The Strength of Dithering in Recommender System

• Main Authors: Xiangyong Liu, Pengfei Yin, Bhuiyan, Md Zakirul Alam, Guojun Wang
• Format: Conference Proceeding
• Language: English
• Subjects: Algorithm design and analysis; Collaboration; Collaborative filtering; Content-based recommendation; Data models; Dithering; Engines; Learn to rank; Libraries; Prediction algorithms; Recommender systems
• ISSN: 2375-527X
• DOI: 10.1109/ISPAN-FCST-ISCC.2017.61

An important goal of a recommender system is to solve the top-k recommendation problem, however, there is no perfect ranking list for any recommender algorithm. Much work has been done on the recommendation list to improve user experience. In this paper, we focus on the technique of dithering which can be used in an online recommendation situation and be neglected in most academic research, and propose a new dithering method for the ranking formulation of recommendation problem. At first step, the system (e.g. recommender engine) collects user's preference and provides a corresponding ranking. The user responds with domain predefined operation (e.g. clicking), and at the following each step, the ranking will be slightly changed, for example, shaking things up by intentionally including in a list of less relevant items. The difference of the list depends on the dithering method and its parameters. We demonstrate the feasibility and utility of proposed methods through a case study.