Sign prediction in social networks based on users reputation and optimism

Online social networks are significant part of real life. Participation in social networks varies based on users needs or interests. Often, people participate in these platforms due to their interests. Social media not only consist of dense connected components (communities), but also these platforms are dynamic. The dynamism also includes formation and deformation of connections. In some online social networks, connections are mapped to positive and negative links. Positive connections are sign of friendship or trust, while negative links show enmity or distrust. Community structures and temporal traces can also be observed in signed networks. Networks with both positive and negative connections occur in various fields of applications. Reliable prediction of edge sign has a significant influence on friendship formation or enmity prevention. Prediction of edge signs have been considerably explored, however, we intend to discover simple and noticeable social properties in order to identify the connections’ future in networks consisting of both positive and negative links. In order to approach this goal, we investigate real-world signed social networks and build several prediction models. Additionally, simple social properties of trust/distrust networks are employed. Two local nodal measures, called reputation and optimism, are introduced. A node’s reputation indicates how popular the node is. Conversely, its optimism measures its voting pattern toward others. To reduce inherent biases in voting, we also introduce an algorithm to compute the nodes’ reputation and optimism. These rank-based metrics are computed based on nodes’ ranking scores in the network. Furthermore, we employ reputation and optimism of trustor and trustee to predict the sign of the edges in a number of real signed networks including Epinions, Slashdot and Wikipedia. Finally, several classifiers are applied for this purpose. Our experiments show that these simple features have superior performance over state of the art methods