CryptoVision: A Comprehensive Dataset for Crypto News and Sentiment Analysis

This dataset contains 188,431 cryptocurrency news articles collected between 2017 and 2025 from several reputable cryptocurrency and blockchain news platforms, including BlockWorks, CoinDesk, Cointelegraph, CryptoPanic, CryptoNews, and Decrypt. The articles were collected using automated Python-based web scraping pipelines for research in cryptocurrency analytics, financial sentiment analysis, natural language processing (NLP), and machine learning. The raw dataset contains six primary attributes: URL, Title, Description, Date Time, Coin, and Type. These attributes provide information about the original news source, article content, publication time, and associated cryptocurrency category.

After preprocessing and cleaning, the final dataset size was reduced to 117,961 high-quality news articles by removing duplicate, incomplete, and noisy records. Several additional NLP-oriented columns were generated during preprocessing. The combined column was created by merging the title and description text into a single feature for textual analysis. The clean_text column contains moderately cleaned text prepared for Transformer-based and deep learning models such as BERT, FinBERT, LSTM, and GRU, where preserving contextual and semantic information is important. The traditional column contains heavily cleaned and normalized text specifically designed for classical machine learning algorithms such as Logistic Regression, Naïve Bayes, Support Vector Machine (SVM), and Random Forest. This column underwent extensive preprocessing, including lowercasing, punctuation removal, stopword elimination, special character filtering, stemming/normalization, and noise reduction to improve feature extraction performance. Finally, sentiment labels were generated using FinBERT, a transformer-based financial language model specialized for financial text analysis, which classified each article into positive, negative, or neutral sentiment categories.

The processed dataset is suitable for cryptocurrency market forecasting, sentiment-aware trading systems, financial NLP research, blockchain-related analytical studies, trend prediction, event-driven market analysis, portfolio risk assessment, and the development of machine learning and deep learning models for cryptocurrency sentiment classification and price movement prediction.