Recommending Hidden Masterpieces: Design and Evaluation of a Web Novel Recommender System

Abstract

Websites accommodating user-generated novels have been becoming popular. An effective and efficient system is however required to adequately link published works and potential consumers. We have been working on a personalized recommender system that makes recommendations by balancing the user's preference uncovered from her/his reading history and the estimated qualities of works. These two aspects, in tandem, contribute to discovering a hidden masterpiece, which often lacks user-provided information due to its yet-to-known status. To adequately capture these aspects, the system calculates the similarity score between works and estimates the quality of a work by referring to sampled sentences and additional information such as genres and keywords. In this paper, the current implementation of the system is described, along with the small-scaled user evaluation results, which generally appreciated the system design principle and enjoyed the system's recommendations. They also provided useful comments to improve system usability.