EDTSUM

Goal

The EDTSUM dataset is designed to advance abstractive summarization of financial news articles. Its primary goals are:

  • To enable headline generation from financial news

  • To support event detection for news-based trading

  • To benchmark news summarization in finance

Description

EDTSUM is a dataset for abstractive summarization of financial news articles, containing news articles paired with their headlines as ground-truth summaries.

Dataset Composition

  • Source: Financial news articles

  • Time Period: Various timeframes covering financial events

  • Sample Size: 2,000 news articles with headlines

  • Content Focus: Corporate events and financial news

Example

News Article:

“Apple Inc. reported quarterly earnings that exceeded Wall Street expectations, with revenue reaching $89.5 billion in the fiscal third quarter. The company’s iPhone sales showed resilience despite supply chain challenges, contributing to a 15% year-over-year growth. CEO Tim Cook highlighted the strong performance of the services division, which grew 27% from the previous year. The tech giant also announced a $90 billion share buyback program, signaling confidence in future growth.”

Generated Headline:

“Apple Beats Earnings Expectations with $89.5B Revenue, Announces $90B Buyback”

Task Description

Abstractive News Summarization

  • Input: Full financial news article

  • Output: Generated headline/summary capturing the essence of the article

  • Challenge: Generating concise, informative summaries that capture key financial events while maintaining factual accuracy

Key Use Cases

  1. Event-Driven Trading: Identifying trading signals from news events

  2. Financial Monitoring: Tracking corporate events and announcements

  3. News Aggregation: Creating concise financial news summaries

Evaluation Metrics

  1. ROUGE Scores: ROUGE-1, ROUGE-2, ROUGE-L for lexical overlap

  2. BERTScore: Semantic similarity measurement

References

Zhihan Zhou, Liqian Ma, and Han Liu. “Trade the Event: Corporate Events Detection for News-Based Event-Driven Trading.” In Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021, pages 2114–2124, 2021.

For dataset access, visit HuggingFace.