Exploratory Cloud Service Recommendation Method Based On Trust Relationship In The Social Network Environment

With the development of big data, internet of things, cloud computing and other Internet technologies in recent years, cloud service is known widely and has a big influence in production and daily life of people. However, when the user faces many functional-similar cloud services, they will be hard to do selection, which is commonly referred to as "information overload". In the field of E-commerce, recommendation system is the best way to solve the "information overload" problem. Recommendation algorithms use the behavior characteristics of users to mine individual preferences of users and recommend the goods which fit their preferences. Recommendation systems have good performances in the Amazon, Tmail and other domestic and international significant e-commerce websites, which bring benefits to companies. However, cloud services and traditional goods have nature differences and particular properties. QoS (Quality of Service) can be used to evaluate cloud service qualities and characteristics objectively. This property is not available in traditional commodities, including response time, throughput and other evaluations, and has regional and temporal characteristics. It’s worth noting that trust relationships between users are rarely taken into account in traditional recommendation.Combining characteristics of cloud service user and cloud service, this paper proposed trust-based recommendation method in the social network environment. Firstly, we use Pearson correlation coefficient and Jacobi coefficient to estimate the objective similarity of the cloud service user. Then, we calculate the trust degree based on interaction degree between cloud service users, which takes direct trust relationship, indirect trust relationship, hybrid trust relationship and dynamic of trust relationship into account. Finally, we use the trust degree to enhance and correct the the objective similarity, and find similar users to predict the null values and then obtain the recommendation list. Experiments shows that the method we propose has the higher prediction accuracy than traditional recommendation methods, and prove the convergence and effectiveness of our method through the parameters adjusting to get the optimal values.