12th EAI International Conference on Testbeds and Research Infrastructures for the Development of Networks & Communities

Research Article

DineTogether: A Social-Aware Group Sequential Recommender System

Download654 downloads
  • @INPROCEEDINGS{10.4108/eai.8-1-2018.155564,
        author={Zheyu Chen and Anqi Hu and Jie Xu and Chi Harold Liu},
        title={DineTogether: A Social-Aware Group Sequential Recommender System},
        proceedings={12th EAI International Conference on Testbeds and Research Infrastructures for the Development of Networks \& Communities},
        publisher={EAI},
        proceedings_a={TRIDENTCOM},
        year={2018},
        month={1},
        keywords={Group recommender system random walk with restart social aware model},
        doi={10.4108/eai.8-1-2018.155564}
    }
    
  • Zheyu Chen
    Anqi Hu
    Jie Xu
    Chi Harold Liu
    Year: 2018
    DineTogether: A Social-Aware Group Sequential Recommender System
    TRIDENTCOM
    EAI
    DOI: 10.4108/eai.8-1-2018.155564
Zheyu Chen1, Anqi Hu1, Jie Xu1, Chi Harold Liu1,*
  • 1: Beijing Institute of Technology, China
*Contact email: liuchi02@gmail.com

Abstract

Group recommendations become important in many practical scenarios, e.g., couples would like to share movies together, families dine together in a restaurant, a group of friends plan to spend vacation in some points of interest. Although some previous research efforts have been done in this area, most of them only consider a small portion of contextual factors but ignore the time series features of the user historical behavior and next recommended location/event to do, that makes the system performance not as satisfactory as expected. Here, we propose a novel group sequential recommender system, called “DineTogether”, for a group of people dinning together. It is able to capture comprehensive contextual, social factors when make recommendations.We design a computational model, called “Social-and-Time-Aware (STA)” model, and a novel algorithm, Generalized Random Walk with Restart (GRWR). Experiment results show that our approach outperforms the state-of-the-art group recommendation approaches.