Serendipity in Recommender System: A Holistic Overview

摘要

Recommender system is a software built to retrieve relevant information based on user interest [1]. Recommender system can determine user's interest by looking at several resources such as user's consumed items, similar users, and search logs. The massive amount of information about users and items coupled with extensive research in increasing recommender system's accuracy resulted in an "over-specialization" problem [2]. In which, recommender systems tend to recommend obvious items or previously known items. These obvious recommendations result in failing to arouse users' long-term interest. For example, a high accuracy travel recommender system will never recommend new places to a user outside of the already visited places. To mitigate this problem, researchers shifted from focusing on accuracy to achieve user satisfaction to a more user central measures known as beyond accuracy measures. These measures believed to increase user satisfaction, and long-term interest.