A Case Study on the Effectiveness of Recommendation Al-gorithms for Sake Review Website

Abstract

Recommendation systems have been used in various online businesses and have been recognized as useful for promoting product purchase and information browsing. However, the system design policies for each target service remain insufficiently clarified. One reason for this is that few case studies in practical environments exist since only the provider of the commercial platform obtains and confirms operational data of the recommendation algorithm and the resulting user behavior. Therefore, in this study, we investigate how some recommendation algorithms affect user behavior in online services for the Japanese Sake system. In the investigation, one of the four recommendation algorithms was assigned to each of the 858 users who logged into the system during 80 days. For each recommendation algorithm, the number of system logins, transitions to the sake description pages, sake brands registered in the sake to-drink list, and posted review comments to each sake brand were analyzed from the system history. Our results show that the recommendation system with association analysis was effective in recommending sake brands for its review sites, considering individual preferences and general popularity.