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

  1. Market Sentiment Analysis: Tracking investor sentiment from news and social media

  2. Trading Signals: Using sentiment as input for algorithmic trading strategies

  3. Risk Monitoring: Identifying negative sentiment around specific assets

Evaluation Metrics

  1. F1-Score: Macro-averaged F1 for sentiment classification

  2. 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.