Entropy-based recommendation trust model for machine to machine communications

Abstract

In a vast data collection and processing applications of machine to machine communications, identifying malicious information and nodes is important, if the collected information is to be utilized in any decision making algorithm. In this process, nodes can learn behaviors of their peers in the form of recommendation from other nodes. These recommendations can be altered due to various motives such as badmouthing honest nodes or ballot stuffing malicious nodes. A receiving node can identify an incorrect recommendation by computing similarity between its own opinion and received recommendations. However, if the ratio of false recommendations is low, the similarity score will be insufficient to detect malicious misbehavior. Therefore in this paper, an entropy-based recommendation trust model is proposed. In this model, a receiving node computes the conditional entropy using consistency and similarity of received recommendations with respect to its own opinions. The computed entropy indicates the trustworthiness of the sender. The proposed model clearly distinguishes malicious nodes from honest nodes by iteratively updating trust values with each message. The performance of the model is validated by a high true positive rate and a false positive rate of zero.