Two-Step Filtering for Rules without Overlapping Conditions

A rule-based system is used in artificial intelligence applications which extract information from data with relatively simple rules. As the rule-based system does not require a learning phase, it has the advantage that it operates immediately when the system is installed. Especially, a rule matching requires a complex process. The previous rule matching algorithm achieves the high performance when the rules have many overlapping conditions. This paper describes an indexing method for reducing the rule retrieval time when the conditions in rules are not overlapped. The proposed method generates a multilevel indexing tree using different values at each level. The number of sensing types in the conditions is used at the second level and the number of ranges of each sensing value is used at the third level. The experimental results show that the proposed method can achieve both the fast search and the high accuracy of the search result at the same time. Its searching time is about twice faster than that of PHREAK method.