CIKM18
Goal
The CIKM18 dataset is designed to advance stock movement prediction using hybrid deep sequential modeling. Its primary goals are:
To predict stock price movements from social text and historical prices
To benchmark sequential modeling approaches for financial forecasting
To evaluate deep learning methods on financial time series
Description
CIKM18 is a stock movement prediction dataset that combines social text-driven signals with deep sequential modeling of historical prices.
Dataset Composition
Source: Social media posts (tweets) and historical stock prices
Time Period: Historical trading data with social media activity
Sample Size: 4,967 instances
Data Modalities: Text (social media) and time series (stock prices)
Example
Stock: Amazon.com Inc. (AMZN)
Historical Prices (sequential 5-day pattern): - Day -5: Close $3,200 (volume: 4.2M) - Day -4: Close $3,185 (volume: 3.8M, -0.47%) - Day -3: Close $3,210 (volume: 5.1M, +0.79%) - Day -2: Close $3,195 (volume: 4.5M, -0.47%) - Day -1: Close $3,220 (volume: 6.2M, +0.78%)
Social Text: - “Amazon Prime Day sales hit record highs, expect strong Q2 earnings” - “$AMZN cloud services AWS gaining market share from competitors” - “Retail analysts upgrading Amazon on strong e-commerce growth”
Sequential Pattern: Increasing volume with positive price momentum
Prediction Target: Rise or Fall (movement > ±0.5%)
Ground Truth: Rise (+1.5% following continued momentum)
Task Description
Stock Movement Prediction with Sequential Modeling
Input: Social media text and sequential historical price data
Output: Binary classification (Rise or Fall)
Challenge: Capturing sequential dependencies in price movements while integrating social media sentiment signals
Key Use Cases
Quantitative Trading: Sequential pattern recognition for trading strategies
Market Prediction: Combining technical analysis with sentiment analysis
Investment Research: Understanding price momentum influenced by social sentiment
Evaluation Metrics
Accuracy (ACC): Percentage of correct predictions
Matthews Correlation Coefficient (MCC): Correlation between predictions and actual outcomes
References
Huizhe Wu, Wei Zhang, Weiwei Shen, and Jun Wang. “Hybrid Deep Sequential Modeling for Social Text-Driven Stock Prediction.” In Proceedings of the 27th ACM International Conference on Information and Knowledge Management (CIKM), pages 1627–1630, 2018.
For dataset access, visit HuggingFace.