Session-based recommendation systems are very common and important in our daily lives. However, current session recommendation systems tend to recommend items that users have interacted with in previous sessions, making it difficult to suggest new items that users have never interacted with. This limitation reduces the user’s interaction range and diminishes their desire to engage. Thus, how to effectively and reasonably recommend new items to users has become an urgent problem that needs to be addressed in session recommendation. This paper proposes a new item recommendation method based on hypergraph enhancement. By utilizing a hypergraph network, global item embedding representations are learned, which enhances the recommendation of new items that have not been generalized in the session. The goal is to obtain more accurate new item embeddings. Additionally, considering the sparsity of sessions, the similarity between sessions is fully taken into account from the perspective of user intent, which enhances the current session representation and enables the system to recommend new items more effectively. Experiments were conducted on two public datasets, and comparisons were made with state-of-the-art models. The experimental results validate the superiority and effectiveness of the proposed method.