Classification and preprocessing of data for television recommendation system

This paper presents a general approach for personalized recommendation system for next generation of smart television. Hence the TV provides a hybrid broadband and broadcast transmission both can collect information. Aim is to combine television with internet content in order to provide interactive and personalized recommendation. This improves user's watching experience. Since these two sources have a different format, data integration is needed. Additionally, the data have to be preprocessed in order to remove so-called “global effects” and improve further classification. Classifier is based on k-nearest neighbors (kNN) improved approach, designed for the Neflix price. It was described for an on-demand-video, where no time consideration is needed. On the other hand, we include time as a factor worth considering as it is typical for TV program schedule. Final recommendation includes also Fuzzy logic based training sequence selection and final weight correction.