Recommender System With Visual Features

People

Song Park, Hyunjung Shim and Jongwuk Lee

Abstract

• The rating database which is used for recommender system is so sparse and lack of information

• A movie poster can reveal some useful things about that movie: actors, genres, atmosphere of the movie, etc.

• Employing visual features of movie posters would be able to enhance the performance of recommender system

Overview

Progress

• Constructed movie posters dataset corresponding to movielens-1m and movielens-100k

• Investigated the appropriate deep architectures for visual features and collaborative filtering

• Implementing integrating network with Tensorflow