Characterizing Semantic Relatedness of Search Query Terms

Mining for semantic information in search engine query logs bears great potential for both the optimization of search engines and bootstrapping Semantic Web applications. The interaction of a user with a search engine (more specifically clicklog information) has recently been viewed as implicit tagging of resources by query terms. The resulting structure – previously called a logsonomy – exhibits structural similarities to folksonomies, which evolve during the expclicit process of annotating resources with freely chosen keywords in social bookmarking systems. For the folksonomy case, appropriate measures of relatedness have shown to be capable to harvest the emerging semantics inherent in the tripartite graph of users, tags and resources. Motivated by the reported structural similarities, in this work we extend this methodology to logsonomies. More specifically, we apply several measures of query term relatedness to the logsonomy graph and provide a semantic characterization for each measure by grounding it against user-validated relatedness measures based on WordNet. Comparing the outcome with prior results of analyzing folksonomy data we find that the formalization of log data in logsonomies retains the semantic information. Some relatedness measures we applied prove to be able to capture these emergent semantics similarly to the folksonomy case, while others exhibit different characteristics. In this way we provide a novel and systematic approach to compare the emergent semantics of user interactions with search engines and social bookmarking systems. We conclude that the type of semantic information inherent in both emerging structures is similar, and inform the choice of an appropriate measure of query term relatedness for a given task.