What Affects Word Changes in Query Reformulation During a Task-based Search Session?

This paper performs an analysis on the influence of different factors on users' choices of specific word changes in query reformulation during a search session. We study three types of word changes: whether to remove or retain a word in the current query; whether or not to add a brand-new word to the query; whether or not to reuse a word (included in previous queries, but removed in the current query). Three types of factors are examined: session-level factors measuring task and user characteristics; query-level factors related to past user activities in a session; word-level factors for the characteristics of the examined word and its relation to the current query and search results. Statistical analysis suggests that: word-level factors strongly influence all three types of word changes; query-level factors only show a clear influence on retaining or removing a word; task-level factors exhibit limited direct influence on all three types of word changes. Analysis also disclose reasons for different word changes: users remove a word to stop exploring a subtask, or to correct bad performing queries; they look for related, unused words from recently viewed result summaries and add to queries; reusing a word usually indicates reverting from a subtask to the main task or another subtask.