High-Quality Financial News Dataset

Description

This repository contains a meticulously scraped dataset from various financial websites. The data extraction process ensures high-quality and accurate text, including content from both the websites and their embedded PDFs.

Dataset Features

• Date: The date of the announcement.
• Subject: The subject of the financial news.
• Content: The full content of the announcement, including text from the website and PDFs.

Additional Processed Fields

We applied the advanced Mixtral 7X8 model to generate the following additional fields:

• ParaphrasedSubject: A paraphrased version of the original subject.
• CompactedSummary: A concise summary limited to 1.5 lines.
• DetailedSummary: A detailed summary of the content.
• Impact: The impact of the announcement, summarized in 2 lines.

Methodology

The prompt used to generate the additional fields was highly effective, thanks to extensive discussions and collaboration with the Mistral AI team. This ensures that the dataset provides valuable insights and is ready for further analysis and model training.

Usage

This dataset can be used for various applications, including but not limited to:

• Financial news analysis
• Abstractive/Exctractive Summarization tasks
• Machine learning model training
• Natural language processing tasks

Dataset

This dataset was created by Piyush Sonawane

Released under Other (specified in description)

Contents

Finance

## Overview

Finance is a dataset for object detection tasks - it contains Objects annotations for 342 images.

## Getting Started

You can download this dataset for use within your own projects, or fork it into a workspace on Roboflow to create your own model.

  ## License

  This dataset is available under the [CC BY 4.0 license](https://creativecommons.org/licenses/CC BY 4.0).

Dataset Card for "lmsys-finance"

This dataset is a curated version of the lmsys-chat-1m dataset, focusing solely on finance-related conversations. The refinement process encompassed:

Removing non-English conversations. Selecting conversations from models: "vicuna-33b", "wizardlm-13b", "gpt-4", "gpt-3.5-turbo", "claude-2", "palm-2", and "claude-instant-1". Excluding conversations with responses under 30 characters. Using 100 financial keywords, choosing conversations with at least… See the full description on the dataset page: https://huggingface.co/datasets/amphora/lmsys-finance.

Quantitative Finance Fine-Tuning Dataset

A dataset of 24 Q&A examples designed to fine-tune large language models (LLMs) for quantitative finance.

  📂 Categories

Category Topics Examples

Volatility Models SABR (corrected), Bergomi, rBergomi, Heston 5

Derivatives Pricing Dupire, VIX, Black-Scholes Greeks, CVaR 5

Interest Rates & Credit HJM, Hull-White, Merton, CDS 4

Numerical Methods Crank-Nicolson, Monte Carlo, FFT, LSM 5

Quant Strategies Momentum, Pairs… See the full description on the dataset page: https://huggingface.co/datasets/mo35/quant-finance-dataset.

Finance Image Text Data Notes

  Dataset summary

A documented Finance data-preparation workflow for Image Text records. The bundled rows demonstrate the schema and validation path rather than pretending to be a full training corpus.

  Included material

load_data.py — loading, cleaning, and split preparation code. dataset_infos.json — schema and split metadata. metadata_sample.jsonl — small, human-readable records for checking the schema. README.md —… See the full description on the dataset page: https://huggingface.co/datasets/SBYNUGROHO/finance-dataset.

All financial transactions made by the Intellectual Property Office as part of the Government’s commitment to transparency in expenditure

Overview

This dataset contains a comprehensive collection of approximately 42,000 articles sourced from the English Wikipedia, specifically curated to cover topics in finance, economics, business, and related fields. It provides a rich, high-quality text corpus for researchers, data scientists, and developers working on Natural Language Processing (NLP) tasks within the financial domain.

The challenge with general-purpose text corpora (like the full Wikipedia dump) is that they are noisy and lack domain-specific focus. This dataset solves that problem by providing pre-filtered, structured articles, ensuring that the content is highly relevant to the world of finance. Each entry includes the full article text, its title, associated categories, and useful metadata like word and character counts.

Data Structure

The dataset is provided in a structured format (e.g., JSONL or CSV). Each record represents a single Wikipedia article and contains the following fields, as seen in the example:

• Title: (string) The title of the Wikipedia article. Example: "IJM Corporation"
• Categories: (list of strings) A list of Wikipedia categories associated with the article. This is extremely * useful for filtering, topic modeling, and classification tasks.
• word_count: (integer) The total number of words in the text_content. Example: 1513
• text_length: (integer) The total number of characters in the text_content. Example: 10697
• text_content: (string) The full, plain-text content of the Wikipedia article, cleaned of templates and irrelevant HTML.

