| Information and Knowledge
Technologies
|
| |
|
|
A Survey of Publications on Sequential Patterns by Kurt Rohloff
Last updated Mar. 7, 2009
This page is intended to contain my survey of papers on computational approaches to discovering, manipulating, processing and forecasting sequential patterns. This list is by no means exhaustive. I'll post papers as I find them and post a few lines describing these works as I read about them. If you have anything that you'd like me to read and/or post, by all means send it to me!
Note: I wrote some of the papers posted on this page and I provide links to pdf versions of these papers. I didn't write most of the papers listed on this page. For the papers I didn't write, I provide links to where most people should be able to find a copy (or a link to a copy) of the paper - most often this is Google Scholar.
This position paper discusses the uses of a sequential pattern methodology in the context of an informatin system to forecast socio-political violence. The system takes in disparate data sources, fuses them into a centralized repository and then uses the sequential pattern methodology discussed here to discover patterns in this data.
This paper introduces a methodology to concisely describe an event sequence patterns. The sequence patterns are constructed from a set of events that can occur at discrete sample times. The methodology discussed in the paper breaks the sequences down into subsequences and describes the event occurrences using a "rate of occurrence" description. The subsequence deconstruction operates using a Minimum Description Length principle. In later work the authors develop the EventSummarizer tool which implements their event sequence discovery process.
This paper introduces a sequential pattern discovery methodology based on backwards chaining. Sequential patterns are defined as quantized combinations of multiple factor values sampled at regular time intervals preceding Events of Interest (EoIs). Patterns are discovered by looking for multiple cases where combinations of equivalent factor values occur at the same times before equivalent EoIs and this pattern does not appear before the non-occurrence of an EoI. Several example patterns are for socio-political violence EoIs such as rebellions.
This paper expands upon previous work for the sequential pattern methodology where sequential patterns are defined as quantized combinations of multiple factor values sampled at regular time intervals preceding Events of Interest (EoIs). The concept of a cover is introduced which is a set of patterns where the for every EoI case in training data, there is at least one pattern that describes the behavior preceding those EoIs and the patterns do not occur before the non-occurrence of an EoI. An example cover is given based on the outbreak of socio-political violence.