<p>With the increasing complexity of academic graph networks, over-fitting has become a great challenge for graph-based paper recommendation. Existing methods mainly focus on knowledge distillation to solve above problem by compressing the volume of graph networks. However, incomplete distillation between two graphs would lead to the neglect of author’s diverse interests, resulting in biases for research interests. Therefore, we propose a new triple-graph interactive distillation recommendation (TIDRec) method for paper recommendation. Specifically, we construct a triple-graph interaction to complete the distillation knowledge between graphs to correct biased research interests of authors. First, a main model is built that integrates knowledge from both graphs (i.e., writing–citation relationship and author–author co-authorship graph) to initialize inner product vectors, capturing global research interests. Then, two auxiliary models with single graph knowledge are constructed to generate distinct inner product vectors, mining local research interests, respectively. Next, a triple-graph interactive distillation approach is designed to continuously correct the global research interests with distill vectors of each other. Finally, papers highly relevant to global research interests are recommended to authors. Extensive experiments prove that TIDRec surpasses state-of-the-art approaches, with an average performance improved by 8–25% for all four metrics.</p>