How to Use AI to Gain Ethereum Market Insights

In this article, we explore the great potential of Large Language Model (LLM)-based AI systems in their application to the dynamic Ethereum market analysis. For easy understanding, we use the AI ​​analysis on Ethereum Non-Fungible Token (NFT) market as an example. Overall, applying AI to gain Ethereum insights requires three steps. First, one must collect important on-chain Ethereum data and associated off-chain metadata. After this, one must compile a specialized database designed for the LLM. The final step is to apply the LLM-based recycling enhanced generation (RAG) approach to analyze data and gain insights.

Collect On-chain and Off-chain NFT Data

In the context of NFTs, the data collection necessitates diving into both the on-chain and off-chain dimensions. OpenSea, a leading NFT marketplace, provides fertile ground for the extraction of on-chain data, such as NFT transaction details and metadata. One can perform the data collection process via OpenSea’s API documentation, a simple approach to accessing on-chain data. At the same time, the off-chain data, such as NFT images or videos, are always stored in InterPlanetary File System (IPFS), a decentralized storage network. The initial step involves identifying the content-addressed IPFS hash, a unique identifier that indicates the NFT image within IPFS. This hash is typically in the NFT metadata or forms part of the transaction details. The next step involves building the HTTP gateway URL. Equipped with the IPFS hash, one can construct the URL, which then enables the sending of an HTTP request. Tools like Axios or the inherent fetch function serve as ideal tools to send an HTTP GET request to the constructed URL and thus retrieve the NFT image data.

Build an LLM knowledge database

Equipped with the LLM with appropriate data for efficient operation, it is essential to establish a comprehensive knowledge database. This knowledge database will serve as a reliable resource for semantic search, enabling the identification and retrieval of the most relevant data. As a result, the LLM is provided with the correct context, facilitating the generation of precise outputs at your request. After obtaining the on- and off-chain data, a systematic cleaning and organization of the collected data is necessary. This process includes identifying relevant properties and features integral to NFT analysis, such as categories, design studios, intellectual property holders and sales history. Many approaches exist to extract these features from images, such as through the NFT metadata or free image recognition software. After successfully extracting and encoding the data features, we can quickly build a custom LLM database.

Use an LLM-based AI model to gain NFT insights

Relying solely on an LLM to generate factual text or even refine the model with your database may not necessarily produce a factually accurate answer. We therefore propose a RAG approach for NFT data analysis. RAG represents a methodology that separates the knowledge database from the language model. This methodology involves asking a question or submitting a request to the AI ​​agent, such as queries related to emerging NFT trends, separate properties, or correlations and relationships between property properties and NFT market performance. Next, a search algorithm, for example Azure Cognitive Search, examines the most pertinent text within the knowledge database, which is likely to contain the required answer. As a final step, a concise request, in the form of instructions to the LLM together with relevant document text, is addressed to the LLM. The model then uses this input to formulate a response that matches the original request, ensuring a fact-based, contextually relevant output.

AI will become a powerful tool in the blockchain industry

This article uses Ethereum NFT analysis as a representative case to demonstrate the effectiveness of using an LLM-based AI methodology for obtaining blockchain insights. In its versatility, this method holds potential for widespread application within the blockchain domain. We expect the future to see an influx of LLM tools to improve and facilitate various aspects of blockchain operation and analysis.

Follow me further Twitter or LinkedIn.