Towards context-aware personalization and a broad perspective on the semantics of news articles

Abstract

We analyze preferences and the reading flow of users of a popular Austrian online newspaper. Unlike traditional news filtering approaches, we postulate that a user's preference for particular articles depends not only on the topic and on propositional contents, but also on the user's current context and on more subtle attributes. Our assumption is motivated by the observation that many people read newspapers because they actually enjoy the process. Such sentiments depend on a complex variety of factors. The present study is part of an ongoing effort to bring more advanced personalization to online media. Towards this end, we present a systematic evaluation of the merit of contextual and non-propositional features based on real-life clickstream and postings data. Furthermore, we assess the impact of different recommendation strategies on the learning performance of our system.

News Recommending Based on Text Similarity and User Behaviour

“…As it was mentioned in paper on news recommending (Jancsary et al, 2010) there are context-sensitive features. To involve these aspects we need to find them in real-time.…”

Towards a context-sensitive online newspaper

“…However, the features that are extracted in this step partially determine the power of our hypothesis space. In previous work [8], we motivated and determined of number of such characteristics. Moreover, we recently conducted a psychological study in order to gain further insights into the factors that really drive user preference in the domain of news articles.…”

“…In earlier work [8], we experimented with the classic Perceptron [12], the family of Passive-Aggressive online algorithms [2] and the Forgetron [3]. The latter can also incorporate a Mercer kernel in order to overcome the linear decision boundary of (2).…”

“…The hyperparameter C controls the penalization of the constraint violation; larger values of C result in stricter enforcement of the constraint and hence more aggressive updates. Previously [8], we found PA-II to be robust in this respect, and we simply use C = 100 in practice.…”

“…In earlier work [8], we presented a study regarding the contribution of a variety of characteristics of articles-and to some extent the user context-to user preference. However, the study was based on clickstream logs and comment postings that had been collected by DERSTANDARD.AT prior to implementation of MAGNIFICENT.…”