Discovering Context-Topic Rules in Search Engine Logs

We present a class of rules, called context-topic rules, for discovering associations between topics and contexts, where a context is defined as a set of features that can be extracted from the log file of a Web search engine. We introduce a notion of rule interestingness that measures the level of relevance of the topic within a context, and provide an algorithm to compute concise representations of interesting context-topic rules. Finally, we present the results of applying the methodology proposed to a large data log of a search engine. Joint work with Mark Levene.