Graph Neural Network-enhanced Sequential Learning for Time Series and Social Media Prediction

Abstract

In recent years due to the emergence of the Graph Convolution Network (GCN) it had been widely used in graph data and can not only use its own information but also consider the nodes connected to itself in the graph to provide more information for model learning However GCN is rarely used in the sequential data We hope that we can utilize the advantage of GCN to imply on two types of sequential data time series data of sensor and retweet sequence in social media to improve the forecasting performance For time series data of sensor we propose a novel GNN-based model Attention-adjusted Graph Spatio-Temporal Network (AGSTN) In AGSTN multi-graph convolution with sequential learning is developed to learn time-evolving spatio-temporal correlation between sensors Fluctuation modulation is realized by a proposed attention adjustment mechanism Experiments on three sensor data air quality bike demand and traffic flow exhibit that AGSTN outperforms the state-of-the-art methods For retweet sequence in social media we focus on dealing with the fake news detection problem We develop a novel GNN-based model Graph-aware Co-Attention Network (GCAN) to achieve the goal Given the source short-text tweet and its retweet users sequence without text comment we aim at predicting whether it is fake or not and generating explanation by highlighting the evidences on suspicious retweeters and the words they concern Experiments on real tweet datasets exhibit that GCAN can significantly outperform the state-of-the-art methods by 16% in accuracy on average and produce reasonable explanation

Date of Award	2020
Original language	English
Supervisor	Cheng-Te Li (Supervisor)