Literature Review

Abstract:

This paper presents TeaDAO's innovative approach to NFT price valuation, leveraging advanced AI-driven methodologies to achieve enhanced accuracy. Recognizing the challenges posed by diverse traits across different projects, we introduce compact parameters, including rarityScore and utilityScore, to encapsulate crucial information. Additionally, we address low transaction volume by enriching data using similarityScore, allowing us to approximate appropriate prices for NFTs with limited transaction history. We also explore ensemble modeling, leading to improved predictive accuracy. The results demonstrate significant progress, with our AI model achieving approximately 83% accuracy. This research lays the foundation for TeaDAO's pioneering role in the NFT valuation space, offering transparent and informed price evaluations to empower NFT enthusiasts and investors.

Literature Review

The valuation of Non-Fungible Tokens (NFTs) has garnered significant attention in the current market, with various approaches being explored. In this literature review, we will examine the two most prevalent methods for NFT valuation, namely the community-driven approach and the AI-driven model approach.

1. Community-driven approach: The community-driven approach relies on crowd intelligence to gather appraisals from community members. One prominent example of this method is Upshot, which initially utilized community-based appraisers and later incorporated Machine Learning (ML) techniques. The advantages of this approach include its ease of implementation and cost-effectiveness. However, it also exhibits some notable drawbacks, such as limited scalability and potential inaccuracies in pricing due to subjective biases.

2. AI-driven model approach: The AI-driven model approach has gained traction in the NFT valuation domain due to its scalability and efficiency. Several platforms, including NFT Bank, BankSea, and Bitscrunch, have adopted this method. Leveraging AI algorithms, this approach offers fast and scalable price predictions for NFTs. However, developing an AI model with accurate predictions for all NFT projects presents a notable challenge.

Considering the two approaches, we have opted to work on the AI-driven model approach for evaluating NFT prices, given its potential for scalability and faster processing times. The proposed methodology encompasses the following general steps, which have been adopted by leading platforms such as Chainlink and BankSea:

ChainLink methodology

BankSea Methodology

1. Prepare Data: The initial step involves data collection and preparation. Data from various sources, including historical NFT transactions, metadata, and any relevant external factors influencing NFT prices, is gathered and processed. Proper data cleaning and preprocessing techniques are applied to ensure data accuracy and consistency.

2. Construct Model: An essential aspect of the AI-driven model approach is the construction of a robust and efficient model. Machine learning algorithms, such as neural networks, random forests, or support vector machines, are commonly utilized for this purpose. The model is trained using the prepared dataset to learn patterns and correlations between NFT attributes and their respective prices.

3. Compute the Price Predictions: Once the model is trained, it can be employed to predict the prices of NFTs based on their attributes. By inputting the relevant features of an NFT, the AI model generates an approximate valuation for the asset.

It is worth noting that specific details of how each platform executes these steps are often proprietary information. As an example, we will outline our approach at TeaDAO:

1. Data Collection and Preparation: We acquire data from various NFT marketplaces and platforms, including transaction histories, token metadata, and sentiment analysis from social media. The data is then processed and cleaned to eliminate any outliers or inconsistencies that might skew the model's predictions.

2. AI Model Development: We employ state-of-the-art deep learning techniques to build a neural network model tailored to NFT price prediction. Our model undergoes extensive training on a large dataset, incorporating techniques like transfer learning and fine-tuning to enhance its predictive capabilities.

3. Price Prediction and Evaluation: With our AI model in place, we can compute price predictions for specific NFTs based on their characteristics. To evaluate the accuracy and effectiveness of our model, we employ cross-validation techniques and assess the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) between predicted and actual prices.

In the next part of our R&D, we aim to improve our AI-driven NFT valuation tool by enhancing the model's architecture, exploring additional data sources, and employing advanced natural language processing techniques to incorporate qualitative aspects affecting NFT prices.