A Multidimensional Network Link Prediction Algorithm and Its Application for Predicting Social Relationships

Abstract In the "We the Media" era, the rules for forming social users' following relationships are complex. Links generated between two social user nodes are influenced by not only the structural information of their social network nodes but also the users’ occupational environments, interests in opinions, topics, social psychology, etc. In the existing studies of link prediction in complex networks, predicting the possibility of link generation between two nodes that have not yet generated edges in a complex network is calculated mainly from the known network nodes and structural information. Such studies, in which the main predictors are the structural similarities among social nodes or user location nodes, cannot fully explore and utilize the social network node users’ public opinion characteristics. To quantitatively identify the influence of different dimensions of public opinion factors on predicting links between social user nodes, we present a study on the prediction of social network links in "We the Media" networks. Starting with the characteristics of the elements found in public opinions on “We the Media” networks, in which public opinions are multidimensional and multilayered and possess multiple attributes, we built a multidimensional network model oriented towards the topology of public opinions on “We the Media” networks. Combined with an analysis of the driving factors in the formation of user-node relationships in social networks, we designed a prediction algorithm that works on multidimensional network links. Furthermore, we conducted an empirical analysis of social relationship prediction, whose effectiveness has also been compared with baseline methods such as the Common-Neighbourhood-Driven model, the Jaccard index, and the SimRank method. We chose the area under the curve (AUC) as the indicator of link prediction and evaluation using “We the Media” public opinion data from Weibo.com. The research findings of this paper can be summarized as follows: (1) The effectiveness of the multidimensional network link prediction algorithm is significantly higher than those of the baseline methods. (2) The prediction algorithm presented in this paper works on multidimensional network links and can evaluate the effects of different dimensions of public opinion factors on the prediction of user-node links in social networks. (3) The element of occupational environment improves the accuracy much more than the element of user interest in opinions and topics when predicting user-nodes’ links, while the element of social psychology reduces the accuracy.