Please use this identifier to cite or link to this item:
Title: Tailoring Recommendations to Groups of Viewers on Smart TV: A Real-Time Profile Generation Approach
Authors: Alam, Iftikhar
Keywords: Computer Science
Computer & IT
Issue Date: 2020
Publisher: University of Peshawar, Peshawar.
Abstract: Smart TV has changed the legacy TV system by providing processing, storage, and connectivity capabilities. It provides better support for watching terrestrial TV channels as well as online web-based live channels, stored videos, and Web 2.0 features. This unique blend makes smart TV an attractive device for many households and thus the market share is increasing day-by-day. However, the unstoppable growth of multimedia content over the Web creates difficulties in finding the desired items and hence contributes to a cognitive overload problem. Browsing and searching for the desired content on a smart TV is not only timeconsuming but also difficult due to its lean-back nature, pull technology, and limited functionalities of its remote-control. Although channels can be searched through Electronic Program Guides (EPGs) but still it is a difficult job due to scrolling a huge list of channels on the remote control. Besides these, recommender systems are also helpful to mitigate the issues of cognitive overload. A recommender system uses different parameters based on user demographics, watching history, and preferences to recommend the most relevant items. These parameters are easy to predict and calculate for a single user on a personalized device including personal computers or smartphones but challenging on smart TV because of its multi-user support and lean-back nature. Therefore, a group recommender system is more appropriate to meet the information needs of group of users. Several approaches have been proposed, such as aggregated predictions, preferences aggregation, etc.; however, these approaches may lead to not only privacy issues but also challenging for predicting the actual member of a group. Moreover, the in-depth analysis of user data on the server-side may further lead to serious 9 Tailoring Recommendations to Groups of Viewers on Smart TV: A Real-Time Profile Generation Approach privacy issues. In an ideal situation, a recommender system should recommend the right items to the right viewer(s). However, in a smart TV watching scenario, achieving this is neither simple nor accurate because of distinct watching behaviour and security concerns. Fortunately, the smart TV comes with not only processing and storage but also with support for camera and microphone that can be used for the secure identity of a user based on “age”, “gender” and “number of viewer(s)” locally and without any connection to the server. This thesis uses such approaches to identify viewer(s) and then generate novel group modeling techniques to identify different profiles in front of a smart TV. It proposes a novel grouping formula and age-gender matrix for generating group profiles based on age, gender, and number information. This information is collected using the Haar-Featured Cascade Classifier (HFCC) and Convolutional Neural Network (CNN) algorithm running on a smart TV to generate groups. A separate profile is generated for each group instead of merging the individual profiles or preferences. The proposed system is trained by using the MovieLens dataset and evaluated using GraphLab-Framework. The results are analyzed statistically and experimentally by using different tests and algorithms. The results showed a significant impact on recommendations to individuals and group users in front of smart TV. This thesis also proposes a statistical method for finding a dominant character in a group based on social metadata i.e. ratings. The finding suggests that existing recommender systems should adapt itself to the varying watching behaviour typically in a smart TV. Improving the recommender system for smart TV may not only contribute to user(s) satisfaction but it may also improve the conversion rate.
Gov't Doc #: 20924
Appears in Collections:PhD Thesis of All Public / Private Sector Universities / DAIs.

Files in This Item:
File Description SizeFormat 
Iftikhar Alam 2020 CS uop peshwar.pdfphd.Thesis3.59 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.