FiQA-SA (Sentiment Analysis)
Description
FiQA-SA is a sentiment analysis dataset for financial question answering, containing financial news headlines and microblog posts. The dataset evaluates sentiment on a continuous scale from -1 (most negative) to 1 (most positive).
Dataset Composition
Source: Financial news headlines and social media posts
Sample Size: 1,173 annotated posts
Sentiment Scale: Continuous values from -1 (negative) to 1 (positive)
Example
Text 1: “Goldman Sachs upgraded the stock to buy with a price target of $150, citing strong fundamentals”
Sentiment Score: 0.8 (positive)
Text 2: “Company faces potential bankruptcy as debt obligations exceed available cash reserves”
Sentiment Score: -0.9 (very negative)
Text 3: “The Federal Reserve maintained interest rates at current levels”
Sentiment Score: 0.0 (neutral)
Task Description
Sentiment Analysis for Financial Texts
Input: Financial news headlines or microblog posts
Output: Sentiment score on a scale of [-1, 1]
Challenge: Requires understanding nuanced financial sentiment and context from both formal news and informal social media
Key Use Cases
Market Sentiment Analysis: Tracking investor sentiment from news and social media
Trading Signals: Using sentiment as input for algorithmic trading strategies
Risk Monitoring: Identifying negative sentiment around specific assets
Evaluation Metrics
F1-Score: Macro-averaged F1 for sentiment classification
Accuracy: Classification accuracy when sentiment is discretized
References
Macedo Maia, Siegfried Handschuh, André Freitas, Brian Davis, Ross McDermott, Manel Zarrouk, and Alexandra Balahur. “WWW’18 Open Challenge: Financial Opinion Mining and Question Answering.” In Companion Proceedings of the The Web Conference 2018 (WWW ‘18), 2018.
For dataset access, visit the official challenge page or HuggingFace.