Personalized Information Access Using Semantic Knowledge - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 182

Fig. 7.1 SERUM interface showing recommended news articles and recognized entities

relevant for modeling the user behavior such as user clicks or mouse-over events.
Events, triggered by the user (e.g., clicks), are linked to news articles and named
entities (e.g., artists in the news article) the user interacted with.
Based on a statistical interaction analysis the user behavior events are aggregated
to identify named entities (e.g., topics, musicians, and genres) the user is interested
in. The analysis includes the last n sessions of the user (in our current system n is set
to five) where the interaction of a user is analyzed and the entities are ranked accord-
ing to the interaction frequency. The analysis also includes a time aspect where an
interaction has a higher weight if the session is a current one. Furthermore, we deploy
semantic data (from Freebase) to extend the knowledge about identified named enti-
ties to produce a richer user model. Thus, musicians recognized to be interesting to
the user are expanded with data about produced albums and collaborating artists. For
example, if the user only stated interest in “Madonna,” we can add genre information
(e.g., pop) and information about collaborations with other artists. These enriched
user profiles are used as the input for our graph-based recommender. The more