Article Preview
Top1. Introduction
In today’s online environment, online retailing and content providers are empowered to offer a huge selection of products and services to meet a variety of consumers' needs and tastes. As a result, consumers are inundated with a wealth of information and choices. For example, if a consumer enters the words “digital camera” into the search box at Amazon's web site, the search result shows over 1 million related items. To mitigate such information overload, recommender systems plays an important role in recommending products that are most likely to be of interest to users (Kuanr & Mohapatra, 2021; Sinha & Dhanalakshmi, 2019). More than one third of online customers that notice these recommendations wind up purchasing the recommended product (Vasa, 2021). Netflix reported that about 75% of the content watched by its subscribers were suggested by its recommendation system (Chong, 2020).
Traditional recommender systems have typically used the metrics such as accuracy and coverage to measure the performance of the recommendations (Adomavicius & Tuzhilin, 2005). For example, the 1 million Netflix Prize (netflixprize.com) sought to substantially improve the accuracy of predictions about how much someone is going to enjoy a movie based on their movie preferences. However, recent studies have argued that the quality of recommendations should go beyond accuracy. As one of the goals of recommender systems is to provide a user with highly idiosyncratic or personalized items, more diverse recommendation list would result in more opportunities for users to be recommended such items (e.g., Adomavicius & Kwon, 2009; Wu et al., 2019; Zhang & Hurley, 2008; Zhang, et al., 2012). Park and Han (2013) find that recommending a diverse list of products enhances user satisfaction and is positively associated with customer retention rates. These, in turn, would lead to the boost of sales revenue and the bottom line. However, one of the challenges is that the methods seeking to increase diversity in recommendation list often sacrifice accuracy (Adomavicius & Kwon, 2009; Hurley & Zhang, 2011; Kuanr & Mohapatra, 2021).
In a nutshell, a basic recommendation method seeks to predict the 'preference' or 'rating' that a user would give to an item, such as music, books, movies, etc. that has not been seen by the user. A rating indicates how a particular user liked a particular item, e.g., Jane Betty gave the book ``Thinking, Fast and Slow'' the rating of 4 (out of 5). These ``particular'' recommendations are often referred as personalized recommendations (Kim et al. 2003). The task of personalized recommendation requires the ability to predict which items will be considered interesting by the user. The personalized recommendations have long been viewed as an important source to assist and augment the natural social recommendation process - in our everyday life, we rely on recommendations from our social circles by word of mouth.
Although recent studies suggest that online social relations influence users' both product choices and ratings (e.g., He et al., 2017; Wang et al., 2018), few studies have examined how they may be utilized to improve the diversity of personalized recommendation list. To address this gap, we explore how online social relations can be used to enhance the diversity of personalized recommendation list while without sacrificing the accuracy in this study. In particular, we utilize a data set collected from an online Web 2.0 site that integrates both online social networking and online product rating functions. In our proposed approach, the ratings from a user's social circle are used as a source of implicit feedback, which is integrated into the recommendation model of collaborative filtering. In order to achieve recommendation diversity while without sacrificing accuracy, we allocate more weight onto the products rated by the focal user's social circle than the others while limiting the degree of dissimilarity in user preferences.