An ABI-based Classification Approach for Ethereum Smart Contracts

Abstract

Due to the rapid increasing of smart contracts in Ethereum, retrieving the related smart contracts has become an important issue. A key step of recognizing the related smart contracts is to classify them. Nowadays most of the classification methods of smart contracts are based on source code or bytecode analysis. However, as the interface of a smart contract, ABI has key information about the smart contract's function and behavior. Aiming to find out the effect of ABI on the classification of smart contracts, we propose an automatic classification method that utilize popular text classification method on ABI granularity. Our experiments are based on basic machine learning algorithms - SVM, XGBoost, Stacking and Bagging, which take TF-IDF vectors of ABIs as input, and the results indicate that ABI-based smart contracts classification is an effective approach. Compared with source code-based classification method, the MCC score of our method increased by more than 0.1 on average.