Collaborative Filtering without Explicit Feedbacks for Digital Recorders

TL;DR

This paper explores a novel collaborative filtering approach for digital recorders that operates without explicit user feedback, using unstructured data and data preprocessing to generate recommendations in IPTV environments.

Contribution

It introduces a new method for collaborative filtering that does not rely on explicit feedback, starting from unstructured recordings and applying data preprocessing.

Findings

Effective recommendations from unstructured data

First approach without explicit feedback in IPTV

Potential for improved user experience

Abstract

Recommendation is usually reduced to a prediction problem over the function $r(u_a, e_i)$ that returns the expected rating of element $e_i$ for user $u_a$. In the IPTV domain, we deal with an environment where the definitions of all the parameters involved in this function (i.e., user profiles, feedback ratings and elements) are controversial. To our knowledge, this paper represents the first attempt to run collaborative filtering algorithms without inner assumptions: we start our analysis from an unstructured set of recordings, before performing a data pre-processing phase in order to extract useful information. Hence, we experiment with a real Digital Video Recorder system where EPG have not been provided to the user for selecting event timings and where explicit feedbacks were not collected.