Recent improvements in computing power and ambient technologies have opened up new perspectives in ambient intelligence and proactive software systems. The need emerges for an ambient and supportive system that deeply understands the user's needs and preferences under the current context. In this work, we propose a recommendation system that aggregates and understands the current situation and continuously learns from the user’s decision history. To leverage a rich pool of contextual data, the system dynamically changes the strategy and scope for contextualization and encodes multimodal data into unified embeddings. A contextual similarity database consisting of these embeddings is leveraged for finding, ranking and comparing scenarios. The proposed recommender is intended to fit into a decentralized service system and addresses challenges in application interaction, user engagement, user privacy and scalability.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.