Adapting web sites by spreading activation in ontologies

In this paper we introduce SPREADR, a model-based tech- nique for creating context-adaptive web applications. In this approach, the domain knowledge as well as context factors are represented by an ontology. Context factors depend on the application domain and may include location, time, user role, weather or any other relevant context information, of- ten information which can automatically be sensed. Both domain objects and context factors are treated as concepts or instances in the ontology, which can be linked to each other, thus forming a semantic network. In addition to this repre- sentation, we introduce scalar activation levels for concepts as well as weights for relations. Concept activation repre- sents the level of user interest whereas activation of con- text factors can be seen as the level of fulfilment based on some measurement. The structure of the semantic network remains the same for each user while activation levels may differ. Recognition of context factors or user actions trigger an activation flow through the network, thus increasing the activation of contextually important nodes. Relations are as- sociated with different weights resulting in different amounts of activation spreading along different relations. The result- ing spreading activation network can be used both for gen- erating a web site as well as for controlling the adaptive be- haviour of the system. Adaptation may include effects such as sorting navigation items by relevance, showing or hiding information depending on the activation value and highlight- ing or recommending important items. The system is also able to learn user preferences through implicit feedback. We present a demonstrator application as well as initial evalua- tion results.