| Summary: | 碩士 === 國立中山大學 === 資訊工程學系研究所 === 99 === Data mining is an important process of extracting desirable knowledge from existing databases for specific purposes. Nearly all transactions in real-world databases involve items bought, quantities of the items, and the time periods in which they appear. In the past, temporal quantitative mining was proposed to find temporal quantitative rules from a temporal quantitative database. However, the quantitative values of items are not suitable to human reasoning. To deal with this, the fuzzy set theory was applied to the temporal quantitative mining because of its simplicity and similarity to human reasoning. In this thesis, we thus handle the problem of mining fuzzy temporal association rules from a publication database, and propose three algorithms to achieve it. The three algorithms handle different lifespan definitions, respectively. In the first algorithm, the lifespan of an item is evaluated from the time of the first transaction with the item to the end time of the whole database. In the second algorithm, an additional publication table, which includes the publication date of each item in stores, is given, and thus the lifespan of an item is measured by its entire publication period. Finally in the third algorithm, the lifespan of an item is calculated from the end time of the whole database to its earliest time in the database for the item to be a fuzzy temporal frequent item within the duration. In addition, an effective itemset table structure is designed to store and get information about itemsets and can thus speed up the execution efficiency of the mining process. At last, experimental results on two simulation datasets compare the mined fuzzy temporal quantitative itemsets and rules with and without consideration of lifespans of items under different parameter settings.
|
|---|
