International Journal of Computer Applications (0975 – 8887) Volume 93 – No 1, May 2014 17 A Secure Online Reputation Defense System from Unfair Ratings using Anomaly Detections Asha baby PG Scholar,Department of CSE S.K.P Engineering College Tiruvannamalai, India A. Kumaresan Professor, Department of CSE S.K.P Engineering College Tiruvannamalai, India K. Vijayakumar Professor, Department of CSE S.K.P Engineering College Tiruvannamalai, India ABSTRACT A reputation system collects feedbacks from users and aggregates these feedbacks as evidence and generates the aggregated results to the normal users. These aggregated results are called reputation scores. We can call this system as online feedback-based reputation system. To protect the reputation system many defense schemes have been developed. In this paper we propose a defense scheme; it is the combination of five modules. Evaluation based filtering, Time domain unfair rating detector, suspicious user correlation analysis, trust analysis based on Dempster-Shafer theory and malicious user identification and reputation recovery. This system identifies the items under attacks, the time when the attacks occur and unfair raters who insert unfair ratings. Compared with existing systems this system achieves detection of high unfair ratings and reduces the detection of false dishonest ratings. General Terms Unfair ratings, user correlation analysis, trust analysis, belief function, Euclidean distance Keywords CUSUM detector, Dempster-Shafer theory, K-mean algorithm 1. INTRODUCTION Many of the people are using internet for their daily life such as entertainment, making personal relationship and business purposes. The internet has created large opportunities for online interactions. However the internet is more vulnerable to attacks, which makes online interactions risky. It will ask a number of questions in online interactions. For example Will a seller at an online shopping site provides the product in correct time? Is Amazon.com site will produce high quality and trustworthy product?[7] Is a video on YouTube really interesting? Here is one problem that satisfies how the online participants protect themselves by identifying the quality of strangers and unfamiliar items . To solve this problem online reputation systems are introduced. A reputation system collects feedbacks as evidence, about individual items, including products, services, and digital contents, aggregate the evidence and disseminates the aggregated results into the normal users. These results are called reputation scores. The system which provides rating is referred to as online feedback-based reputation system [3].In order to protect reputation systems, many defense schemes have been developed. The efficiency of the defense scheme depends on the accuracy of the reputation system. Without the proper defense scheme items reputation score increase or decrease rapidly. It will reduce the quality of the reputation system. In this paper we propose a defense scheme. It consists of 5 modules, Evaluation based filtering, A Time domain unfair rating detector[1], Suspicious user correlation analysis, Trust analysis based on Dempster-Shafer theory and finally Malicious user identification and reputation recovery. This system accepts time domain rating sequences as input, an unfair rating detector will identify whether changes occur due to rapidly or changes accumulated over time. Here revised CUSUM detector[9] is used. Which detects the change intervals, it is given to the correlation analysis module, and from suspicious user intervals this module forming the clusters containing suspicious users with a smaller distance using k-mean clustering algorithm. After forming the clusters, group rating will be calculated by taking average rating score of each users in this group. This module filtering out number of malicious users. Even though these clusters contain malicious users. In order to improve the detection rate again these clusters are given to the 3 rd module trust analysis .It will calculate the trust value based on the Dempster-Shafer theory[9] and find out all malicious users based on their trust value. Here trust value is calculated by collecting history of individual users. Lower trust value indicates malicious user and finally accurate reputation score will be calculated by considering only honest raters. The performance of this system is evaluated with existing systems such as [3], [4], [5], [6]. Our system provides better detection rate and low false alarm rate. The remaining of the paper is arranged as follows section 2 describes the proposed mechanism, section 3 deals with the results and final section 4 gives the conclusion.