Predicting User Engagement with Direct DisplaysrnUsing Mouse Cursor Information

摘要

Predicting user engagement with direct displays (DD) isrnof paramount importance to commercial search engines, asrnwell as to search performance evaluation. However, under-rnstanding within-content engagement on a web page is not arntrivial task mainly because of two reasons: (1) engagementrnis subjective and different users may exhibit different be-rnhavioural patterns; (2) existing proxies of user engagementrn(e.g., clicks, dwell time) suffer from certain caveats, suchrnas the well-known position bias, and are not as effective inrndiscriminating between useful and non-useful components.rnIn this paper, we conduct a crowdsourcing study and exam-rnine how users engage with a prominent web search enginerncomponent such as the knowledge module (KM) display. Tornthis end, we collect and analyse more than 115k mouse cursorrnpositions from 300 users, who perform a series of search tasks.rnFurthermore, we engineer a large number of meta-featuresrnwhich we use to predict different proxies of user engagement,rnincluding attention and usefulness. In our experiments, werndemonstrate that our approach is able to predict more ac-rncurately different levels of user engagement and outperformrnexisting baselines.