shell and knowledge base, application In the blasphemous system, the startup is quick], and the knowledge base is also incorporated as part of a dedicated program in executable format, making the knowledge base invisible and eliminating the huge cost of construction. It becomes possible to strongly protect the knowledge base. Furthermore, the execution module generated as a dedicated program includes knowledge base development support modules such as a knowledge base editor for building a knowledge base that is necessary for knowledge base development but is not required when running the app. Since it is not included,
There is also the advantage that the size of the execution module is smaller than when operating and distributing the diagnostic machine shell and the knowledge base separately.

In this embodiment, the C language is used as the procedure*m language, but other procedure*m languages may also be used.

[Brief explanation of the drawing]

The drawings all relate to an embodiment of the present invention; FIG. 1 is a schematic diagram of the system configuration, FIG. 2 is a diagram showing an example of knowledge description, FIG. Figure 4 is a diagram showing an example of a declaration statement of a fact structure, Figure 5 is a flowchart showing an overview of knowledge base conversion processing, and Figure 6 is a flowchart showing an overview of knowledge base conversion processing.
9 and 9 are diagrams showing an example of a structure generation table,
FIGS. 7 and 8 are diagrams showing seven steps for creating a structure generation table and knowledge structure data, and FIGS. 10 and 11 are diagrams showing different examples of seven-act structure data.

Claims (1)

[Scope of Claims] 1) Facts indicating facts and findings necessary for making an inference, and criteria for drawing a conclusion based on intermediate hypotheses and facts indicating intermediate facts and hypotheses. A knowledge base in which knowledge for diagnostic problem solving is expressed and accumulated by conclusion hypotheses summarized in a format, and diagnostic reasoning that makes inferences to solve problems or support them based on the knowledge in the knowledge base. a module, a knowledge base editor for constructing the knowledge base, a knowledge base conversion means for converting the knowledge accumulated in the knowledge base into structured data in a procedural language, and reasoning similar to the diagnostic reasoning module. A knowledge base system comprising: an inference function library for the procedural language; and a compiler that translates the structure data and combines it with the inference function library. 2) An attribute registration table consisting of attributes corresponding to the knowledge data type of knowledge in the knowledge base in which knowledge is expressed using diagnostic knowledge expression is provided, and the knowledge data type of knowledge in the knowledge base is determined, and the determined knowledge data Creating a structure generation table in which the attributes of the attribute registration table corresponding to the type correspond to the attribute values of the knowledge, and converting the knowledge into structure data in a procedural language according to the structure generation table. Characteristic knowledge base conversion method.