KR102062560B1 - Method and system for integrated knowledge database construction based on production rules - Google Patents

Abstract

The present invention relates to a method for a server to build a rules-based knowledge database. More specifically, the present invention provides a method for selecting a data set from an external data set source, a step b for generating a classification model using a decision tree algorithm corresponding to the data set, and extracting conditions and results from the classification model. step c, step d for generating an XML model using the conditions and results, step e for extracting details of the XML model, step f for generating and verifying a production rule using the details, and steps a to f Repeating the steps to store the production rules generated in one or more data sets in the knowledge database.

Description

Rule-based knowledge database construction method and system {METHOD AND SYSTEM FOR INTEGRATED KNOWLEDGE DATABASE CONSTRUCTION BASED ON PRODUCTION RULES}

The present invention relates to a method of building a knowledge database, and more particularly, to a method and system for building a rule-based knowledge database using a data set.

Knowledge engineering is one of the key areas of research to build knowledge-based databases to create better decisions. In the research field, research is being focused on knowledge expression techniques such as decision trees, production rules, and decision graphs. Among these, production rules are most commonly used because they are concise, easy to understand, predictable, and reliable in supporting decision making.

However, as the size of data grows rapidly, professionals may have difficulty building knowledge databases using personal knowledge. Therefore, a decision tree algorithm can be used to build a knowledge database using such a huge amount of data.

Various systems have been proposed as a method for generating production rules using a decision tree algorithm. In most systems, only decision trees based on J48 classes can generate XML models. That is, the conventional method has a disadvantage in that a production rule is generated based on one decision tree without considering a plurality of decision trees.

The present invention has been made to solve the above-mentioned problem, and an object thereof is to build a knowledge database based on production rules.

It is also an object of the present invention to create a reliable production rule from a data set.

It is also an object of the present invention to generate a production rule using one or more decision tree algorithms.

In addition, the present invention is to improve the reliability by verifying the production rules from experts.

In order to achieve the above object, the present invention provides a method for building a rule-based knowledge database by a server, the method comprising selecting a data set from an external data set source and classifying using a decision tree algorithm corresponding to the data set. Step b of generating a model, step c of extracting conditions and results from the classification model, step d of generating an XML model using the conditions and results, step e of extracting detailed information of the XML model, the detail information By using step f to generate and verify a production rule, and repeating the steps a to f characterized in that it comprises the step of storing the production rule generated in one or more data sets in the knowledge database.

In addition, the decision tree algorithm may include at least one of BFTree, J48, RandomTree, REPTree, or SimpleCart.

Further, the step c is characterized by removing header information of the classification model and extracting conditions and results by using text trimming, text segmentation, and special character replacement techniques in the classification model.

Also, the d step may include converting the XML model to the XML model by performing an operator array, an indentation, and a file conversion operation on the classification model.

Furthermore, step e may include analyzing the XML model using a DOM analyzer, and generating the production rule including attributes, operators, and values using the analysis result of the XML model. do.

In addition, the step f, characterized in that it comprises the step of verifying the production rule, the external expert to verify that the production rule is operating normally.

Further, the checking of the production rule may include: comparing the production rule with an existing production rule previously stored in the knowledge database, and confirming the rule; if there is a rule overlapping the production rule and the existing production rule, the external expert It characterized in that it comprises the step of providing the overlapping rule to.

In addition, the present invention provides a system for building a rule-based knowledge database, the model generator for generating a classification model by selecting a data set from an external data set source, the decision tree algorithm corresponding to the data set, A preprocessing unit for extracting conditions and results from a classification model, a transformation unit for generating an XML model using the conditions and results, and a rule for generating a production rule by extracting and analyzing detailed information of the XML model and verifying the production rule. The generation unit, characterized in that it comprises a knowledge database for storing the production rule.

Furthermore, the decision tree algorithm may include at least one of BFTree, J48, RandomTree, REPTree, or SimpleCart.

The preprocessor may remove the header information of the classification model and extract conditions and results by using text trimming, text segmentation, and special character replacement techniques in the classification model.

