Harmonic attentive multimodal neural network for movie recommendation. (May 2020)

• Record Type: Journal Article
• Authors: Wang, J X; Guo, Y; Qi, T M; Tang, Q F
• Is Part Of: Journal of physics. Volume 1550:Number 2(2020)
• Publication Date: 2020-05
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1550/2/022018
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: A multimodal neural network based on harmonic self-attention was proposed for movie recommendation. This method can deal with the multi-source data and learn representations of users and items well. The multimodal neural network mainly consists of three sub-networks, ResNet, Bert, and LSTM. In addition, the harmonic self-attention mechanism can explore the user's preference from the perspective of time. Experimental results show that compared with other three latest recommendation methods, HSMNN shows better performance in terms of HR and NDCG.