kurry/sp500_tone_dataset

SP500 Tone Dataset

Dataset summary

The SP500 Tone Dataset contains tone dispersion metrics extracted from earnings calls of S&P 500 companies. Each record represents a quarterly earnings call, capturing the variability in tone and sentiment expressed during the call. This dataset is useful for financial sentiment analysis, market prediction models, and studies on executive communication strategies.

Supported tasks

• Regression (predicting tone dispersion)

Languages

• English

Dataset structure

Field	Type	Description
symbol	string	Stock ticker symbol for the company
company_id	int64	Unique identifier for the company
year	int64	Year of the earnings call
quarter	int64	Quarter of the earnings call (1-4)
date	string	Date of the earnings call
tone_dispersion	float64	Measure of variability in tone during the call
call_key	string	Unique identifier for the specific earnings call

Splits

This dataset has a single split:

• train: All records of tone dispersion metrics

Usage

from datasets import load_dataset

dataset = load_dataset("kurrytran/sp500_tone_dataset")

data = dataset["train"]

# Example: Compute average tone dispersion by year
from collections import defaultdict
avg_tone_by_year = defaultdict(lambda: {"total": 0, "count": 0})
for item in data:
    year = item["year"]
    avg_tone_by_year[year]["total"] += item["tone_dispersion"]
    avg_tone_by_year[year]["count"] += 1

for year, values in sorted(avg_tone_by_year.items()):
    print(f"Year {year}: Average tone dispersion = {values['total'] / values['count']:.4f}")

Citation

If you use this dataset in your research, please cite it as follows:

@dataset{sp500_tone_dataset,
  author    = {kurrytran},
  title     = {SP500 Tone Dataset},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/kurrytran/sp500_tone_dataset}
}

License

This dataset is licensed under CC BY 4.0. For more details, see https://creativecommons.org/licenses/by/4.0/.