Efficient Recommendation Systems for Movies Based on BERT4Rec

Abstract

Translate Article

The existing recommendation systems ususally uses collaborative filtering or content filtering for providing recommendation for users. The system is trained and tested with The MovieLens 25M Dataset, which contains a few million data pairs. The success of the recommendation system through the training of BERT4Rec would help the user to filter out unsuitable information. The BERT4Rec is a bidirectional model based on transformer, which allows the training of the recommendation system to be efficient. The model is consisted of 3 layers, the embedding layer, which process the data into information that is later used in the transformer layer, the transformer layer will process the data and forward it to the output layer, where the results would be received by the user. Through testing of a few samples, the recommendation system suggested accurately, which reflects the success of the training and testing phase of this recommendation system. For future improvements, the system could be trained with a data set that includes the ticketing information, which would more accurately reflect what the user had watched in the past. Thus, creating a more precise recommendation system that would give better suggestions for the user.