An efficient interval-based sequential pattern mining

• Main Authors: Jiang, Ji-Chiang, 姜季強
• Other Authors: Lee, Suh-Yin
• Format: Others
• Language: en_US
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/96142387729499523193

碩士 === 國立交通大學 === 資訊科學與工程研究所 === 98 === Existing sequential pattern mining algorithms assume that events occur instantaneously. However, events in real world applications usually have durations which are called interval-based events. But complex relationship among event intervals causes difficulty in designing an efficient interval-based event mining algorithm. Therefore, the concept of “coincidence-slice” is proposed to solve the problem caused by the complex relationship among event intervals. The event intervals are incised to disjoint smaller “event slices” according to the coincidences among event intervals, that is, several event slices may occur in the same time period called “coincidence”. Therefore, an original event sequence can be represented as a list of ordered “coincidences” which contains event slices. This new representation proposed is called “coincidence sequence representation”. We transform the problem of complex relationship among event interval to consider the simple relationship among event slices. The proposed interval-based sequential pattern mining algorithm called CTMiner is based on the “coincidence sequence representation”. The CTMier also uses the concept of well-known sequential pattern mining algorithm PrefixSpan to find temporal patterns without candidate generation. Finally, to comprehend the frequent temporal pattern represented by “coincidence sequence representation”, we discover and use relation list to present all the relationships in a pattern. We also implement some pruning strategies to improve the performance of CTMiner by considering the characteristics of the “Coincidence-slice”. Experiments on both synthetic datasets and real dataset of library lending indicate the efficiency and scalability of the proposed algorithm.