Mining Access Logs Using Visual Linking of Entity Characterizations

Abstract We illustrate a technique that facilitates visualization of meaningful relationships between entities represented in large multidimensional databases. Our approach uses entity characterizations, which are textual descriptors combined with statistical metrics. As a data mining technique, visually linking entity characterizations integrates information across multiple entities, allowing a user with significant domain familiarity to “see” patterns in the data that would not otherwise be apparent. The technique described here supports navigation within the data, while keeping a clear view of the distinct entities represented. We use access log analysis as an example problem, outlining our approach in this domain.