Research On Shilling Attack Detection Algorithm Based On Hidden Markov Modeland Clustering

Collaborative filtering recommendation system is widely used in the field of e-commerce because of its personalized recommendation,which effectively alleviates the problem of "information overload" on the network.However,some attackers inject fake user's rating into the system for the purpose of profit,which will affect the recommendation result and achieve the goal of malicious competition.Such behavior seriously threatens the security of the recommendation system and reduces the user's trust in the recommendation system.To protect the safety of recommendation system and provide users with reliable and authentic recommendation results become the hot spot of scholars at home and abroad.In response to this problem,this paper proposes two kinds of shilling attack detection algorithms from the perspective of unsupervised.At first,aiming at the limitation of existing unsupervised detection method in detecting shilling attack,this paper proposes a method of unsupervised attack detection based on hidden Markov model from the perspective of user behavior.By analyzing the user's historical rating data,the attack user is detected from the difference of the user's rating habits.The detection method firstly uses the Hidden Markov Model to acquire the user's preference sequence and proposes a method for calculating the user's matching degree by analyzing the preference sequence of each user and further calculates the user's suspicious degree by analyzing the difference between the real user and the fake user's rating behavior Finally,the hierarchical clustering method is used to cluster the users to get the set of attack users.Secondly,aiming at the problem of misjudgment of true users in unsupervised attack detection,this paper proposes a method of unsupervised attack detection based on improved K-means clustering and item's popularity.The algorithm analyzes the user's rating behavior from the perspective of clustering and attempts to improve on the traditional K-means clustering algorithm,so that the improved K-means clustering algorithm can gather most of the attack users and gather as few real users as possible to reduce the impact of shilling attacks on the recommendation system.Finally,the corresponding experiments are designed for the above two types of shilling attack detection algorithms.The experimental results verify the effectiveness of the proposed algorithm by comparing with the existing attack detection algorithm.