Proceedings of the 2013 International News Recommender Systems Workshop and Challenge

News article recommendation differs in several ways from other well-known types of recommender systems such as for music and movies. First, freshness represents an important aspect. Sometimes, freshness is deemed more important than relevancy. Second, similarity between news articles does not necessarily reflect their relatedness. For instance, two news articles might share a majority of words. Still, their actual topic might differ. Third, news are typically published in a rather unstructured format. In contrast, structured data such as social graphs facilitate pre-processing steps. Fourth, news readers might have special preferences on some particular events which recommender systems can barely predict. Fifth, serendipities (i. e., variety in recommended news articles) represent a crucial property of a news recommender system. Contrarily to music recommendations, users avoid to re-consume an item. Thus, news recommender systems are required to provide diverse sets of items in order to assure not to recommend monotonously. Sixth, breaking or trendy news might have a high relevance even though the appear completely unrelated to the individual user profile. Seventh, the has not yet established consensus on how to evaluate news recommender systems. We typically observe implicit preferences as users interact with news portals. Those preferences do not exhibit a graded relevance. Thus, well-established evaluation criteria based on ratings (e.g., root mean squared error) cannot be applied. Another set of challenges arises from the context of news recommendation. Recommender systems are known to struggle with so-called "cold-start users" (i. e., users form whom no preferences are available yet). News portals typically refrain to require users to login prior to read news articles. Hence, there is a large fraction of users who appear to be "cold-start users". The lack of sufficiently many interaction to established trustworthy user profiles entails further challenges. Inferring interest signals suffers from incomplete profiles. Additionally, items' relevance is time dependent and diminishes with time progressing. Another challenge related to the context arises from the increasingly frequent use of mobile devices to read news articles such as tablets and smart phones. Those have a limited space available to display news and related recommendations. News recommender systems have to deal with this issue thus applying suited layout mechanisms to fit the content to the available screen. Finally, news recommender systems face numerous technical challenges. Those challenges include minimizing response time improve user experience, scaling to the large amount of requests to avoid time outs, providing flexibility to incorporate new recommendation methods or adjust parameter settings for existing implementations, and guarantee a reliable service whom user can access at any time