Synapse: Meta Recommendation System Using Contextual and Non-Stationary Bandits

DOI

Bibliographic Information

Other Title
  • Synapse: 文脈と時間経過に応じて推薦手法の選択を最適化するメタ推薦システム

Description

Since the effectiveness of a method depends on various contexts and varies with time, the timely selection from among many recommendation methods appropriate to contexts has become a critical issue for EC vendors. In this paper, we propose a practical recommendation system that continuously selects recommendation methods suitable for contexts and corresponding to temporal change using a multi-armed bandit policy. Our proposed system quickly detects changes in the effectiveness of recommendation methods using aggressive exploration. The experimental results using a real-world dataset collected from an EC site demonstrate that the proposed system outperforms baseline methods.

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top