Development of a Hybrid Recommendation System for NFTs Using Deep Learning Techniques

Abstract

Recommender systems are widely used in domains such as movies, music, and e-commerce. Non-Fungible Tokens (NFTs), introduced through blockchain technology, have become a remarkable research topic due to their technological characteristics such as uniqueness, proof of ownership, immutability, and traceability. They are used in various fields such as art, finance, and education. However, research on NFT recommendation systems remains limited. NFTs introduce unique challenges due to their high sparsity of user-item interactions, diverse data types such as images, textual information, and transaction data, and blockchain anonymity, which leads to a lack of demographic and score data. These factors complicate the development of personalized recommendations. In this study, a personalized recommendation system for NFTs was developed using deep learning methods, leveraging the distinctive technological features of NFTs and addressing the challenges of the NFT domain. The proposed model, named NFT-NCFAE, utilizes Neural Collaborative Filtering (NCF) to capture user-item interactions and employs AutoEncoder (AE) to integrate diverse NFT-related data, such as images, text, prices, and transaction history, alongside user data. To evaluate the specific contribution of the AE within the developed model, an additional analysis was conducted using only NCF, focusing on user-item interactions without incorporating additional NFT-related data. Both models were tested on a dataset utilized in a previous study from the literature, and the results were thoroughly evaluated. The findings indicate that the NFT-NCFAE model outperforms both the existing study in the literature and the NCF model. Consequently, the NFT-NCFAE model has the potential to contribute significantly to the development of personalized NFT recommendation systems.