With the rapid development of academic social media, the problem of node influence diffusion in scholar social networks has increasingly received extensive attention from the field of influence maximization (IM). Existing learning-based methods for solving the IM problem usually rely solely on network topology or individual node activities, lacking comprehensive consideration of both network topology and important information of nodes, leading to poor model performance. By comprehensively considering the network topology as well as the global information and importance of nodes, we propose a deep reinforcement learning (DRL) framework, named APGD-IM, which is based on an adaptive DeepWalk algorithm and a prior-enhanced graph neural network (GNN), aiming to optimize the performance degradation caused by the above issue. Specifically, we propose an adaptive DeepWalk algorithm DRA based on attention mechanism and node importance information, along with a prior-enhanced GNN module PGNN, for generating node embeddings. These embeddings are then used to learn parameters by combining double deep Q-network to address scholar influence maximization problem in social networks. Experimental results on four real-world social networks demonstrate that our proposed model outperforms other baseline methods and maintains stable performance advantages across different diffusion models.