In this paper, we explore the application of Knowledge Graph Embedding (KGE) techniques and Few-Shot Relational Learning (FSL) methods to the domain of digital assets, particularly focusing on NFT recommender systems. We evaluate the effectiveness of various KGE approaches, including TransE, Node2Vec, and GraphSAGE, to model user-token interactions and improve recommendation accuracy. Additionally, we address the new token prediction problem, a challenge inherent to NFT platforms and blockchain transactions, where new tokens with little interaction history need to be recommended. Two implementations of the proposed models were tested on a comprehensive dataset from the year 2023, allowing for robust evaluation of their performance. The results demonstrate the potential of combining KGE and FSL for enhancing NFT recommendations and predicting token relationships in dynamic digital asset markets.