TOSI: یک روش ارزیابی اثر شبکه اجتماعی اعتماد محور در شبکه های اجتماعی زمینه ای TOSI: A Trust-Oriented Social Influence Evaluation Method in Contextual Social Networks

۱٫ Introduction ۱٫۱٫ Background Online Social Networks (OSNs) are becoming more and more popular and have been used as the means in a variety of applications, like employment, CRM and e-Commerce. In these applications, the social influence of a participant can affect others’ decision-making. For example, at Epinions (epinions.com), an OSN based e-commerce platform, a buyer can write a product review to rate the products and corresponding seller. This review can be viewed by other buyers and thus can impact their decision making in purchasing the same products. As indicated in studies of Social Psychology [1, 2, 3] and Computer Science [4], a person is more likely to accept the recommendations given by participants with higher social influence (named as Influencers) in a specific domain. Therefore, it is significant to accurate evaluate the social influence of participants and identify those Influencers from social networks. In the literature, many social influence evaluation methods have been proposed [5, 6, 7, 8, 9, 10, 11, 12, 13, 14], in which, Independent Cascade (IC) model [5] is a typical model to find the Top-K nodes who have the maximal social influence in a network. Subsequently, some important works [11, 9] are proposed to improve the scalability of  IC model. In addition, in recent years, the Local Influence Maximization method [15] has been proposed to evaluate the social influence of a specific participant in OSNs. Furthermore, as some OSNs are becoming a large real-time generator of social data-streams, some streaming methods [16, 17] have proposed to evaluate the social influence of participants in OSNs. ۱٫۲٫ The Problem and Motivation As illustrated in Social Psychology [18, 19, 20], the social trust between participants (e.g., students trust their lecturers in a specific research area), the social relationship between participants (e.g., the relationship between a father and his song), and the preference similarity between participants (e.g., they all like to play basketball) have significant influence on participants’ decision-making, and thus impact their social influence. However, these important social contexts are not fully considered by the existing social influence evaluation methods. Thus, these methods cannot deliver accurate social influence evaluation results. In addition, with the growth of network scale and complexity, social networks become susceptible to different types of unwanted and malicious spammer or hacker actions [21]. We propose a trust-oriented ranking strategy to defend against this kind of attack. Example 1: Figure 1 depicts a social network from Epinions, which contains five participants (i.e., P 1 to P5, they are all buyers). The trust relationship (represented as arrows with solid lines) between P1 and P3, P2 and P3, and P4 and P3 can be established based on the quality of the product review of P3. Their social relationship and preferences can be mined from the their profiles and purchase history [22]. Suppose P 1 has closer social relationships, and has more similar preferences to P3 than that of P2, then P3 can more likely affect the purchasing behavior of P1 than P2, which is not identified by the existing social influence evaluation methods. However, in traditional social influence evaluation methods [5, 6, 7, 8, 9, 10], the probability of the influence between two nodes is random in LT model or unified in IC model, which cannot reflect the realistic influence of participants. In addition, if P 5 is a spammer, in tradition social influence evaluation models, like the triggering model [5] and iterative model [23], he/she can utilize plenty of spam neighbors to establish fake strong social influence to affect P1’s decision-making. The above mentioned problems motivate us to develop a social influence evaluation method to accurately evaluate participants’ social influence in OSNs. In this paper, with considering the above mentioned important social contexts, we propose a Trust-Oriented Social Influence evaluation method, called TOSI by adopting iterative method. Since our method is convergent fast, thus we can deliver accurate social influence evaluation results with good efficiency.