Constructing Temporal Equivalence Partitionings for Keyword Sets

Abstract

Identifying keyword associations from text and search sources is often used to facilitate many tasks such as understanding relationships among concepts, extracting relevant documents, matching advertisements to web pages, expanding user queries, etc. However, these keyword associations change continually change with time. In this paper, the authors define an equivalence relationship among keywords and develop methods to construct a temporal view of the equivalence relationship by constructing optimal temporal equivalence partitionings for keyword sets. They describe efficient algorithms to construct an optimal temporal equivalence partitioning for a keyword pair. They use the fact that the equivalence relationship is transitive to extend these algorithms to obtain an optimal temporal equivalence partitioning for a larger set of keywords. The authors show the effectiveness of the approach by constructing the temporal equivalence partitionings of several sets of keywords from the Multi-Domain Sentiment data set and the ICWS2009 Spinn3r data set.