Decentral Item-Based Collaborative Filtering for Recommending Images on Mobile Devices

Decentral recommender systems appear well suited for mobile scenarios, but have not been investigated thoroughly or implemented very much so far. We have designed and implemented a system to recommend images on personal digital assistants (PDAs). Our approach also incorporates recommending for groups of users that are present at a public shared display. The system exchanges rating vectors among PDAs, computes local matrices of item similarity and utilizes them to generate recommendations. Our innovation in comparison to existing systems includes improving the extensibility of the data model by introducing versioned rating vectors. In addition, we have optimized the storage requirements on the mobile device. We have evaluated the approach in a small user study. Furthermore, the scalability of our system was analyzed using a standard recommender data set resulting in positive findings.