User Device Interaction Prediction Via Relational Gated Graph Attention Network and Intent-aware Encoder

Jingyu Xiao,Qingsong Zou,Qing Li,Dan Zhao,Kang Li,Wenxin Tang,Runjie Zhou,Yong Jiang

DOI: https://doi.org/10.5555/3545946.3598820

2023-01-01

Abstract:With the booming of smart home market, intelligent Internet of Things (IoT) devices have been increasingly more involved in home life. To improve the user experience of smart home, some prior works have explored how to use time series analysis technology for predicting the interaction between users and devices. However, existing solutions have inferior User Device Interaction (UDI) prediction accuracy, as they fail to consider the complex heterogeneous device transitions, multiple intents of a user and multi-level periodicity of user behaviors. In this paper, we present DeepUDI, a novel approach for accurate UDI prediction. First, we propose Relational Gated Graph Attention Network (RGGAT) to learn embedding of device and device control while considering complex heterogeneous temporal transitions. Second, we propose Intent-aware Encoder (IAE) to encode multiple intents of users via capsule networks. Third, we design a Historical Attention Mechanism (HAM) to capture the multi-level periodicity by aggregating the current sequence and the historical sequence representations through the attention mechanism. Comprehensive experiments on four realworld datasets show that DeepUDI consistently outperforms state-of-the-art baselines and also offers highly interpretable results.