Predicting user-item interactions in e-commerce platforms plays an important role in improving recommendation systems and optimizing user experience. This paper presents an in-depth exploration of using temporal graph networks (TGNs) for user-item interaction prediction. Using TGN on datasets, Amazon, LastFM, and RetailMarket, we experiment node classification and edge prediction tasks. Our approach achieves an accuracy of 80% and an AUC score of 0.93. Our findings prove the effectiveness of TGNs in identifying patterns and predicting interactions, thus advocating for their integration into e-commerce platforms to improve recommendation systems and user satisfaction.