Determination of trust rates of individuals in social networks

The increasing importance of social networks has increased research on trust estimation and interpretation of trust between entities (individuals) in the network. It is important to estimate the inter-user trust rate to minimize the risks in the interaction of users. This thesis expresses trust rates numerically, allowing an entity to identify the most and least trusted entities of the network. In this thesis, three different methods were used to answer the questions "does entity X trust entity Y?" and "which entity is trustworthy and which is not?" in the social network. In the first method, all paths between two nodes are found and the probabilities of each edge on these paths are found using a Markov chain. These probability values are multiplied along the path and a ratio is found. Then, a trust score is obtained by summing the trust ratios between the node and all other nodes. In the second method, all paths between all individuals are found using breadth-first search. All paths to a node are summed up and paths from each node are proportioned to this sum and trust is expressed as a percentage. According to the percentages found, the entities that the entity under study trusts and does not trust in the network were identified. As a third method, the trust ratio between individuals was calculated using linear algebra techniques to reduce the arithmetic overhead involved in calculating trust. When a process similar to the Markov process is applied to the adjacency matrix of an undirected graph with OTG (Ratio-based trust), the resulting matrix is treated as a probability transition matrix. Here the kth power of the matrix represents the trust at distance k between pairs of nodes in the graph. Finding the confidence values in this way has a high time cost of . Linear algebra is used to avoid these costs. In PTG (Path-based trust), the number of walks between individuals is calculated by taking the power of the walk length of the adjacency matrix of the graph. Taking the walk length power of the adjacency matrix is quite costly. Again, this cost is eliminated by using linear algebra. The trust detection algorithms known in the literature use shortest paths and similar methods to eliminate unimportant paths, which makes the result controversial due to data loss. In all three methods we used, we expressed the trust ratios numerically without data loss and identified the most trustworthy and most untrustworthy entities in the network.