Session based recommendations using recurrent neural networks - Long short-term memory

• Main Authors: Dobrovolny, Michal, Selamat, Ali, Krejcar, Ondrej
• Format: Conference or Workshop Item
• Published: 2021
• Subjects: T Technology (General)
• Online Access: http://eprints.utm.my/id/eprint/98062/; http://dx.doi.org/10.1007/978-3-030-73280-6_5
• Institution: Universiti Teknologi Malaysia

This paper describes the use of long short-term memory (LSTM) for session-based recommendations. This paper aims to test and propose the best solution using word-level LSTM as a real-time recommendation service. Our method is for general use. Our model is composed of embedding, two LSTM layers and dense layer. We employ the mean of squared errors to assess the prediction results. Also, we tested our prediction of recall and precision metrics. The best performing network has been a trainer for the last year of likes on an image-based social platform and contained about 2000 classes. Our best model has resulted in recall value 0.0213 and precision value 0.0052 on twenty items.