Privacy-Preserving Assessment of Social Network Data Trustworthiness

Extracting useful knowledge from social network datasets is a challenging problem. While large online social networks such as Facebook and LinkedIn are well known and gather millions of users, small social networks are today becoming increasingly common. Many corporations already use existing social networks to connect to their customers. Seeing the increasing usage of small social networks, such companies will likely start to create in-house online social networks where they will own the data shared by customers. The trustworthiness of these online social networks is essentially important for decision making of those companies. In this paper, our goal is to assess the trustworthiness of local social network data by referencing external social networks. To add to the difficulty of this problem, privacy concerns that exist for many social network datasets have restricted the ability to analyze these networks and consequently to maximize the knowledge that can be extracted from them. This paper addresses this issue by introducing the problem of data trustworthiness in social networks when repositories of anonymized social networks exist that can be used to assess such trustworthiness. Three trust score computation models (absolute, relative, and weighted) that can be instantiated for specific anonymization models are defined and algorithms to calculate these trust scores are developed. Using both real and synthetic social networks, the usefulness of the trust score computation is validated through a series of experiments.