Movie Genre Prediction and Recommendation Using Deep Visual Features from Movie Trailers
V M Radhika, K. P. Swaraj · 2020 International Conference on Power, Instrumentation, Control and Computing (PICC) · 2020
Most of the existing video recommendation models use only the textual contents for recommending interesting videos to users. The textual contents are often missing for videos created by users. Humans are attracted to visual data than any other type of data. Plenty of visual feature information offers people useful inferences. The video classification and recommendation using deep learning is a challenging task. This work proposes a method for movie genre prediction and recommendation using visual features from trailers. This approach segments, each trailer into visual shots of 1 second and extracts the deep visual features of the video frames using the Resnet-152 model. To capture the temporal sequences, it uses the LSTM (Long Short Term Memory) architecture. The output obtained from the Resnet-LSTM model is a vector of the probability distribution of the video segment over the genres (action, comedy, horror, romance). For the recommendation task, this method uses the MovieLens 25M dataset along with the extracted visual features. Finally, the recommendation system will recommend movies based on the genre and rating. This method outperforms current state-of art-approaches.