- 2017 IEEE. This paper compares alternative user interest models created by aggregating an individual's interest expressed through their interactions with multiple everyday applications. A local service unobtrusively observes user interactions with these applications as well as the content authored, annotated, and consumed in them to understand the user interests expressed through these applications. An open question is the relative importance of the authored-text, annotated-text, and implicit feedback generated in each application when identifying users' real interests. This paper evaluates the effectiveness of the recommendation support from semi-explicit user interest models (authored/annotated text) and unified user interest models (implicit feedback + semiexplicit). Results indicate both these models are successful in allowing users to locate the content easily based on subtle changes of user's indirect and semi-direct interest indicators.