Tag Recommendation for Movie and Music Contents

Tag recommendations are popular for labelling online web content. It usually involves the process of content classification, organization, and information retrieval. With the increase in the consumption of online content, recommender systems have been developed to study the behavioural patterns of users and use these results to recommend content. The cold start problem is an issue being faced in developing recommender systems due to insufficient data regarding recommended items. In this paper, we propose a hybrid approach of a Media Convolutional Neural Network with a Latent Dirichlet Allocation topic modelling (MCNN-LDA) algorithm for recommending tags for movie and music contents in trying to address the multi-label classification issue of the cold start problem. The final model is evaluated on the Internet Movie Database (IMDB) movie dataset and GTZAN audio dataset. The high accuracy results of the developed model achieved at a threshold of 0.90 indicate a high performance on the quality of tags recommended which is confirmed by the evaluation metrics performed.