Furthermore, the conversion unit may generate the XML model by performing an operator array, indentation, and file conversion operation on the classification model.

The rule generator may include a model analyzer that analyzes the XML model using a DOM analyzer, and a production rule generator that generates a production rule including an attribute, an operator, and a value using an analysis result of the XML model. It features one.

Furthermore, the rule generation unit may further include a production rule verifying unit including a rule verifying unit for verifying the production rule by an external expert and a rule confirming unit for confirming whether the configuration of the production rule is correctly written.

The rule checking unit may compare the production rule with a pre-stored existing production rule and check a rule. When the rule is confirmed, the rule checking unit may duplicate the production rule and the existing production rule. It is characterized by providing a rule that becomes.

According to the present invention as described above, it is possible to build a knowledge database based on production rules.

The invention can also generate reliable production rules from data sets.

In addition, the present invention may generate production rules using one or more decision tree algorithms.

In addition, the present invention can improve the reliability by verifying the production rules from experts.

1 is a diagram illustrating an overall process of building a knowledge database according to an embodiment of the present invention.
2 is a diagram illustrating a configuration of a knowledge database building system according to an embodiment of the present invention.
3 is a view for explaining a knowledge database construction method according to an embodiment of the present invention.
4 is a diagram illustrating a process of extracting a condition and a result required to generate an XML model according to an embodiment of the present invention.
5 is a view for explaining a process of generating a production rule according to an embodiment of the present invention.
6 is a view for explaining a process of verifying a production rule according to an embodiment of the present invention.
7 is a diagram for describing a classification model generated using a REPTree according to an embodiment of the present invention.
8 is a diagram illustrating an XML model according to an embodiment of the present invention.
9 is a view for explaining a method of analyzing an XML model according to an embodiment of the present invention.
10 is a view for explaining a production rule according to an embodiment of the present invention.

The above objects, features, and advantages will be described in detail with reference to the accompanying drawings, whereby those skilled in the art may easily implement the technical idea of the present invention. In describing the present invention, when it is determined that the detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

The same reference numerals in the drawings are used to indicate the same or similar components, all combinations described in the specification and claims may be combined in any way. And unless specified otherwise, reference to the singular may include one or more, and reference to the singular may also include the plural expression.

The terminology used herein is for the purpose of describing particular example embodiments only and is not intended to be limiting. Singular expressions as used herein may also be intended to include the plural meanings, unless the context clearly indicates otherwise. The term "and / or," "and / or" includes any and all combinations of the items listed therein. The terms "comprising", "comprising", "comprising", "comprising", "having", "having", and the like have implicit meanings, and as such, these terms are defined by their features, integers, It specifies steps, actions, elements, and / or components and does not exclude the presence or addition of one or more other features, integers, steps, actions, elements, components, and / or groups thereof. The steps, processes, and operations of a method described herein should not be construed as necessarily in their particular order as discussed or illustrated, unless the order of execution is specifically determined. . It should also be understood that additional or alternative steps may be used.

In addition, each component may be implemented as a hardware processor, each of the above components may be integrated into a single hardware processor, or the above components may be combined with each other to be implemented as a plurality of hardware processors.

The decision tree used to build the knowledge database is a typical method used for data mining analysis. The decision tree can be expressed visually by algorithms. The decision tree can be composed of the root of the tree which is the largest condition, the branch of the tree which is the detailed condition, and the leaf of the tree for the purpose of classifying the data.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a diagram illustrating an overall process of building a knowledge database according to an embodiment of the present invention.

The server may select a decision tree algorithm corresponding to the data set selected from an external data set source. The server may generate a classification model by applying a decision tree algorithm to the data set through the model generator. When the classification model is generated, the server may generate a preprocessing model by performing preprocessing to extract conditions and results from the classification model through the preprocessor. When the preprocessing model is generated, the server may generate an XML model by using the condition and the result of the preprocessing model through the transformation unit. The server can extract attributes, operators, and values from the XML model, generate production rules, and make validation requests to experts. After the expert completes the verification, the server can store the production rules in a knowledge database. More details on building a knowledge database will be described later with reference to FIG. 2.

