Bundling recommendation enhances user experience and boosts merchant sales performance by offering predefined sets of product combinations. It also plays a significant role in various service ecosystems such as video-on-demand and music playlist generation. Existing bundling recommendation methods often rely on shared model parameters or multi-task learning schemes, neglecting the deep-level connections among users, items, and bundles, which leads to information loss and impacts the performance of recommendation systems. To address these issues, this paper proposes a complete hypergraph neural network (CHNN) . First, the framework constructs a complete hypergraph to express the ternary relationships among users, items, and bundles. These ternary relationships not only include the interconnections among users, items, and bundles but also encompass the internal connections within users and bundles, effectively describing the relationship between product bundles and user preferences. Second, the model consists of an initialization layer, a triple convolution layer, and a prediction layer. The initialization layer generates embedding vectors for each user, item, and bundle. The triple convolution layer extracts information from the complete hypergraph and leverages the user-bundle graph and item-bundle graph to enhance the representations of users, items, and bundles. The prediction layer provides recommendations based on the final embedding vectors. Through multi-layer rich convolution operations, the model fully explores the associations contained in the complete hypergraph to achieve more accurate recommendations. Experiments on two real-world datasets, NetEase and YouShu, demonstrate that CHNN achieves an average improvement of 2.4% in Recall and 2.75% in NDCG, outperforming existing baseline models and showcasing its effectiveness in the field of bundling recommendation.