Size Matters: Finding the Most Informative Set of Window Lengths

摘要: Event sequences often contain continuous variability at different levels. In other words, their properties and characteristics change rates, concurrently. For example, the sales of a product may slowly become more frequent over period several weeks, but there be interesting variation within week same time. To provide an accurate robust "view" such multi-level structural behavior, one needs to determine appropriate levels granularity for analyzing underlying sequence. We introduce novel problem finding best set window lengths discrete event sequences. define suitable criteria choosing propose efficient method solve problem. give examples tasks that demonstrate applicability present extensive experiments on both synthetic data real from two domains: text DNA. find optimal sets themselves can new insight into data, e.g., burstiness events affects measuring frequencies.