Online social networks (OSNs) provide community platforms that engage users. The public page is a popular example. These pages are connected through “like” relationships, creating online community networks and neighborhoods. We investigated the pivotal features influencing link formation and neighborhood structuring within the page graph by exploring a series of potential features, both graph-based and content-based. Our methodology combines node similarity and Graph Neural Networks to perform link prediction. We identified the page state label as the single most accurate predictor in link prediction tasks, which also is the most efficient feature with the smallest number of classes. Moreover, we observe that augmenting the page state label feature with the page node degree and page city population features further enhances link prediction accuracy. Page location label shows a strong effect on pages connecting with their neighbors.