ACL18
Goal
The ACL18 dataset is designed to advance stock movement prediction using social media and historical prices. Its primary goals are:
To predict stock price movements from tweets and historical data
To benchmark multimodal financial forecasting combining text and time series
To evaluate sentiment-driven trading signals
Description
ACL18 is a stock movement prediction dataset containing tweets and historical stock prices. Models must predict whether a stock’s price will rise or fall based on social media sentiment and price trends.
Dataset Composition
Source: Twitter posts and historical stock prices
Time Period: Historical trading data with corresponding tweets
Sample Size: 27,053 instances
Data Modalities: Text (tweets) and time series (stock prices)
Example
Stock: Apple Inc. (AAPL)
Historical Prices (5-day window): - Day -5: $120.50 → $119.80 (-0.58%) - Day -4: $119.80 → $121.30 (+1.25%) - Day -3: $121.30 → $120.90 (-0.33%) - Day -2: $120.90 → $122.10 (+0.99%) - Day -1: $122.10 → $121.75 (-0.29%)
Tweets: - “$AAPL new iPhone pre-orders exceed expectations! Strong demand across all models” - “Apple suppliers ramping up production, bullish signal for Q4” - “Analysts raising price targets on $AAPL after product launch”
Prediction Target: Rise or Fall (next day movement > ±0.5%)
Ground Truth: Rise (+1.2% next day)
Task Description
Stock Movement Prediction
Input: Historical stock prices (time series) and related tweets (text)
Output: Binary classification (Rise or Fall)
Challenge: Combining noisy social media signals with price patterns to predict short-term movements
Key Use Cases
Algorithmic Trading: Short-term trading signals from social sentiment
Market Sentiment Analysis: Tracking investor sentiment on social media
Risk Management: Early warning signals from social media discussions
Evaluation Metrics
Accuracy (ACC): Percentage of correct predictions
Matthews Correlation Coefficient (MCC): Balanced measure for binary classification
References
Yumo Xu and Shay B. Cohen. “Stock Movement Prediction from Tweets and Historical Prices.” In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1970–1979, 2018.
For dataset access, visit HuggingFace.