Behaviour-driven clustering based on event-sequence similarity metrics
2010 (English)Manuscript (preprint) (Other academic)
When analysing event data two key objectives are to first identify interesting subsequences in the data records and then to retrieve groups of records that exhibit similar behaviour. This is especially true when the focus of the exploration is the human, for example when using activity diaries to reveal sub-populations with similar behaviour, medical records to identify groups with similar medical conditions, or web sessions to find groups with similar web-surfing habits. In this paper we propose a visual exploration approach, based on sequence similarity metrics and clustering techniques, that will allow an analyst to interactively explore the distribution of sequences along event data records as well as group the results according to user-selected similarity preferences. We have identified a set of similarity metrics that are specific to event-sequences which we use as input into a clustering algorithm. The user can choose which metrics to use and assign weighting factors to them, which results in groupings that exhibit similar behaviour according to their definition of similarity and interestingness. The resulting clusters can be interactively explored in a multiple linked-view environment showing the clusters, the cluster quality, the similarity metrics and meta (background) information describing the clustered individuals in order to make comparisons within and between groups. Using such an interactive approach that considers user preferences and takes advantage of background knowledge gives a basis for enhanced analytical reasoning by providing a more complete understanding of the retrieved groupings and can lead to a more thorough analysis and accurate assessments.
Place, publisher, year, edition, pages
Event-based data, activity diary data, event sequences, similarity metrics, clustering, interactive exploration
Engineering and Technology
IdentifiersURN: urn:nbn:se:liu:diva-58310OAI: oai:DiVA.org:liu-58310DiVA: diva2:338116