Finance Sentiment
Multilingual financial sentiment analysis
2 min readupdated
The model evaluates financial texts (headlines, earnings call quotes, market posts) and classifies their sentiment from an investor's perspective as positive, negative, or neutral. It features 308 million parameters, runs on CPU, and supports 15 languages, including Russian. The model labels match popular FinBERT (a model for financial text analysis) exactly, allowing drop-in replacement without changing code. The release is useful for developers of financial dashboards, analysts, and news researchers.
Key features
- Support for 15 languages: English, Russian, Chinese, German, and others.
- Compatible labels: uses standard
positive,negative, andneutralcategories. - Simple integration: loads via the standard
transformerslibrary pipeline without custom code. - Browser execution: ONNX weights available in full precision and int8 format for transformers.js.
- Apache 2.0 license: permits free commercial use.
What you can do
- Apply in bots and services for monitoring financial news and reports in Russian or other languages.
- Integrate into dashboards for automated sentiment scoring of market posts.
- Run the model locally on a regular CPU for fast text stream processing.
- Use in web applications via ONNX Runtime and transformers.js.
Example
from transformers import pipeline
classifier = pipeline("text-classification", model="empiriolabsai/finance-sentiment")
classifier([
"Quarterly revenue rose 12% and the company raised its full-year guidance.",
"Shares fell 8% after the company cut its dividend and warned on margins.",
"Die Bank meldete im dritten Quartal einen unerwarteten Verlust.",
])
# labels: positive, negative, negative
Read more
DetailsMacro-F1 on Russian translations of financial sentences: 0.900.
Suitable for CPU execution, weights optimized in int8 format.
Operates at the individual sentence level and does not account for sarcasm or multi-step reasoning.
Not an investment recommendation.










Comments