| Summary: | 碩士 === 國立高雄大學 === 資訊工程學系碩士班 === 101 === Sequential utility pattern mining is an interesting and emerging issue in knowledge discovery with wide applications. It considers not only the quantities and profits of items but also the occurrence order of items in sequences to get useful patterns, which may then help managers find out potential marketing knowledge and make appropriate decisions. In the first part of the thesis, we handle quantitative sequential utility pattern mining with the consideration of quantitative values of items. The problem is more difficult than traditional sequential utility pattern mining since it has to discover proper quantitative intervals of items and doesn’t satisfy the downward-closure property. Hence, an effective upper-bound model is developed to keep the property, and a merging method is designed to decide quantitative intervals of items in sequences. Moreover, to make the patterns simple and comprehensible to human beings, in the second part of the thesis, the fuzzy set concept is adopted to find interesting high utility se-quential patterns with linguistic meaning. We thus introduce another new research issue named fuzzy utility sequential pattern mining. In particular, a new fuzzy utility function is developed to evaluate the linguistic benefit of a pattern in a sequence, and an effective fuzzy upper-bound model is developed to reduce the search space of finding high fuzzy utility sequential patterns from a quantitative sequence database. The experimental results show the effectiveness of the two types of patterns and the performance of the proposed mining approaches under different parameter settings.
|
|---|
