Information fusion in a TV program recommendation system

In a previous paper, we presented a conceptual framework for high-level information fusion. The approach relies on the use of the Conceptual Graphs model for both knowledge representation and information fusion. We introduced the use of fusion heuristics in the framework. This paper validates the viability of the approach through several experimentations. We use the fusion platform implementing the framework within a recommendation system for smart television. TV program descriptions coming from different sources of information are fused. To evaluate the quality of the fusion process, we compare the results to TV program descriptions recorded by the INA institute. We consider both the preciseness of the schedule and the proportion of programs correctly predicted. The experimentations emphasize the importance of the heuristics that we introduced in the conceptual framework.