Belief Dynamics and Biases in Web Search

Abstract

We investigate how beliefs about the efficacy of medical interventions are influenced by searchers' exposure to information on retrieved Web pages. We present a methodology for measuring participants' beliefs and confidence about the efficacy of treatment before, during, and after search episodes. We consider interventions studied in the Cochrane collection of meta-analyses. We extract related queries from search engine logs and consider the Cochrane assessments as ground truth. We analyze the dynamics of belief over time and show the influence of prior beliefs and confidence at the end of sessions. We present evidence for confirmation bias and for anchoring-and-adjustment during search and retrieval. Then, we build predictive models to estimate postsearch beliefs using sets of features about behavior and content. The findings provide insights about the influence of Web content on the beliefs of people and have implications for the design of search systems.