SPREADR

As the predecessor of DISCOVR, SPREADR was a prototype of a context-aware shopping portal incorporating a series of spreading activation-based adaptation techniques to recommend products to the user that would not only fit to his or her personal preferences, but also to the current usage context. Starting with a set of automatically determined internal or external context factors like the user’s click history, the current season or upcoming holidays, the system spreads ‘activation energy’ to related concepts that are modeled using a pre-defined vocabulary in an ontology. For example, if the user has been looking at a series of DVD movies and Christmas holidays are approaching, he or she might possibly be interested in Christmas DVDs because these concepts are semantically related with each other. Thus, in contrast to other recommending techniques like collaborative filtering, spreading activation directly mimics the way in which the human brain works, by having certain ‘neurons’ co-fire if they share the same input triggers.

In SPREADR, this approach was additionally enhanced with a rule-based reasoner that would also work upon the given ontology, as well as an explanation and feedback mechanism where the user could request an explanation of how the presented recommendations have been generated, and in turn tell the system whether the provided recommendations were good or not. Making use of its integrated learning mechanism, the system could thus adapt to the user by strengthening those types of connections between semantic concepts that were of special interest to the user. In a user study, it was shown that the system in the given scenario produced significantly better recommendations than traditional recommendation methods.

Scope

WISE

Novel approaches for Web Information and Service Engineering

Contact

Jürgen Ziegler

Full Professor

Contributors

Tim Hussein

Former team member

Werner Gaulke

Former team member

Timm Linder

Former team member

Publications

Explanation of spreading activation based recommendations

Context-aware Recommendations

Adapting web sites by spreading activation in ontologies

Context-adaption based on ontologies and spreading activation