Research Area:  Machine Learning
The sequences of users behaviors generally indicate their preferences, and they can be used to improve next-item prediction in sequential recommendation. Unfortunately, users behaviors may change over time, making it difficult to capture users dynamic preferences directly from recent sequences of behaviors. Traditional methods such as Markov Chains (MC), Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM) networks only consider the relative order of items in a sequence and ignore important time information such as the time interval and duration in the sequence. In this paper, we propose a novel sequential recommendation model, named Interval- and Duration-aware LSTM with Embedding layer and Coupled input and forget gate (IDLSTM-EC), which leverages time interval and duration information to accurately capture users long-term and short-term preferences. In particular, the model incorporates global context information about sequences in the input layer to make better use of long-term memory. Furthermore, the model introduces the coupled input and forget gate and embedding layer to further improve efficiency and effectiveness. Experiments on real-world datasets show that the proposed approaches outperform the state-of-the-art baselines and can handle the problem of data sparsity effectively.
Author(s) Name:  Dongjing Wang, Dengwei Xu, Dongjin Yu & Guandong Xu
Journal name:  Applied Intelligence
Publisher name:  Springer
Volume Information:  volume 51, pages 906–920 (2021)
Paper Link:   https://link.springer.com/article/10.1007/s10489-020-01820-2