Predicting Tags for Movies from Plot Synopses Using Deep Learning Techniques

Abstract

Automatically generating or predicting tags for movies can help recommendation engines improve retrieval of similar movies, and help viewers know what to expect from a movie in advance. It improves the search results of a movie recommender system by predicting high weighted tags from a movie's plot synopsis. We propose a model in which we at rst perform pre-processing of data(stopwords eradication, stemming of data etc.) and then tokenize the data by a technique called BERT and then vectorize it by TF-IDF process and then input those pre-processed data to a deep learning technique to give us a prediction tag scores from a set of tags for movies. We compare our system's result with an already proposed model with emotion ow encoded neural network and found that our model's perfor- mance shows improvement in result(TL, TR and F1 measure) specially due to pre-processing of data and for using the techniques like BERT and TF-IDF.