Research Of Trust-based Recommend Algorithm In E-business

With new technique and new application appearing ceaselessly in internet, e-commerce nowadays gets more opportunities and challenges. E-commerce have some unique requirements for recommend technique, include three important area: social recommend mechanism, personalized recommendation and robustness. Traditional algorithm has met bottleneck. To solve these problems, import trust into recommend system is a kind of very effective method. How to qualitative and quantitative trust is this paper’s main work.The current recommended system based on trust focused mainly on trust and the spread of trust said, but ignored how to calculate trust. The trust degree they used all gets from users. Based on past experience, according to previous research neglecting social recommend mechanism problems, this paper put forward the method of introduce community found into recommended system, find the user trust evaluation on others automatically. This is a theory that based on the achievement in trust study. Through judging the similarity and familiarity between users to find out the users’ trust relationship. This method does not require user actively submit the trust valuation of other users, but through the user ratings between social information mining.To the recommended system, user model is very important;lots of algorithms are based on user model launch. This paper will propose a kind of user model combined user personal model and community model. The user community model based on the information flow way structure user model via social information. This kind model can reflect user’s sphere of activities and the relationship between users much better. With user model we can figure out the user’s trust degree, and then find out user’s closest neighbor. This paper also put forward a method of user personal model updating algorithm, by introducing time factor to keywords weights calculations, to update keywords out the purpose of adaptive user’s interests. Experimental results shows that it solves the data sparse and cold start-up problem in the collaborative filtering algorithm because this system does not require user submit trust score. Compared with traditional collaborative filtering systems, the given system is better in accuracy and capacity of resisting the attack.