BigData22
Goal
The BigData22 dataset is designed to advance stock movement prediction with self-supervised learning from sparse noisy tweets. Its primary goals are:
To predict stock price movements from noisy social media data
To handle sparse and unreliable social media signals
To benchmark robust forecasting methods for financial markets
Description
BigData22 is a stock movement prediction dataset focusing on learning from sparse and noisy tweets combined with historical stock prices.
Dataset Composition
Source: Twitter posts and historical stock prices
Time Period: Recent trading periods with social media activity
Sample Size: 7,164 instances
Data Modalities: Text (tweets) and time series (stock prices)
Example
Stock: Tesla Inc. (TSLA)
Historical Prices (3-day window): - Day -3: Open $850.00, Close $845.00 (-0.59%) - Day -2: Open $847.00, Close $862.00 (+1.77%) - Day -1: Open $863.00, Close $858.00 (-0.58%)
Sparse Tweets (noisy data): - “$TSLA to the moon 🚀🚀” (3 likes, low credibility) - “Elon musk factory tour tomorrow” (15 likes) - “$TSLA delivery numbers looking strong this quarter - analyst” (250 likes, higher credibility)
Prediction Target: Rise or Fall (movement > ±0.55%)
Ground Truth: Rise (+2.1% next day based on actual delivery numbers)
Task Description
Stock Movement Prediction from Noisy Data
Input: Sparse, noisy tweets and historical stock prices
Output: Binary classification (Rise or Fall)
Challenge: Filtering noise from sparse social media data while extracting meaningful signals for prediction
Key Use Cases
Retail Trading Signals: Understanding retail investor sentiment from social media
Noise Filtering: Developing robust models that handle unreliable data sources
Social Media Analytics: Measuring impact of social chatter on stock prices
Evaluation Metrics
Accuracy (ACC): Percentage of correct predictions
Matthews Correlation Coefficient (MCC): Balanced measure accounting for class imbalance
References
Yejun Soun, Jaemin Yoo, Minyong Cho, Jihyeong Jeon, and U Kang. “Accurate Stock Movement Prediction with Self-supervised Learning from Sparse Noisy Tweets.” In 2022 IEEE International Conference on Big Data (Big Data), pages 1691–1700, 2022.
For dataset access, visit HuggingFace.