Understanding searchers' query dynamics online

Abstract

An important issue in studying relevance judgment is order effect, which refers to the different relevance judgment of a document when it appears in different positions in a list. Part I proposes a set of order effect forming mechanisms including the learning effect, sub-need scheduling effect, and cursoriness effect based on the conceptualization of dynamic relevance and elaboration likelihood model. The empirical study in an interactive information retrieval setting indicates that the actual pattern of order effect conforms to the combinatory effect of the three mechanisms. Part II propose and evaluate a novel method for automatic query expansion by modeling usersa?? search plan and manifested query specifications with Hidden Markov Model (HMM). Experimental results show that our HMM method has achieved competitive performance against query-similarity based query expansion method and outperforms well known pseudo relevance feedback technique.