Many online news publishers display on the bottom of their articles a small widget box labelled “You might also be interested in”, “Recommended articles”, or similar where users can find a list of recommended news articles. Dependent on the actual content provider, these recommendations often consist of a small picture and accompanying text snippets.
While some publishers provide their own recommendations, more and more providers rely on the expertise of external companies such as plista, a data-driven media company which provides content and advertising recommendations for thousands of premium websites (e.g., news portals, entertainment portals). Whenever a user reads an article on one of their customers’ web portals, the plista service provides a list of related articles. In order to outsource this recommendation task to plista, the publishers firstly have to inform them about newly created articles and updates on already existing articles on their news portal. In addition, whenever a user visits one of these online articles, the content provider forwards this request to plista. These clicks on articles are also referred to as impressions. Plista determines related articles which are then forwarded to the user and displayed in above mentioned widget box as recommendations.
Having a large customer base, plista processes millions of user visits in real time on a daily basis. NEWSREEL provides research teams the opportunity to deliver some of these recommendations. In the second iteration of NEWSREEL which is organised as a campaign-style evaluation lab of CLEF 2015, we provide two tasks that address the challenge of real-time news recommendation. The first task allows benchmarking news recommendation algorithms in a living lab environment. The second task simulates the real-time recommendation task using a novel recommender systems reference framework which has been developed within the FP7 project CrowdRec.