Ένα μοντέλο σύστασης για υπηρεσίες IPTV με χρόνο - επίγνωση: Μία κατά τμήματα γραμμική επέκταση του πρωτότυπου αλγορίθμου SVD++.
A time-aware recommendation model for IPTV: A partially – linear extension to the original SVD++ algorithm
This work addresses the problem of recommendation within the context of developing user individualization services for IPTV, involving vast volumes of timely zapping data that span a wide spectrum of users and associated multimedia items. Specifically, the primary objective of this work lies upon the extension of the linear models for the users’ baseline predictors which are embedded in the original formulation of the SVD++ recommendation algorithm. The fundamental characteristic of such models concerns the uniform manipulation of the timeline of interactions for each user, thus failing to accurately model the variation of user ratings over time. This work, on the contrary, introduces a more flexible user baseline predictor by utilizing a partially linear model which is built by dividing the timeline of interactions in such a way that a predefined fraction of ratings fall within each time interval. Therefore, the proposed model aims at capturing in a more efficient way significant modifications of user’s viewing behavior that otherwise would be ignored. A series of experiments conducted verify the efficiency of the proposed model as measured in terms of Means Absolute Error.