Privacy Preserving Reputation Systems for Decentralized Environments

Reputation systems are a key technology for making users accountable for their behavior in online communities. A reputation system computes the reputation of a user based on the feedback provided by the community. If the user draws negative feedback, it loses good reputation and may eventually be excluded from the network. It has been observed that users in a reputation system often hesitate in providing negative feedback due to the fear of retaliation. A solution to this issue is privacy preserving reputation systems, which compute reputation such that the individual feedback of any user is not revealed. However, many existing privacy preserving reputation systems rely on centralized constructs and are thus not suitable for decentralized environments. The ones that are decentralized are either limited to specialized platforms (such as anonymous networks and trusted platforms), rely on trusted third parties, do not protect privacy under strict adversarial models, or are expensive in terms of resource utilization (for example, O(n) messages, where n is the number of users providing feedback). In this thesis, we present privacy preserving reputation protocols, that are decentralized, do not require specialized platforms nor trusted third parties, protect privacy under a range of adversarial models (semi-honest, non-disruptive malicious, disruptive malicious), and are more efficient than comparable protocols (the most expensive protocol requires O(n) + O(log N) messages, where N is the total number of users). The techniques that we utilize include trust awareness, data perturbation, secret sharing, secure multi-party computation, additive homomorphic cryptosystems, and zero-knowledge proofs. Another key innovation in our protocols is that an entity is able to quantify the probability of disclosure of its feedback. If the risk is unacceptable, the entity can abstain from submitting feedback. The protocols are thus able to offer up to perfect privacy, which has been previously assumed to be impossible in protocols for decentralized environments that compute reputation in an additive manner. We also address some issues related to trust recommendation and propagation. In particular, we present a solution to the problem of subjectivity in trust recommendation. Experimental results indicate the effectiveness of the proposed strategies.