Reproduce “EDGAR-CORPUS: Billions of Tokens Make The World Go Round” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Lefteris Loukas, Manos Fergadiotis, Ion Androutsopoulos, Prodromos Malakasiotis
Loukas et al. developed and released `edgar-corpus`, the largest publicly available financial NLP corpus comprising 6.5 billion tokens from SEC 10-K filings, alongside `edgar-crawler` for data acquisition and `edgar-w2v` domain-specific word embeddings. The `edgar-w2v` embeddings consistently outperformed generic and other financial embeddings across hypernym classification, financial tagging, and sentiment analysis tasks, improving accuracy by up to 2.6 percentage points and R² by 0.057.
Key insights
- Large-scale, meticulously pre-processed, and domain-specific corpora are critical for developing high-performing NLP models capable of capturing the unique semantics in specialized fields like finance.
- Careful data processing, including the removal of tables and precise itemization, along with avoiding aggressive techniques like stemming, is essential for preserving the fine-grained semantic distinctions in financial text.
- Providing open-source tools and comprehensive datasets significantly democratizes research, lowering the barrier to entry for academic and smaller research groups in complex, data-intensive domains such as financial NLP.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.