You are here

News Recommenders: Real-Time, Real-Life Experiences


Doychin Doychev, Rachael Rafter, Aonghus Lawlor, Barry Smyth

Publication Type: 
Refereed Conference Meeting Proceeding
Recommender systems have become an essential part of our daily lives. In this work we present our early experiences when it comes to a real-world news recommendation task in the guise of the 2014 NewsREEL challenge [6]. NewsREEL participants respond to recommendation requests with their suggestions and, crucially, received live-user feedback in the form of click-through data, providing an opportunity for a large-scale evaluation of recommender systems in the wild. In contrast to other domains (books, movies, etc.), news has a greater item churn [2,4] and users are much more sensitive to article recency [4]. Furthermore, user profiles are typically constrained to the current session [11] or at best defined by browser cookies [7]. Moreover, it is atypical for users to evaluate articles explicitly and feedback is commonly collected implicitly by observing user behaviour [8]. Finally, there is a low consumption cost associated with reading a news article which can result in a larger than normal diversity of the consumed items [10]. Within the context of the NewsREEL challenge [5] (formerly, The News Recommender System Challenge (NRS’13)), Said et al. [9] study the performance of similarity- and recency-based algorithms. They find that both types perform slightly better when recommending in general news sites than in more topic focussed sites. Others [4,5] recommend representing a news article using a combination of metadata and contextual features; see [7] for a full list of such features available for the NewsREEL challenge.
Conference Name: 
The 23rd Conference on User Modelling, Adaptation and Personalization
Digital Object Identifer (DOI): 
Publication Date: 
Conference Location: 
Research Group: 
National University of Ireland, Dublin (UCD)
Open access repository: