Paper Info
Title
GMTL: A GART Based Multi-task Learning Model for Multi-Social-Temporal Prediction in Online Games
Abstract
Multi-social-temporal (MST) data, which represent multi-attributed time series corresponding to the entities in multi-relational social network series, are ubiquitous in real-world and virtual-world dynamic systems, such as online games. Predictions over MST data such as social time series prediction and temporal link weight prediction are of great importance but challenging. They are affected by many complex factors, including temporal characteristics, social characteristics, collaborative characteristics, task characteristics and the intrinsic causality between them. In this paper, we propose a graph attention recurrent network (GART) based multi-task learning model (GMTL) to fuse information across multiple social-temporal prediction tasks. Experiments on an MMORPG dataset demonstrate that GMTL outperforms the state-of-the-art baselines and can significantly improve performances of specific social-temporal prediction task with additional information from others. Our work has been deployed to several MMORPGs in practice and can also expand to many related multi-social-temporal prediction tasks in real-world applications. Case studies on applications for multi-social-temporal prediction show that GMTL produces great value in the actual business in NetEase Games.
Year
DOI
Venue
2019
10.1145/3357384.3357830
Proceedings of the 28th ACM International Conference on Information and Knowledge Management
Keywords
Field
DocType
graph attention network, link weight prediction, multi-task learning, online game, recurrent neural network, time series prediction
Multi-task learning,Information retrieval,Computer science,Artificial intelligence,Machine learning
Conference
ISBN
Citations 
PageRank 
978-1-4503-6976-3
0
0.34
References 
Authors
0
6
Name
Order
Citations
PageRank
Jianrong Tao15111.96
Linxia Gong271.17
Changjie Fan35721.37
Longbiao Chen412310.60
Dezhi Ye550.83
Sha Zhao6489.96