Generating Pseudo Search History Data In The Absence Of Real Search History

Previous studies in Information Retrieval literature have shown that users' search history can be leveraged to improve current search results. However sometimes we have little to no search history available. In such cases, it would be helpful to obtain data similar to search history data. One way of doing this is by simulating previous search interactions. In the present study, we focus on generating simulated "related queries" that can serve as an additional source of information about the current search [1]. Assuming that users reformulate their queries by leveraging some of the terms and key phrases they find in ranked documents during their search, we proposed simple models for generating such related queries.