2 is a diagram illustrating a configuration of a knowledge database building system according to an embodiment of the present invention.

The knowledge database construction system may include a model generator 100, a preprocessor 200, a converter 300, a rule generator 400, a storage 500, and a knowledge database 600.

The model generator 100 may generate a classification model by selecting a data set from an external data set source and selecting a decision tree algorithm corresponding to the data set. The model generator 100 may include a data selector 110, an algorithm selector 130, and a classification model generator 150.

The data selector 110 may select a data set for generating a production rule from an external data set source.

The algorithm selector 130 may select a decision tree algorithm suitable for the data source selected by the data selector 110. In this case, the decision tree algorithm may be one of BFTree, J48, RandomTree, REPTree, or SimpleCart.

The classification model generator 150 may generate a classification model by applying a tenth order cross validation method to a data set and a decision tree algorithm corresponding to the data set. Cross-validation is a technique that uses part of the data set to generate the model and the remaining part to verify the model, which has the advantage of avoiding overfitting. Tenth order cross-validation can separate the data set into ten groups, using nine groups to generate the model and one group to validate the model. That is, since the 10th order cross-validation method uses all groups included in the 10 groups for verification, the classification model can be generated by repeating the process 10 times. The classification model generated by the classification model generator 150 using REPTree is as shown in FIG. 7.

The classification model generator 150 may store the generated classification model in the classification model storage 510.

The preprocessor 200 may extract conditions and results from the generated classification model. The preprocessor 200 may include a classification model selector 210 and a model preprocessor 230.

The classification model selector 210 may select a classification model to perform preprocessing in the classification model storage 510.

The model preprocessor 230 may perform a preprocessing operation on the classification model by using text trimming, text segmentation, and special character replacement techniques. The model preprocessor 230 may remove header information included in, for example, a classification model. Referring to Figure 1, the classification model may include header information such as (a / b) [c / d]. Where a is the total order of all instances up to the current leaf, b is the total order of the misclassified instance of all instances up to the current leaf, and c is the total of all inappropriate instances terminated at the current leaf D means the total order of the misclassified instance of all inappropriate instances terminated in the current leaf. The model preprocessor 230 may remove header information, which is unnecessary information for generating an XML model. Eliminating unnecessary information can have the effect of using less memory.

The model preprocessor 230 may extract conditions and results from the classification model by performing preprocessing on the classification model. In this specification, the preprocessed classification model is referred to as a preprocessing model. That is, the model preprocessor 230 may generate a preprocessing model including conditions and results.

The model preprocessor 230 may store the preprocessing model in the preprocessing model storage 530.

The converter 300 may generate an XML model using the preprocessing model generated by the preprocessor 200. More specifically, the converter 300 may include a preprocessing model selector 310 and an XML model converter 330.

The preprocessing model selector 310 may select a preprocessing model to be converted into an XML model in the preprocessing model storage 530.

The XML model converter 330 may generate an XML model by performing an operator array, indentation, and file conversion on the preprocessing model. The XML model is a standardized language for interoperability in decision trees, and can be used without limitations on hard disks, operating systems and programming languages by applying independent Unicode across all platforms. In other words, the XML model can have the same configuration regardless of the data set and decision tree algorithm.

The XML model generated by the XML model converter 330 is illustrated in FIG. 8. Referring to FIG. 8, the XML model may be configured of at least one of an attribute, an operator, or a value. For example, the condition extracted by the XML model converter 330 from the preprocessing model may be expressed as an attribute = SituationCategory ", an operator = ", and a value = " Sitting ". Furthermore, the result extracted from the preprocessing model may be expressed as an output decision (value) = "Stretching".

The XML model converter 330 may store the XML model in the XML model storage 550.

The rule generator 400 extracts and analyzes detailed information of the XML model to generate a production rule and verify the production rule. In more detail, the rule generator 400 may include an XML model analyzer 410, a production rule generator 430, and a production rule verifier 450.

