You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

ZettaQuant Financial Sentiment Analysis Model (`financial-sentiment-general`)

Fine-tuned version of FinRoBERTa for financial sentiment analysis — classifying financial text as positive, negative, or neutral.

What is `financial-sentiment-general`?

financial-sentiment-general is a financial text classification model built on a FinRoBERTa continual pre-trained version of RoBERTa, trained on large financial text and adapted for the language patterns of financial content. It identifies the sentiment expressed in financial text and assigns one of three labels: positive, negative, or neutral.

Financial language does not behave like everyday language. A phrase like "strong headwinds" is negative. "Missed estimates by a slim margin" is negative. "Revenue was flat" in a growth-expected context is negative. General-purpose sentiment models — trained on product reviews, social media, or news — routinely misread these signals because they are not trained on financial vocabulary or context.

At ZettaQuant, we specifically train models for different data and use cases, and those models are available at the appropriate deployment or access channel. For enterprise customers, we also train custom models on their data for enhanced performance.

This model is built for developers, data scientists, quantitative researchers, and financial teams who need reliable sentiment analysis for finance as a component in larger NLP pipelines.

Use Cases for Financial Sentiment Analysis

Financial News Monitoring Classify headlines and articles from financial news sources to track shifts in market sentiment around specific companies, sectors, or macroeconomic events.

Market Commentary Analysis Process daily or weekly market commentary — analyst notes, investment memos, and strategy reports — to surface sentiment signals at scale without manual review.

Corporate Disclosure Processing Analyze language in corporate communications — press releases, earnings announcements, and official statements — to detect sentiment that may affect investor perception.

NLP Pipeline Component Use as a classification layer inside larger financial NLP workflows, including retrieval-augmented generation (RAG) systems, alert pipelines, and data enrichment pipelines.

Key Capabilities

Financial domain adaptation — built for financial terminology and context, not repurposed from generic text
FinRoBERTa backbone — contextual understanding via the FinRoBERTa transformer architecture
Three-class output — returns positive, negative, or neutral
Transformers-compatible — works directly with the Hugging Face pipeline API for enterprise licensed clients only
API access available — use via the ZettaQuant Platform with no model download required
Safetensors format — fast, safe weight loading for production environments

How to Use the Financial Sentiment Model

Option 1 — ZettaQuant Platform API (No Download Required)

curl -X POST \
  -H "x-api-key: $ZQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model_id": "fpb_sentiment_label", "instances": [{"text": "Your text here"}]}' \
  $ZQ_BASE_URL/v1/models/infer

Full API reference: ZettaQuant Platform Documentation

Option 2 — Hugging Face Transformers (Enterprise Licensed Clients Only)

from transformers import pipeline

# Load the financial sentiment analysis pipeline
classifier = pipeline(
    "text-classification",
    model="zettaquant/financial-sentiment-general"
)

# Single input
result = classifier("The company reported stronger-than-expected earnings this quarter.")
print(result)

# Batch input
texts = [
    "The company reported stronger-than-expected earnings this quarter.",
    "Management issued a warning citing significant revenue headwinds.",
    "The board approved the proposed merger on Tuesday."
]

results = classifier(texts)
for text, result in zip(texts, results):
    print(f"{result['label']} — {text}")

Expected Inputs and Outputs

Input: A string of financial text in English — a sentence, headline, or short paragraph.

Output: A sentiment label.

Label	Meaning
`positive`	The text indicates favorable or beneficial developments for a company or investors.
`neutral`	The text is factual or balanced and does not signal a clearly positive or negative outcome.
`negative`	The text indicates unfavorable or harmful developments for a company or investors.

Example output:

[{"label": "positive"}]

Access Policy & Production API

Gated Model

This model is gated. Access to model weights requires a review by the ZettaQuant team. The repository is publicly listed, but files and weights are not available for general download from Hugging Face.

To request access:

Click Request Access at the top of this page.
Fill in your name, organization, and title.
The ZettaQuant team will review and respond to your request.

Production API (No Download Required)

This model is available via the ZettaQuant Platform API. No GPU, no weight download, no infrastructure setup required.

Detail	Value
Model ID	`fpb_sentiment_label`
Task	Sequence Classification
Tier	Free
Documentation	docs.zettaquant.ai

Disclosures

This model is designed for English-language financial text only. Performance on other languages is not supported.
Inputs should be sentence or short paragraph length. Very long documents should be chunked before inference.
Like all domain-adapted models, performance may vary on text that is structurally different from typical financial content (e.g., informal social media posts, highly technical regulatory language).
Model output should be treated as one input in a broader analysis process and not as the sole basis for financial decisions.

Frequently Asked Questions

What output labels does this model return? The model returns one of three labels — positive, negative, or neutral. Positive indicates favorable developments for a company or investors. Negative indicates unfavorable developments. Neutral indicates factual or balanced content without a clear directional signal.

How is this different from a general-purpose sentiment model? General-purpose models are trained on everyday text such as product reviews and social media posts. They frequently misclassify financial language because the vocabulary and context are domain-specific. For example, phrases like "missed estimates," "margin compression," or "guidance cut" are clearly negative in finance but may read as neutral to a general model. This model is adapted specifically for financial text.

What kinds of financial text is this model designed for? The model is designed for financial text in English, including market commentary, financial news, and corporate disclosures such as press releases and earnings announcements.

Can I use this model without downloading the weights? Yes. The ZettaQuant Platform provides a free-tier REST API for this model. You can call it via a standard HTTP POST request using your API key and base URL. See the API documentation for full details.

Can this model be used for commercial purposes? Yes, the ZettaQuant API can be used for commercial purposes, subject to ZettaQuant terms and access approval.

What language does this model support? English only. The model is not designed or tested for multilingual financial text.

How do I get access to the model weights? Click Request Access at the top of this page. Fill in your name, organization name, and title. The ZettaQuant team reviews access requests and will respond with next steps.

Is the ZettaQuant Platform API free to use? Yes. This model is available on the Free tier of the ZettaQuant Platform. See the platform documentation for rate limits and usage details.

What is the model ID for the ZettaQuant API? The model ID for use with the ZettaQuant Platform API is fpb_sentiment_label.

About ZettaQuant

ZettaQuant builds AI-powered financial intelligence tools that help convert unstructured financial text into actionable quantitative insights.

Website: www.zettaquant.ai
Platform Docs: docs.zettaquant.ai
Hugging Face Org: huggingface.co/zettaquant

Downloads last month: 2

Safetensors

Model size

0.1B params

Tensor type

F32