A social recommender mechanism for e-commerce: Combining similarity, trust, and relationship

Highlights

• Success of e-commerce depends greatly on an effective product recommender design.

• The social recommender system can accurately generate personalized products.

• Recommendation sources include the factors of preference, trust, and relationships.

• The proposed framework can effectively promote retailer's products and services.

Abstract

Online business transactions and the success of e-commerce depend greatly on the effective design of a product recommender mechanism. This study proposes a social recommender system that can generate personalized product recommendations based on preference similarity, recommendation trust, and social relations. Compared with traditional collaborative filtering approaches, the advantage of the proposed mechanism is its comprehensive consideration of recommendation sources. Accordingly, our experimental results show that the proposed model outperforms other benchmark methodologies in terms of recommendation accuracy. The proposed framework can also be effectively applied to e-commerce retailers to promote their products and services.

Introduction

“Social is not just about sharing connections, it's about providing different ways for people to interact…. Social commerce excites me — we already know how powerful recommendations from friends can be and the group shopping experience can easily be replicated through social commerce.”

Sophy Silver, Facebook's UK & Ireland Public Relations Chief

With booming social networking technologies and platforms, most e-commerce companies are creating social network profiles of their own. J.P. Morgan anticipates that global e-commerce revenue will reach $963 billion by 2013 [28]. The report forecasts that e-commerce revenue will grow to $680 billion worldwide, up 18.9% from 2010 revenue, and online retail commerce in the U.S. alone will grow 13.2% to $187 billion.

For many people, shopping is a social experience, and they often want to get their friends' opinions before buying. Social commerce is helping people buy where they connect. It integrates social media into e-retail sites and adds e-commerce functionality to social networks. For online storeowners, social commerce is becoming a way of thinking about transacting business online. Some e-commerce sites use your friends' preferences to help you make better purchasing decisions. Amazon, for example, helps you find records and books by the artists and authors your friends have listed in their Facebook profiles.

Recommender systems assist users in making choices from various alternatives; the goal of these systems is to estimate user preferences and provide predictions of appropriate information. Social recommender systems aim to relieve information and interaction overload by applying various techniques that ultimately present the most relevant and attractive information to users. These personalized recommendations based on social interactions or preferences are viewed as a huge opportunity for vendors. Indeed, a survey of online retailers in 2010 found that over half planned on implementing recommendation features on their sites [20].

To date, a variety of recommendation techniques has been developed. To our best knowledge, collaborative filtering, content-based, and hybrid approaches are three popular approaches that have been used to generate recommendations [23]. An approach that has received less attention is using the social relations on individuals as an additional source of information. The principle of homophily from the social network field suggests that “similarity breeds connection.” In other words, users share many attributes with the people close to them. This suggests that if we have information about the connections in a person's network, we can infer some of that person's attributes. Most commercial recommender systems are strongly supported by the demographic information of users. Since some of the similarities within a network are caused by the influence and interactions of its members, it would be reasonable and feasible to develop a social recommendation based on the connections of individual users. In reality, people tend to be affected by the opinions of and suggestions by people with similar interests, shopping experts, and close friends.

However, most of current social networking platform, such as Facebook and twitter, and electronic commerce platform, such as Amazon and Yahoo! Shopping, are independently operated. The supporting recommender systems are also independently deployed on the two kinds of platforms based on social factors and purchase history respectively. As a result, the electronic commerce (retailing) platforms generally do not consider social factors such as relationships and trust etc. among the users and the power of social influence is not exploited. Contrarily, social networking platforms generally do not consider online shopping related factors such as purchase history and product rating etc.

To address these issues, in this research, we synthesize the features of social networking and electronic commerce platforms to design a social recommender mechanism that considers the factors of preference similarity, recommendation trust, and social relationship in order to increase the prediction accuracy of product recommendations in e-commerce. The factors of human interactions and relations (e.g. trusts [30], [61], reputations [40], [64], and social relationship [34], [68]) have been applied separately in different application contexts. In this research, by building several new social metric formulas, we exploit and consolidate various types of consulting information source to generate product recommendations.

The proposed mechanism allows us to identify suitable products for individual customers by utilizing the collective intelligence from social networks and to balance the consulted sources based on these personalized preferences. Our experimental results based on users' evaluations in Yahoo! Shopping show that the proposed model could enhance recommendation accuracy. The proposed model could be practically applied to new emerging social commerce platforms.

The remainder of this paper is organized as follows. In Section 2, we discuss the existing literature related to our research topics. In Section 3, we discuss the factors that contribute to the proposed social recommendation framework. Section 4 describes the experimental data source, settings, and procedures. The experimental results and evaluations are discussed in Section 5. Section 6 concludes our research contributions and presents future research directions.