tahamajs/bitcoin_tweets_news_market_data_combined_v1

Bitcoin Tweets, News, and Market Data Dataset (2015–2021)

This dataset combines Bitcoin tweets, Bitcoin news, and Bitcoin market data to create a rich, multi-source dataset for training machine learning models aimed at predicting Bitcoin's future price movements.

Dataset Overview

This dataset consists of the following components:

• Bitcoin Tweets: 16 million tweets labeled with sentiment tags (positive, negative, or neutral) from 2015 to 2021.
• Bitcoin News: Relevant Bitcoin-related news articles spanning 2015–2021.
• Market Data: Historical market data for Bitcoin, including technical indicators (RSI, EMA, Volume, etc.) from 2015 to 2021.

The dataset has been enhanced to generate prediction prompts using market data, sentiment analysis from tweets, and Bitcoin news headlines, forming the basis for machine learning training tasks, especially for predicting Bitcoin's closing prices.

---

Dataset Components

The dataset includes the following columns:

1. Bitcoin Tweets (from Kaggle dataset gauravduttakiit/bitcoin-tweets-16m-tweets-with-sentiment-tagged):

• Date: The date when the tweet was posted.
• text: The tweet content.
• Sentiment: Sentiment of the tweet (Positive, Negative, or Neutral).

2. Bitcoin News (from Hugging Face dataset edaschau/bitcoin_news):

• news_date: The date when the news article was published.
• title: The headline of the news article.
• article_text: The full text of the article.

3. Bitcoin Market Data (from Yahoo Finance and custom calculations):

• BTC_Close: Bitcoin's closing price for that day.
• BTC_Volume: Bitcoin's trading volume for that day.
• BTC_EMA_50: 50-day Exponential Moving Average (EMA) of Bitcoin's closing price.
• BTC_EMA_200: 200-day Exponential Moving Average (EMA) of Bitcoin's closing price.
• BTC_RSI_14: 14-day Relative Strength Index (RSI) for Bitcoin's price.
• SP500_Close: S&P 500 index closing price.
• Gold_Close: Gold's closing price.
• Oil_Close: Oil's closing price.
• USD_Index_Close: US Dollar Index closing value.

---

Data Processing

1. Tweets:

   • 10 tweets are sampled per day to ensure daily representation of sentiment related to Bitcoin.

2. News:

   • Bitcoin news articles are aligned with the corresponding days in the tweet dataset.

3. Market Data:

   • Technical Indicators like EMA (50 and 200), RSI (14), and other related market data are calculated using historical data from Bitcoin (BTC-USD), Gold (GC=F), Oil (CL=F), S&P 500 (^GSPC), and US Dollar Index (DX-Y.NYB).

---

Data Generation & Structure

For each day:

• The historical 60 days of Bitcoin data (closing prices, volume, technical indicators).
• The Bitcoin news headlines (up to 10 articles).
• Sampled tweets (up to 10 tweets) are included.

This data is formatted into prompts for fine-tuning models, such that:

• Input: The prompt includes historical market data, news articles, and sampled tweets.
• Output: The prediction is the next day's Bitcoin closing price.

---

Usage Instructions

1. Install Dependencies:

Before running the scripts, install the required libraries:

pip install kagglehub yfinance pandas scikit-learn datasets

2. Download the Dataset:

The Bitcoin tweets dataset can be downloaded using KaggleHub:

import kagglehub

# Download the Bitcoin tweet dataset
path = kagglehub.dataset_download("gauravduttakiit/bitcoin-tweets-16m-tweets-with-sentiment-tagged")
print("Path to dataset files:", path)

3. Load and Process the Data:

The data can be processed by running the code to combine the datasets, calculate technical indicators, and generate fine-tuning prompts.

4. Example Usage:

Once the dataset is processed, you can split it into training and validation datasets. Here is an example of how to load and split the data:

from sklearn.model_selection import train_test_split
import json

# Load the generated JSON data (after running the processing code)
with open('bitcoin_finetuning_data_with_news.json', 'r') as f:
    full_dataset = json.load(f)

# Split the dataset into training and validation sets
train_data, val_data = train_test_split(full_dataset, test_size=0.1, random_state=42)

# Save the datasets
with open('train_dataset_with_news.json', 'w') as f:
    json.dump(train_data, f, indent=4)

with open('val_dataset_with_news.json', 'w') as f:
    json.dump(val_data, f, indent=4)

---

Dataset Usage Example

For fine-tuning models, here's an example prompt generated from the data:

Prompt:

Historical Prices (Last 60 days): 60000.12, 61000.23, 59500.45, ..., 63000.01
Technical Analysis for 2021-12-10: BTC Volume was 350000. The 14-day RSI is 68.32.
The price is above the 50-day EMA (61000.25) and above the 200-day EMA (58000.12).
Macro-Economic Context: S&P 500 closed at 4500.12; Gold at 1800.25; Oil at 70.25; US Dollar Index at 93.12.
Bitcoin News: Title: "Bitcoin hits all-time high" | Article: "Bitcoin has reached new heights, breaking the 60000 USD barrier..."
Sample Tweets: "Bitcoin hits new ATH || Bulls are back, price going up || Buy Bitcoin now || Hold BTC, it's going up || BTC news updates".

Output:

65000.00

---

Licensing and Usage

The dataset is provided under the MIT License.

• Bitcoin Tweets: Available under the Kaggle Terms of Service.
• Bitcoin News: Available under the Hugging Face License.
• Market Data: Data from Yahoo Finance is provided under the Yahoo Terms of Service.

---

Contributing

If you'd like to contribute to the dataset or improve the processing code, feel free to open an issue or submit a pull request.

---

Contact

For any questions or suggestions, please contact [Your Name or Organization Name].