In mainstream search systems, search interactions typically assume the form of search queries and result selections. This information can be useful as implicit feedback to improve search performance when aggregated across many searchers (Joachims, 2002; Agichtein et al., 2006). Implicit feedback can also be collected at a personal level and used to update the search experience directly in real time (e.g., rearranging available information [White et al., 2005a; White et al., 2005b]), or to tailor the results for the current query using personalization or contextualization (Dou et al., 2007; Teevan et al., 2011b; Bennett et al., 2012). Trails that people follow through document collections (Bush, 1945; Bilenko and White, 2008) and trail destinations (White et al., 2007) can also be employed to help searchers understand the corpus and the domain (White and Huang, 2010), and ultimately to attain their search goals. In addition to being used to refine search results or other features (such as related searches or query auto-completions [shown dynamically as searchers compose queries]), recorded traces of search interaction can also be used as a diagnostic tool to understand when searchers are satisfied or frustrated with their search experience (Hassan et al., 2010; Aula et al., 2010; Feild et al., 2010), and consequently, where the search system could improve its performance or provide missing search support. Sections 2.12.3 discuss components of the search interaction process, including SERP interactions and sequences of actions extending beyond the search engine and into the corpus being searched.

There is a growing set of opportunities for search engines to learn from aggregated search activity as a new range of interface features emerge to integrate new capabilities, such as touch and gesture. Section 2.4 discusses mechanisms for collecting and representing interests and intentions beyond queries and clicks. Other advances, such as eye-gaze tracking and spoken dialog, will change both the manner and the settings in which people interact with search systems. Because interactions depend on both software and hardware, device-dependent models of relevance may be needed to accurately associate interaction events on each device with searchers’ common interests and intentions. Some progress has been made on modeling relevance on mobile devices using touch interactions (Guo et al., 2013), but more research in this area is anticipated.