Authors: Prakhar Srivastava; Pankaj Srivastava
Addresses: Department of Information Technology, ABV – Indian Institute of Information Technology and Management, Gwalior, Madhya Pradesh, India ' Department of Applied Physics, ABV – Indian Institute of Information Technology and Management, Gwalior, Madhya Pradesh, India
Abstract: Movie plot summaries are highly indicative of the genre to which they belong. Depending upon the words present in the plot summaries, we can easily decide which emotion is being portrayed in the movie. In this paper, we predict the movie genres by feeding the plot summaries to our proposed model. For making word representations that can be understood by our model, we use Facebook's fasttext library. Our model uses a Bi-LSTM network and a ranking system depending upon posterior probability scores to determine the movie genre. We split the plot summary into sentences and predict the genre associated with each sentence using our model. We then fuse the decision from all the sentences to make a collective decision for a particular plot summary. We use the majority voting algorithm for making this decision. We try document-level and sentence-level approaches for predicting the movie genres. Post comparison of results, we found sentence-level approach using Bi-LSTM network performs better than the document-level approach using the same network. For the baseline models, we used recurrent neural networks (RNN) and logistic regression (LR) and compared the results with our proposed model.
Keywords: NLP; movie genre prediction; Bi-LSTM; word embeddings; recurrent neural networks; fasttext; majority voting algorithm.
International Journal of Swarm Intelligence, 2021 Vol.6 No.2, pp.168 - 177
Received: 29 Jun 2020
Accepted: 27 Nov 2020
Published online: 29 Oct 2021 *