Distributed collaborative information systems allow multiple organizations to share resources. Information sharing or transferring among different organizations is achieved through a collaborative social network. The network is a combination of multiple small social networks, each of which belongs to an organization. The objective of this paper is to construct a collaborative social network that transfers accurate information among multiple small networks, without disclosing the private structures of the small networks. This is comparable to building a backbone router to share information among multiple branch routers, without revealing any private structures of the branch router.

The author proposes a privacy-preserving protocol for collaborative social network construction that also considers the accuracy of the shared information. The author shows that the constructed collaborative social network achieves a good balance between privacy risk and information utility.

The goal of the protocol is to decide whether or not to keep the edge between two nodes (persons) by comparing the total communication count with the given threshold (in this paper, 500). The count is the private data a party wants to protect. Assuming there are n parties, and that the total count of the n parties is more than 500, then the edge between these two nodes will be kept.

The protocol consists of four steps:

(1) A party P₁ generates a cryptographic key pair (e, d) by using a homomorphic encryption scheme; e represents the encryption scheme and d represents private data.

(2) “P₁ sends P₂ the encryption of the private value that is masked by a digital envelope.”

(3) “P₂ computes the multiplication between the received term and the encryption of the masked private value by another digital envelope.”

(4) Repeat until P₁ and P_n collaboratively obtain the result of whether the total count is more than a threshold T or not.

The time complexity of the protocol is O(2α(n+k)), where α is the total bytes for each communication, n is the total number of parties, and k is the total number of random integer numbers in step four of the protocol.

The author verified the proposed protocol’s performances on the Enron email dataset and a research article dataset. There are three different experiments. The first focuses on how the percentage of the whole dataset influences the time complexity of the protocol. The second focuses on how the number of parties influences the time complexity and communication cost of the protocol. The third focuses on how the total number of random integer numbers used in step four of the protocol influences the time complexity and probability for the “advantage of one party to gain the other party’s private data.”

The experimental results of both datasets show that both time complexity and communication costs increase with the number of parties, data size, and the total number of random integer numbers. The probability of the advantage that each party can gain the information about other parties is very small on both datasets, which indicates that the proposed protocol can efficiently protect private information.

The paper has some limitations. Perhaps the most important is that although “the total weighted count of communications between two nodes” is an important notation in the proposed protocol, its definition is not clear. Furthermore, the author does not clarify the relation between this communication count and the local weighted count of a party, which is referred to as the private data that a party wants to protect.

Also, there are some issues with the experiments’ datasets. For instance, the author lists that there are 150 users and 192,829 edges in the Enron email dataset. However, for a directed graph, given 150 nodes, the maximum number of edges is 150*(150-1), which is less than 192,829. In the research article dataset, the author claims that edges with a weight of less than 20 were removed from the dataset; however, this somewhat high threshold of 20 will make the new dataset very small, which may not be appropriate for a protocol study.