Link prediction methods based on strengthening projection model in large-scale bipartite networks

Many complex networks created from the real-world data are bipartite structured by nature. Bipartite networks give important ideas about the complex nature of the real-world systems, they are also complex network types that represent the interactions between the different node groups. Recently, the link prediction studies on large-scale and complex networks have particularly become the focus of interest for researchers in various scientific fields. They have a constantly changing and evolving structure. A system that would propose popular topic titles among these networks that are related with an author's fields of work would simplify the work of many researchers. Recently, a lot of work has been done which address recommending academic topics to authors as a link prediction problem. Only a few of these studies use the bipartite networks despite its high practical interest and the specific challenges it raises. The majority of the previous works on the link prediction in bipartite networks focus on using the properties of traditional unimodal projection networks to predict the relations between the node pairs. The traditional projection methods involve many node pairs with weak relationships when used in networks constructed from real-world data. The analysis of these weak and unnecessary information requires high computation time. To cope with this problems, we proposed in this study the notion of strengthening projection networks which is the backbone of the network instead of traditional unimodal projection networks in bipartite graphs. Then, a link prediction method based on strengthening projection network have been developed to predict the potential links. Finally, two new proximity measure algorithms such as time-aware proximity measure algorithm and time-ignored proximity measure algorithm was proposed to evaluate the quality of the potential link. We test the proposed method on the academic information bipartite network constructed with collected data from IEEE XPlorer. Experiments on a real network demonstrate that the success of the proposed method is promising and it is possible to obtain fast and high-quality link prediction results from a large-scale network. Key Words: Bipartite Networks, Complex Network Analysis, Link Prediction, Academic Topic Recommendation, Strengthening Weighted Projection, Time-aware Proximity Measure