The classification of cryptocurrencies remains an open challenge to make valid decisions due to their diverse technical structures, financial applications and evolving use cases. The scientific literature does not provide a simple technical categorization that facilitates asset comparison, enhances risk measurement, provides a structured approach to understanding the dependencies between different crypto-assets and facilitates decision-making processes among a wide range of stakeholders. This study proposes a technical categorization framework that classifies cryptocurrencies based on their underlying blockchain infrastructure or smart contract functionalities.
The authors designed the categories and classify the top 100 market cap cryptocurrencies with them. To validate the proposal, the same task was executed by using multiple large language models (LLMs), including ChatGPT, Perplexity, Claude and Gemini; with zero-shot classification approach.
The results indicate that, when prompted with predefined categories, LLMs achieve substantial agreement with human classification, with ChatGPT demonstrating the best results. Moreover, categorization without any guidance is inconsistent across models, often defaulting to use-case- based groupings. Notably, providing additional information about cryptocurrencies or detailed definitions of categories does not significantly alter classification outcomes, suggesting that LLMs rely predominantly on their internal knowledge base.
Future research should focus on refining empirical measures for decentralization, expanding classification testing with human participants and leveraging advancements in LLMs for improved categorization accuracy.
This study highlights the potential of LLMs as tools for the systematic classification of cryptocurrencies, a key part of an important organizational decision-making process. It is remarked that the importance of having structured categories of cryptocurrencies is for all kinds of decision-makers, including investors, industry stakeholders, fund managers and regulators. Future research should focus on refining empirical measures for decentralization, expanding classification testing with human participants and leveraging advancements in LLMs for improved categorization accuracy.
- (1)
There is a gap in cryptocurrency classification from the asset perspective
- (2)
The proposed technical categorization classifies cryptocurrencies based on their technicalities
- (3)
ChatGPT and Perplexity zero-shot achieve substantial agreement with human classifiers
- (4)
LLMs understand the categories without further explanation and help in decision-making
- (5)
Providing additional information about the assets does not alter the LLMs classification
