Efficient Algorithms for Mining Frequent Target Episodes from Complex Event Sequences

• Main Authors: Pei-WenJiang, 蔣佩雯
• Other Authors: Shin-Mu Tseng
• Format: Others
• Language: en_US
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/18031867620347458599

碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 101 === Mining frequent episodes from event sequences is an important topic in data mining fields. Most of existing related researches focused on mining frequent episodes from a single event sequence. However, sequences containing simultaneous events are frequently encountered, and we refer to such sequences as complex event sequences. For some practical applications, users are often interested in target episodes where the last event of an episode is the target event type. In this thesis, we address the problem on mining frequent target episodes from complex event sequences. To avoid restricting the length of episode, the support of an episode is counted by the size of minimal occurrences set. We extend the state-of-the-art algorithm PPS to PPS+ as a base method for mining episodes from complex event sequences, and propose the EM-SES algorithm to overcome the drawback of PPS+. Furthermore, the TEM-SES algorithm is proposed to reduce the cost of mining a complete set of frequent episodes. Experimental evaluation on both synthetic and real datasets demonstrates that our algorithms deliver good performance in terms of efficiency and effectiveness.