README.md

metadata

license: odc-by
task_categories:
  - text-classification
language:
  - en
tags:
  - tense-prediction
  - finance
  - nlp
size_categories:
  - 100K<n<1M

NOSIBLE Prediction Dataset

Changelog

• v1.0.0: Initial version

Who is NOSIBLE?

NOSIBLE is a vertical web-scale search engine. Our worldwide media surveillance products help companies build AI systems that see every worldwide event and act with complete situational awareness. In short, we help companies know everything, all the time. The financial institutions we work with rely on us to deliver media intelligence from every country in every language in real-time. Shortcomings in existing financial datasets and financial models are what inspired us to release this dataset and related models.

• NOSIBLE Prediction v1.1 Base

What is it?

The NOSIBLE Prediction Dataset is an open collection of 100,000 cleaned and deduplicated news samples curated to identify predictive language. Each sample is annotated with one of two labels: prediction, indicating that the text includes a prediction, estimate, or forecast, and not-prediction, indicating that it does not.

All text is sourced from the NOSIBLE Search Feeds product using a curated set of finance-related queries. Sentiment labels are assigned through a multi-stage, LLM-based annotation pipeline (described below).

How to use it

Using the HuggingFace datasets library:

Install the dataset library with pip install datasets, then load the dataset:

from datasets import load_dataset

dataset = load_dataset("NOSIBLE/prediction")
print(dataset)

Expected Output

DatasetDict({
    train: Dataset({
        features: ['text', 'label', 'netloc', 'url'],
        num_rows: 100000
    })
})

You can also access this dataset through any interface supported by Hugging Face.

Dataset Structure

Data Instances

The following is an example sample from the dataset:

{
    "text": "Offshore staff HOUSTON \u2013 VAALCO Energy is looking to bring in a floating storage and offloading (FSO) unit at the Etame Marin oil field offshore Gabon. The company has signed a non-binding letter of intent with Omni Offshore Terminals to supply and operate the vessel at Etame for up to 11 years, following the expiry of the current FPSO Petr\u00f3leo Nautipa contract with BW Offshore in September 2022. Omni has provided a preliminary proposal for leasing and operating the FSO, which could reduce VAALCO's operating costs by 15-25%, compared with the current FPSO contract during the term of the proposed agreement. Maintaining the current FPSO beyond its contract or transitioning to a different FPSO, VAALCO added, would require substantial capex investments.",
    "label": "prediction",
    "netloc": "www.offshore-mag.com",
    "url": "https://www.offshore-mag.com/rigs-vessels/article/14202267/vaalco-contemplating-switch-to-fso-at-etame-offshore-gabon"
}

Data Fields

• text (string): A text chunk from a search result.
• label (string): Whether the text contains a predition or not.
• netloc (string): The domain name of the source document.
• url (string): The URL of the document.

Dataset creation

Data source

The dataset was sampled from the NOSIBLE Search Feeds, which provides web-scale surveillance data to customers. Samples consist of top-ranked search results from the NOSIBLE search engine in response to safe, curated, and finance-specific queries. All data is sourced exclusively from the public web.

Relabeling algorithm

Labels were first generated using multiple LLM annotators and were then refined using an active-learning–based relabeling loop.

The algorithm outline is as follows:

1. Hand-label ~200 samples to tune prompts for the LLM annotators.
2. Label a set of 100k samples with LLM labelers:
   • xAI: Grok 4 Fast
   • xAI: Grok 4 Fast (reasoning enabled)
   • Google: Gemini 2.5 Flash
   • OpenAI: GPT-5 Nano
   • OpenAI: GPT-4.1 Mini
   • OpenAI: gpt-oss-120b
   • Meta: Llama 4 Maverick
   • Qwen: Qwen3 32B
3. Train multiple linear models to predict the majority-vote of the LLM labelers. The features are the text embeddings of the following models:
   • Qwen3-Embedding-8B
   • Qwen3-Embedding-4B
   • Qwen3-Embedding-0.6B
   • OpenAI: Text Embedding 3 Large
   • Google: Gemini Embedding 001
   • Mistral: Mistral Embed 2312
4. Perform iterative relabeling:
   • Compare all the linear models' predictions to the majority-vote label.
   • Identify disagreements where all linear models agree but the majority-vote label does not.
   • Use a larger LLM (“oracle”) to evaluate ambiguous cases and relabel when appropriate.
   • Drop the worst performing linear models from the ensemble.
   • Repeat until no additional samples require relabeling.
5. This is the final dataset used for training the NOSIBLE Prediction v1.1 Base model.

We used OpenAI: GPT-5.1 as the oracle.

Additional information

License

The dataset is released under the Open Data Commons Attribution License (ODC-By) v1.0 license.

Attribution

• NOSIBLE Inc Team

Contributors

This dataset was developed and maintained by the following team:

• Matthew Dicks
• Simon van Dyk
• Stuart Reid

Citations

Coming soon, we're working on a white paper.