Article Preview
Top1. Introduction
With the mature of Internet technologies and the rapid growth of the social network, microblogs have emerged as a new open channel of communication or publication for people on the Internet to read, make comments, cite, socialize and even make new connections beyond their social networks (Huang et al., 2009). Microblogs represent the opinions of the populace and can be regarded as a reaction to current events on the Internet, such as breaking news and gossips. According to report from Technorati.com, over 150 million microblogs have been recorded, over 400 million pieces of tagged social media exist, and over 1.9 million microblog entries are updated every day. Microblogs have become a connected community and social network, and microblogs are increasingly being considered to be an important marketing domain induced by online person to person interactions (Liu et al., 2011). Besides, there is a huge number of information in microblog spaces, including articles, profile, pictures and other multimedia resources, which has become the “information overload” problem for microblog users. This problem brings bloggers plethora of choices and options available that often varies in quality.
The need for solving the problem of “information overload” has led to the prevalence of personalized recommendation. Recommender systems are the most successful application of personalized recommendation. They receive information from users about items that they are interested in, and then recommend to them items that may fit their needs. The core of recommender systems depends on two well-known filtering algorithms: content-based filtering (CBF) and collaborative filtering (CF) (Sarwar et al., 2001). In contrast, CF only depends on historical information about whether or not a given target user who has previously preferred an item, and does not necessarily required any analysis on the actual content of an item. Therefore, CF has an advantage over CBF in situations where it is hard to analyze the underlying content. However, the microblog recommender system differs from other kinds of recommender systems. Such as, recommendation targets vary from products, multimedia, news and articles to all kinds of service and even virtual community (Kumar & Mishra, 2009), and the inappropriate recommendation may result in unpleasant attitudes towards the recommendation target (Lee et al., 2007); and microblog content is very subjective and textual-sensitive for recommenders. In other words, in the scenario of microblog recommendation, it is important for recommender systems to introduce interesting articles to microblog readers. Despite its advances, CF-based recommender systems are unable to describe and distinguish the relationships between different users (Adomavicius et al., 2005). This problem severely affected the recommendation quality. Therefore, how to utilize social network information in making recommendations should yield valuable information.
In this paper, an improved approach called SNbR is developed to enhance the effectiveness of microblog recommendation by combining social network information and CF methods. Firstly, the data about users’ preferences/interests and their relationships are collected. Then, microblog users are ranked by their browsing behaviors in the scenario of social network, and user ranking is incorporated into the similarity computation for the advance of the recommendation quality. Finally, we generated recommendations about items using CF and the suggested neighbour groups, and compared the performances of diverse algorithms