Abstract
This paper presents an approach to build a TV recommendation system called PersonalTV that enables the use of multiple classifiers, each one specialized on selected attributes of detailed program information. For generating adequate recommendations, the system makes use of content filtering and the preferences directly specified by the user within an MPEG-7 profile. By tracking user actions and interpreting their semantics, the system is able to individually weight these actions and dynamically adjusts the process to the user’s evolving preferences. We show how specialized spam fighting methods can successfully be transferred to the area of recommendation systems and adapted accordingly. Being lightweight, these methods are especially applicable in resource-constrained environments such as TV set-top boxes or mobile devices. Moreover, the use of the variance of the beta-distribution as a confidence value for each recommendation is presented.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have