Explaining and Combining Recommender Algorithms for Decentralized Architectures (ENCORE)
Collaborative filtering is a new method for retrieving information. Instead of basing the search on the content in the information units, a collaborative filter or recommender system uses databases of user ratings as a basis for predictions. From our user studies we see that collaborative filtering can be of great help to users when they search for information. In order to be truly useful collaborative filtering needs more research. Specifically, we have seen that it can be improved by exploring new ways of explaining to users why they get a particular recommendation. Secondly, for some application scenarios, collaborative filtering has to work in networks where there is no central server Ð thus requiring a decentralized recommender model. Finally, it is necessary to find ways of combining traditional content-based methods for information filtering with collaborative information to improve the filtering process and solve some of the problems with collaborative filtering.
ENCORE aims to tackle these three problem areas and package the resulting solutions in a general-purpose recommender platform. The platform will be applied in two very different applications to investigate its feasibility and integration qualities in different domains and for different tasks. Together with IFS AB (Industrial and Financial Systems AB) we will integrate the platform in their main product IFS Applications, which is a professional system for customer management and resource planning. The second application is a research prototype for position-based mobile information filtering. In both cases, the applications will be evaluated with end users.
Dowload the ENCORE project description here