Current TV recommender systems typically rely on genre or actors, e.g., recommendations referring to other “suspenseful genres” or “documentaries”. However, less obvious relationships between programs (e.g., relationships between actors and directors, or settings and locations) exist that can be found in linked data and exploited for more serendipitous recommendations. However, to date, little is known how these “linked data patterns” can be exploited to maximize the serendipity of TV recommendations. Accordingly, the present project seeks to fill this research gap. More specifically, we ask:
- Back-end: What linked data patterns effectively maximize serendipity? And how should linked data patterns be ranked in order to maximize serendipity? (Computer Science)
- Front-end: What discourse profile for recommendations enhances serendipity? What content and form of TV recommendations results in maximum serendipity? (Language & Cognition)
- Effect: How do TV recommendations that differ in their underlying linked data patterns and portrayals affect user’s perceptions of serendipity and subsequent TV choice behavior? (Communication Science)
Two empirical studies are central to this project:
- A conjoint analysis (online, MTurk) with “made-up/exemplars” of recommendations will allow to test the relative importance of a wide range of different features of them (e.g., both on back-end and front-end level) on serendipity. The analysis will illuminate the most important features for achieving serendipitous TV recommendations.
- A subsequent usability test (online experiment; MTurk) will (1) experimentally vary those most important features and (2) actual design those features. Accordingly, the experiment/usability test will examine the effect of actual/fully functional TV recommendations on serendipity and users’ TV choice behavior.