Community detection means dividing the nodes in complex networks into different groups. Nodes within the same group are closely connected, while nodes in different groups have fewer connections. Community detection is fundamental problem in network analysis, aiming to uncover the underlying structures and organization within complex networks. Traditional methods focus on network topology, neglecting valuable information contained in different types of links. Improving the network structure purposefully can result in better outcomes in community detection. In this study, we have utilized mixed link prediction as a technique to enhance the network structure. Our goal was to eliminate any noise in the network and restore any missed links without altering the number of nodes and edges. We then proceeded to apply various community detection algorithms to compare the quality of the results. To ensure the generality of our approach, we chose the most popular community detection methods (Louvain, Giravan Newman, and Fast Greedy) and link prediction ranking formulas (Common Neighbors, Jaccard Coefficient, Adamic/Adar, Preferential Attachment, and Recourse Allocation) as the core of mixed link prediction. To evaluate the effectiveness of our proposed method, we test it on four different real-world datasets from various domains based on modularity and normalized mutual information measures. Our findings demonstrate that our novel framework for community detection using mixed link prediction improves community detection results in most cases. The success rate also depends on the network properties. Furthermore, this approach has the potential to be extended to stronger community detection and link prediction methods in future researches.
