Hypergraph which is used to model a social network can more accurately represent diverse relationships compared to general graph. Studying the influence maximization (IM for short) problem in a hypergraph helps to identify and optimize key nodes and strategies for information dissemination in social networks. In this paper, we proposed a hypergraph neural network-based approach for the IM problem (HGNNIM for short) in hypergraphs. First, HGNNIM integrates intrinsic attributes of node with the hyperedge attributes it belongs to, employing feature fusion techniques to generate high-dimensional embeddings. Next, these embeddings are subsequently fed into regression model in HGNNIM to predict node influence values, which are trained under both Independent Cascade (IC for short) and Susceptible-Infected-Recovered (SIR for short) diffusion models to simulate real-world information spread dynamics. Finally, the trained model is then deployed on target networks to identify high-impact seed nodes and predict their potential influence coverage. The experimental results indicate that the nodes identified by HGNNIM can achieve a wider range of influence in given hypergraph compared to the existing baselines.