Valuation of non-fungible tokens

NFTs signify a fundamental change in our understanding and engagement with digital assets, introducing a fresh phase in blockchain technology defined by distinctive ownership, decentralized marketplaces, and empowered creators. NFTs have already become a market worth billions of dollars, highlighting the critical necessity of robust and widely accepted valuation mechanisms within the NFT realm for artists, collectors, and investment managers. Although NFTs are currently in their early stages, we anticipate significant growth that will see them capture a substantial portion of the art, collectibles, gaming, and community markets. Moreover, we foresee their expansion into various non-fungible real-world goods and services. In essence, we predict that by the end of this decade, NFTs will evolve into a market valued in the trillions of dollars.

Inspired by the significant potential impact outlined above, we have strategically woven NFTs into our research agenda, with a particular emphasis on their data-related dimensions. Our focus lies within the realms of intelligent analytics and machine learning. We seek not only to understand their implications but also to unlock novel avenues for innovation and progress at the intersection of technology and finance.

Our research draws extensively on the data resources provided by NFTValuations, a specialized platform focused on NFT valuation methods.’s mission is centered on achieving two key objectives: firstly, to develop a robust and defensible methodology for calculating market capitalizations on an ecosystem-wide scale; and secondly, to establish a reliable and automated approach for valuing individual NFTs, complete with relevant confidence intervals. Overall, the unbiased valuation offered by creates a novel understanding of the NFT space.

Although valuation presents an intriguing research challenge, our exploration extends beyond this realm. Leveraging the data resources mentioned earlier, we have embarked on a comprehensive scientific inquiry focused on two main pillars: representational learning and machine learning. This endeavor encompasses a vertical investigation into fundamental scientific questions, facilitated by the wealth of data at our disposal.

Representational learning stands at the forefront of our research endeavors, serving as the vanguard in handling input data. It prioritizes the extraction of pertinent features essential for driving the subsequent modeling phase. This encompasses a diverse array of topics, including but not limited to the following:

  • Feature Learning: The process involves extracting significant features and representations not only from the content of Non-Fungible Tokens (NFTs) themselves but also from associated datasets. This encompasses the identification and extraction of relevant information essential for various tasks, including NFT valuation.
  • Manifolds: Due to the inherent high-dimensional complexity of potential feature spaces, the objective is to establish resilient low-dimensional representations, such as embeddings. These embeddings facilitate robustness against noise and facilitate the capture of nonlinear relationships among features, crucial for tasks like NFT valuation.
  • Sparse Coding: Complementary to manifold-based representations, sparse coding endeavors to manage feature representation by promoting sparsity based on linear relationships. It aims to efficiently represent features of interest while minimizing redundancy, which is particularly pertinent in the context of NFT analysis.
  • Transfer Learning: This technique aims to leverage knowledge gained from related tasks to mitigate the need for extensive feature engineering when tackling new tasks. By transferring learned features, models can adapt more effectively to novel tasks, including NFT valuation, thereby enhancing efficiency and performance.
  • Multimodal/Crossmodal Learning: Given the diverse nature of features associated with NFTs across different modalities, the challenge lies in developing representations that not only reconcile feature spaces derived from disparate modalities but also establish cohesive joint multimodal representations. This facilitates a comprehensive understanding of NFT content, aiding in valuation and analysis.
  • Zero/One-Shot Learning: This constitutes a particularly demanding area of research within the field, aiming to establish feature spaces conducive to deploying models when labeled data is scarce or absent. The objective is to enable effective model training even with limited or no labeled data, addressing challenges inherent in scenarios like NFT analysis and valuation.
  • LLMs Attention: Delving into one of the foundational concepts powering Large Language Models (LLMs) – attention – with the objective of capturing intricate long-term dependencies among features.

Machine learning constitutes the cornerstone of our research pursuits, embracing both unsupervised and supervised approaches to tackle various tasks. This comprehensive approach is instrumental in addressing a diverse range of topics, which include, but are not limited to, the following:

  • Abnormal Pattern Detection: Our focus lies in identifying anomalies or novelties within datasets, applicable across diverse contexts such as NFT pricing dynamics and market-related occurrences.
  • Latent Structure Revelation: Employing unsupervised techniques like hierarchical clustering, we delve into uncovering patterns reflective of the intrinsic structure of the NFT market. Augmenting our investigation with tools from graph theory aids in quantifying metrics such as decentralization.
  • Valuation Estimation: Aligned with the objectives of, our research extends towards achieving unbiased NFT valuation estimates. We delve deeper by exploring the potential integration of NLP models, particularly in sentiment analysis, to refine valuation processes.
  • Fraudulent Pattern Classification: Acknowledging the susceptibility of the NFT space to fraudulent activities, we approach this challenge as a classification problem. Our efforts entail developing task-specific classifiers, crucial for identifying fraudulent practices like wash trading within the NFT market.
  • Model Fusion: Research across the aforementioned domains generates multiple models, prompting exploration into fusion methodologies. This research direction encompasses various fusion strategies, including amalgamating models for shared tasks and integrating diverse task-specific models in the context of broader tasks, thereby offering a comprehensive framework for leveraging multiple models effectively.
  • Deep Learning: Recognizing the transformative potential of deep neural networks, we emphasize their fundamental properties, such as regularization techniques and attention mechanisms inspired by LLMs. We investigate how these properties can be leveraged in addressing NFT-specific challenges, particularly in valuation and classification tasks.


The following list includes an indicative number of related peer-reviewed publications.

  • Christodoulou, K., Katelaris, L., Themistocleous, M., Christodoulou, P. and Iosif, E. “NFTs and the metaverse revolution: research perspectives and open challenges”. In “Blockchains and the Token Economy: Theory and Practice”.
  • • Zanardo, E., Domiziani, G.P., Iosif, E. and Christodoulou, K. Identification of Illicit Blockchain Transactions Using Hyperparameters Auto-tuning. In “Principles and Practice of Blockchains”. Cham: Springer International Publishing.
  • Themistocleous, M., Christodoulou, K. and Katelaris, L. “An Educational Metaverse Experiment: The First On-Chain and In-Metaverse Academic Course”. In Proceeding s of European, Mediterranean, and Middle Eastern Conference on Information Systems. Cham: Springer Nature Switzerland

Furthermore, in a broader context, indicative papers which focus on the extraction of business intelligence utilizing unsupervised learning for web mining, are listed below.