Deep Heterogeneous Network for Temporal Set Prediction

Abstract

Given historical sequential sets of elements such as purchasing items from time to time, receiving clinical records in subsequent visits to the hospitals, could be formalized as sequential sets, namely temporal sets. Forecasting the elements in the subsequent set according to historical sequence of sets is denoted as the problem of temporal sets prediction, which can provide useful and effective information for decision-making and planning in different domains. However, due to the difficulty of set representation, dynamic temporal dependence of historical sets and fusion of multi-level user preference , it is challenging to model and predict such temporal sets. To address these issues above, we propose a novel Deep Heterogeneous Network for Temporal Sets Prediction (DHNTSP) in this paper. Firstly, we design a set representation method based on Heterogeneous Information Network (HIN) embedding, where HIN is used to model the multiple relationships among sets, items, users and categories, and matrix factorization is used to vectorize the nodes of HIN. Then, an attention-based recurrent module is designed to learn the temporal dependence of the historical sets. Last, a fusion module is designed to combine sequential-level and individual-level of user preference to further improve the performance. The experiments are conducted on real-world datasets, and results demonstrate that DHNTSP outperforms both classical and the state-of-the-art methods.