Movie genre prediction from subtitle using deep learning

Nowadays, in the era of big data, prediction plays an important role in finding correct estimations about related data. People need to find correct data relationships from other data resources by using other related data. The movie genre is one of the areas to predict and categorize movies using subtitles. Other movie prediction approaches use different data models to solve the problem. This thesis aims to predict movie genre from subtitle files by using deep learning methods. The dataset used in the study was obtained from the IMDb and OpenSubtitle websites. This dataset contains subtitle files in XML format for different movie/series and corresponding movie categories. The subtitle files were preprocessed with a conversion technique and each subtitle was converted to 100.000 dimensional vector. The LSTM deep learning model developed in this thesis based on the preprocessed dataset was tested with 5 fold cross-validation technique and the results were measured and presented with different methods such as Area Under the ROC Curve (AUC) and Hamming loss. The suggested model resulted in an accuracy rate of 93.97%, 0.2392 Exact Match Score and 0.0602 Hamming Loss. The model employed in this work shows that a high prediction ratio for the movie genre is obtained by using deep learning techniques.