A novel complex network link prediction framework via combining mutual information with local naive Bayes.

As an important research direction of complex networks and data mining, link prediction has attracted more and more scholars' attention. In the early research, the common neighbor is regarded as a key factor affecting the formation of links, and the prediction accuracy is improved by distinguishing the contribution of each common neighbor more accurately. However, there is a drawback that the interactions between common neighbors are ignored. Actually, it is not just the interactions between common neighbors, but all the interactions between neighbor sets contribute to the formation of links. Therefore, the core of this work is how to better quantify and balance the contributions caused by common neighbors and the interactions between neighbor sets, so as to improve the accuracy of prediction. Specifically, local naive Bayes and mutual information are utilized to quantify the influence of the two aspects, and an adjustable parameter is introduced to distinguish the two contributions in this paper. Subsequently, the mutual information-based local naive Bayes algorithm is proposed. Simulation experiments are conducted on 5 datasets belonging to different fields, and 9 indexes are utilized for comparison. Numerical simulation results verify the effectiveness of the proposed algorithm for improving link prediction performance. Published under license by AIP Publishing.