Decoding Market Emotion from Blockchain Activity: A Data-Driven Sentiment Classifier

This study investigates whether blockchain activity and financial metrics can explain Bitcoin market sentiment, moving away from traditional price-forecasting objectives. The authors propose a data-driven framework that integrates on-chain data (such as transaction counts, fees, and network difficulty) with daily sentiment labels derived from Twitter. By treating on-chain activity as an explanatory variable for market mood, the researchers aim to provide a transparent, interpretable indicator of whether the market is in a bullish or bearish phase.

The research pipeline involves three primary stages: data collection, feature engineering, and model training. The authors collected historical Bitcoin network data from the Blockchair platform and merged it with daily sentiment classifications from a Kaggle-hosted Twitter dataset. To ensure data quality, the researchers simplified the sentiment labels into a binary classification (positive vs. negative), discarding neutral sentiments to reduce noise. The data were normalized using a block-based MinMax Scaler to ensure consistency across variables. Several machine learning models were tested using 5-fold cross-validation, with XGBoost selected as the most reliable classifier. The study also employed SHAP (SHapley Additive exPlanations) to quantify the influence of specific on-chain features on the model's predictions.

The XGBoost model achieved an average F1-score of 0.84, suggesting that blockchain activity contains meaningful signals that correlate with public sentiment. By using SHAP values, the authors provide a level of model interpretability that is often missing in complex deep learning approaches. This work is significant because it offers a simplified, robust alternative for market analysis that does not require real-time access to social media APIs or heavy natural language processing, potentially aiding investors in making more informed, data-backed decisions.