Extracting Sequential Patterns from Progressive Databases: A Weighted Approach

摘要

Research on pattern mining has deduced that progressive sequential pattern mining approach can be used to obtain the most updated frequent sequential patterns. However, no existing sequential pattern mining algorithms provide a metric to quantify the importance of the extracted sequential patterns. The support count, which can be used as metric, may be altered to assign priorities to patterns by assigning weights to individual items or to specific timestamps in the period of interest. This paper proposes a method to assign weights to patterns using the fact that, the time period over which a pattern is spread affects the significance of the pattern. As the period over which the pattern spans increases, the probability of the occurrence of the pattern reduces. In order to increase practical usage, the method also assigns importance to timestamps, so that the presence or absence of a pattern on that timestamp may help to weigh the pattern. The weighted patterns may hence be obtained by modifying the support count of a pattern by measuring the time period over which the pattern occurs.