Multifaceted Analysis of News Articles for Intelligent User- and Context-Sensitive Presentation
Little is known in hard facts about why readers of online newspapers prefer some articles over others. Current news filtering systems assume that the topic of an article is the only factor that determines user satisfaction. But content accounts only for about 40% of a story's satisfaction rating. Factors that determine the remaining 60% can be as diverse as readability concerns, writing style, the type of a story, visual complexity, proper use of photographs, or, even less concretely, the appeal of a story. Contextual information, like previously read articles or the overall popularity and recentness of articles, needs to be considered as well.
The goal of MAGNIFICENT is to gain deep insight into both the relevant parameters of stories and the adaptive training of user profiles along these parameters. Ultimately, we want to improve a user's reading experience through personalized presentation of articles, taking into account personal reading behavior and use of the medium. From a broader perspective, our approach emphasizes a more comprehensive view of semantics. The meaning of a text goes far beyond its basic propositional content. It comprises a rich mix of diverse factors, including rhetorical structure, style, standpoint, and many other aspects writers routinely convey to their readers.
The departure from merely modelling the topical preferences of a reader, however, leads to a situation where commonly used techniques from information retrieval no longer seem appropriate. We will thoroughly investigate which parameters of news articles determine reader preference, rather than topical relevance. Then, robust and scalable machine learning techniques have to be selected and applied that are flexible enough to incorporate the additional information. Statistical models that capture a reader's behavior through implicit or explicit feedback will be trained and adapted regularly.
FIT-IT Programme (Semantic Systems)