tahamajs/bitcoin-daily-raw-news-tweets-financials

Daily Bitcoin Multimodal & LLM-Augmented Financial Dataset

Dataset Description

This is a rich, time-series dataset designed for multimodal analysis and forecasting of the Bitcoin market. It aggregates a wide array of daily data from early 2015 to the end of 2022, with each row representing a single day.

The dataset combines several data dimensions:

• —Textual Data: Raw text from news articles, social media (tweets and Reddit), providing daily public discourse and sentiment.
• —Financial Market Data: Daily closing prices for Bitcoin, Gold, and Oil to capture macroeconomic context.
• —On-Chain Metrics: Fundamental indicators of the Bitcoin network's health, security, and usage, such as hash rate, transaction volume, and active addresses.
• —Social & AI-Generated Sentiment: High-level sentiment indicators like the Fear & Greed Index, complemented by sentiment and trading action classifications from a Large Language Model (LLM).

The dataset is structured with historical features (e.g., the last 60 days of prices) and a future prediction target (the next 10 days of prices), making it ideal for developing and testing sophisticated forecasting models.

Data Fields

The dataset includes the following columns:

Time & Identifier

• —date (date): The specific date for the data record in YYYY-MM-DD format. This is the primary key for the time series.

Textual Features

• —daily_news_full_text (list of strings): A collection of up to 30 raw, full-text news articles related to Bitcoin published on the given date.
• —daily_tweets (list of strings): A collection of up to 30 raw, full-text tweets mentioning Bitcoin posted on the given date.
• —contextual_past_article (string): The complete text of a single, randomly selected news article from the 60 days prior to the date, intended to provide historical context.
• —text_cointelegraph (string): Aggregated daily news text from Cointelegraph.
• —text_bitcoin_news (string): Aggregated daily news text from news.bitcoin.com.
• —text_reddit_posts (string): Aggregated daily posts from Bitcoin-related subreddits.

Financial & Commodity Features

• —gold_close_price (float): The closing price of Gold futures (GC=F) for the day.
• —oil_close_price (float): The closing price of WTI Crude Oil futures (CL=F) for the day.
• —btc_price_history_60d (list of floats): A list containing the daily closing prices of Bitcoin (BTC-USD) for the 60 days leading up to the date. This serves as the primary historical feature set for time-series models.
• —btc_price_target_10d (list of floats): A list containing the daily closing prices of Bitcoin (BTC-USD) for the 10 days following the date. This is the intended prediction target for forecasting models.

On-Chain & Network Metrics

• —btc_market_cap (float): The total market capitalization of Bitcoin in USD.
• —btc_total_supply (float): The total number of bitcoins in circulation.
• —btc_hash_rate (float): The estimated number of tera-hashes per second (TH/s) the Bitcoin network is performing. A key measure of network security.
• —btc_difficulty (float): A relative measure of how difficult it is to mine a new block. This adjusts to keep block production time stable.
• —btc_transactions (float): The total number of confirmed transactions on the given day. A measure of network activity.
• —btc_unique_addresses (float): The number of unique addresses that were active on the network that day, as either a sender or receiver.
• —btc_estimated_tx_volume_usd (float): The estimated value of on-chain transactions in USD.

Social & LLM-Generated Metrics

• —fear_and_greed_index (float): The numerical value (0-100) of the popular crypto Fear & Greed Index, where 0 is "Extreme Fear" and 100 is "Extreme Greed".
• —cbbi_index (float): The Colin Talks Crypto Bitcoin Bull Run Index, a metric to gauge Bitcoin's position in a macro cycle.
• —llm_sentiment_class (string): The overall daily sentiment (Positive, Negative, Neutral) as determined by an LLM, providing an AI-based sentiment score.

Data Curation

• —Data Sources: The dataset was created by aggregating data from multiple sources:
• —Prices & Commodities: Yahoo Finance (yfinance).
• —On-Chain, Social, LLM, and Additional Text: danilocorsi/LLMs-Sentiment-Augmented-Bitcoin-Dataset.
• —Primary News Source: edaschau/bitcoin_news.
• —Primary Tweet Source: A user-provided mbsa.csv file.
• —Missing Data: Gaps in daily numerical data (due to market closures, reporting lags, etc.) have been filled using a forward-fill (ffill) strategy to ensure continuity.
• —Raw Text: All text fields contain unprocessed data and have not been summarized. They may require cleaning, preprocessing, or embedding for use in models.