The XML model analyzer 410 may extract detailed information of the XML model. The XML model analyzer 410 may select an XML model from the XML model storage unit 550 and then syntactically analyze the XML model using a Document Object Model (DOM) analyzer. The DOM parser can construct an XML model into an object tree to extract instances of the object tree. The XML model analyzer 410 may extract detailed information of the XML model using the algorithm shown in FIG. 9.

The algorithm of FIG. 9 may extract the detailed information of the XML model by examining the information of nodes such as the relationship between the root of the tree, sibling nodes and parent nodes. In other words, you can search the tree by extracting all conditions along the result from the path of the object tree. In other words, the XML model analyzer 410 may extract attributes, operators, and values from the XML model using the DOM analyzer.

The production rule generator 430 may generate a production rule using attributes, operators, and values extracted by the XML model analyzer 410. For example, the production rule generator 430 may generate a production rule as shown in FIG. 10 from a classification model generated using REPTree and J48. Referring to FIG. 10, the production rule may consist of a condition IF and a result THEN.

The production rule generator 430 may store the generated production rule in the production rule storage 570.

The production rule verification unit 450 may enable an external expert to verify the production rule, and may verify that the configuration of the production rule is correctly written. More specifically, the production rule verifier 450 may include a rule verifier (not shown) and a rule checker (not shown).

The rule verifier selects a production rule stored in the production rule storage 570 so that an external expert can verify the rule. The expert can verify the production rules and store them in the rule-based knowledge database 600. The expert can verify that the production rules have been created in a suitable structure.

The rule checking unit checks the production rule by comparing the production rule with a pre-stored existing production rule. If there is a duplicate rule between the production rule and the existing production rule, the rule checker may provide a duplicate rule to an external expert.

The storage unit 500 may store the model and the production rule generated by the model generator 100, the preprocessor 200, the converter 300, and the rule generator 400. More specifically, the storage unit 500 includes a classification model storage unit 510 for storing the classification model generated by the model generator 100 and a preprocessing model storage unit 530 for storing the preprocessing model generated by the preprocessor 200. ), An XML model storage unit 550 that stores the XML model generated by the conversion unit 300, and a production rule storage unit 570 that stores the production rule generated by the rule generation unit 400.

The knowledge database 600 may store the production rule verified from the production rule verifier 450.

Hereinafter, a knowledge database construction method according to an embodiment of the present invention will be described with reference to FIGS. 3 to 6. In the description of the knowledge database construction method, detailed embodiments overlapping with the above-described knowledge database construction system may be omitted.

3 is a view for explaining a knowledge database construction method according to an embodiment of the present invention.

Referring to FIG. 3, the server may select a data set from an external data set source (S100).

When the data set is selected, the server may select a decision tree algorithm corresponding to the data set (S200). The decision tree algorithm may be one of BFTree, J48, RandomTree, REPTree or SimpleCart.

The server may generate a classification model by applying a tenth order cross validation method to the data set and the decision tree algorithm (S300).

The server may extract conditions and results included in the generated classification model (S400). The server can use text trimming, text segmentation, and special character substitution techniques to extract conditions and results from the classification model.

The server may generate an XML model using the conditions and the results extracted from the classification model (S500).

When the XML model is generated, the server may generate the production rule by extracting the detailed information from the XML model (S600).

The server may store the production rule in the knowledge database when it is determined to be a normal production rule by verifying and confirming the generated production rule (S700).

4 is a diagram illustrating a process of extracting a condition and a result required to generate an XML model according to an embodiment of the present invention.

The server may cut the text of one region by performing text trimming on the region including the condition and the result in the classification model (S410), and divide the cut out text to distinguish the condition and the result (S420).

The server may extract the conditions and the results of the classification model (S440) by replacing the special characters included in the classification model according to the XML model (S430).

5 is a view for explaining a process of generating a production rule according to an embodiment of the present invention.

The server may analyze the XML model using a DOM (Document Object Model) analyzer (S610). The DOM parser converts an XML model into an object tree, allowing access to instances of the object tree.

The server may extract detailed information including an attribute, an operator, and a value by examining an instance of an object tree representing a value of an XML model (S620).

The server may generate a production rule using the extracted detailed information (S630).

6 is a view for explaining a process of verifying a production rule according to an embodiment of the present invention.

When the production rule is generated, the server may make a request for verification of the production rule to an external expert (S640). External experts can determine whether the structure of the production rules is appropriate.

If the production rule is verified, the server may check whether the production rule operates normally (S650). Furthermore, the server can check the production rules by comparing the production rules with existing production rules stored in the knowledge database. At this time, if it is determined that there is a duplicate rule in the existing production rule and the generation rule, the server may provide a notification for the duplicate rule to the external expert (S660).

Embodiments of the present invention disclosed in the specification and drawings are only specific examples to easily explain the technical contents of the present invention and aid the understanding of the present invention, and are not intended to limit the scope of the present invention. It will be apparent to those skilled in the art that other modifications based on the technical idea of the present invention can be carried out in addition to the embodiments disclosed herein.

Claims (14)

In how the server builds a rules-based knowledge database,
Selecting a data set from an external data set source;
Generating a classification model using a decision tree algorithm corresponding to the data set;
C) extracting conditions and results from the classification model;
Generating an XML model using the conditions and results;
E step of extracting detailed information of the XML model;
F generating and verifying a production rule using the detailed information;
Repeating steps a through f to store the production rules generated in one or more data sets in the knowledge database.
The method of claim 1,
And the decision tree algorithm comprises at least one of BFTree, J48, RandomTree, REPTree, or SimpleCart.
The method of claim 1,
Step c,
Using a text trimming, text segmentation, and special character substitution technique in said classification model to remove header information of said classification model and extract conditions and results.
The method of claim 1,
The d step,
And converting the XML model into the XML model by performing an operator array, indentation, and file conversion operation on the classification model.
The method of claim 1,
The step e,
Analyzing the XML model using a DOM analyzer;
And generating the production rule including attributes, operators, and values using the analysis results of the XML model.
The method of claim 1,
The f step,
Verifying the production rule by an external expert;
Verifying that the production rule is operating normally.
The method of claim 6,
Checking the production rule,
Checking a rule by comparing the production rule with an existing production rule previously stored in the knowledge database;
And providing the overlapping rule to the external expert if there is a overlapping rule in the production rule and the existing production rule.
In a system for building a rule-based knowledge database,
A model generator for selecting a data set from an external data set source and generating a classification model by selecting a decision tree algorithm corresponding to the data set;
A preprocessor extracting a condition and a result from the classification model;
A transformation unit generating an XML model using the conditions and results;
A rule generation unit for generating a production rule by extracting and analyzing detailed information of the XML model and verifying the production rule;
Knowledge database building system comprising a knowledge database for storing the production rule.
The method of claim 8,
The decision tree algorithm includes at least one of BFTree, J48, RandomTree, REPTree, or SimpleCart.
The method of claim 8,
The preprocessing unit,
A knowledge database construction system for removing header information of the classification model and extracting conditions and results by using text trimming, text segmentation and special character substitution techniques in the classification model.
The method of claim 8,
The conversion unit,
A knowledge database construction system for generating the XML model by performing an operator array, indentation, and file conversion on the classification model.
The method of claim 8,
The rule generation unit,
An XML model analyzer for analyzing the XML model using a DOM analyzer;
And a production rule generator for generating a production rule including attributes, operators, and values using the analysis result of the XML model.
The method of claim 8,
The rule generation unit,
A rule verifier for verifying the production rule by an external expert;
And a production rule verifying unit including a rule checking unit for confirming whether a configuration of the production rule is correctly written.
The method of claim 13,
The rule check unit,
The rule is checked by comparing the production rule with the previously stored production rule.
And when the rule is confirmed, if there is a rule overlapping the production rule and the existing production rule, providing the overlapping rule to the external expert.

Images