Potential Use Cases / Inspiration

This domain-specific corpus is ideal for a wide range of tasks. Here are a few ideas to get you started:

• Domain-Specific Language Modeling: Fine-tune pre-trained language models like BERT, GPT-2, or RoBERTa on this corpus to create a powerful "FinanceBERT" that understands the nuances of financial terminology.
• Named Entity Recognition (NER): Train models to identify and extract entities like company names, financial instruments, stock tickers, and key personnel.
• Text Classification & Topic Modeling: Classify articles into sub-topics (e.g., "Corporate Finance," "Macroeconomics," "FinTech") or use unsupervised methods like LDA to discover latent themes in financial discourse.
• Question Answering & Information Retrieval: Build a finance-specific search engine or a Q&A system that can answer questions like "What is quantitative easing?" or "What are IJM Corporation's main business segments?".
• Sentiment Analysis: Analyze the sentiment of articles about specific companies or market events, although Wikipedia's neutral point-of-view policy makes this a unique challenge. Summarization: Develop and test abstractive or extractive summarization models tailored for long-form financial documents.

Acknowledgements

This dataset was created using data from Wikipedia, the free encyclopedia. The content is available under the Creative Commons Attribution-ShareAlike 3.0 Unported License (CC BY-SA 3.0).

Please be sure to acknowledge Wikipedia as the source of the text content in any work that uses this dataset.

This dataset provides detailed, household-level records of income and expenses, including transaction categories, payment methods, recurrence patterns, and basic household demographics. It enables comprehensive budgeting analysis, supports financial literacy initiatives, and can power personalized financial recommendations and research into household spending habits.

Graph and download economic data for All Employees: Financial Activities: Finance and Insurance in Chicago-Naperville-Arlington Heights, IL (MD) (SMU17169745552000001SA) from Jan 1990 to Dec 2024 about Chicago, finance companies, companies, finance, insurance, IL, financial, employment, and USA.

AutoScientist Finance — Adapted Fine-tuning Dataset

Adapted training data (Adaption Labs AutoScientist pipeline, v5) used to fine-tune Rishidar/autoscientist-finance-qlora (Qwen2.5-0.5B-Instruct, QLoRA SFT + DPO).

finance_adapted.jsonl — SFT prompt/completion pairs (enhanced prompts + reasoning traces). finance_v5_raw.csv — full v5 output incl. chosen/rejected preference pairs used for DPO.

Also mirrored on Kaggle: rishidard/autoscientist-finance-dataset.

Personal Finance Management Software Market analysis indicates the market was valued at USD 131 Billion in 2025 and is anticipated to reach USD 245.9 Billion by 2035 with a CAGR of 6.5% over the forecast timeline.

A comprehensive list of the columns in your dataset, along with descriptions for each:

The campaign finance database is the San Francisco Ethics Commission's repository for campaign finance filings. It can answer questions about who is contributing money, who is receiving money, and how it is being spent. Use the campaign finance database to research campaign contributions and expenditures reported on forms provided by the Fair Political Practices Commission. The database provides live access to the Ethics Commission's records. Filings are accessible once processed/posted by the Ethics Commission.Forms filed with the Ethics Commission can be downloaded in PDF format. Forms filed electronically can be searched and exported in Microsoft Excel format. The following Excel exports are available:- Excel file based on a search of itemized transactions up to 2,000 rows (Updated immediately, with the exception of FPPC filing deadlines -- within 48 hours);- Excel file by year or for entire life of a single committee (Updated immediately upon filing submission); and- Excel file by year for all committees in a single calendar year (Updated every 24 hours).

visey07/finance dataset hosted on Hugging Face and contributed by the HF Datasets community

The OECD’s development finance statistics provide comprehensive data on the volume, origin and types of foreign aid and other resource flows from provider countries to partner countries.

Graph and download economic data for Job Openings: Finance and Insurance (JTU5200JOR) from Dec 2000 to Aug 2026 about job openings, vacancy, finance, insurance, and USA.

In 2023, global climate finance reached a record high of *** trillion U.S. dollars. However, this falls well short of the average annual climate finance needed from 2024 through 2030 under a 1.5 °C scenario. Despite the considerable increase in climate finance requirements, the amount needed is only a fraction of the economic losses due to climate change estimated under a business-as-usual scenario.

This dataset contains the predicted prices of the asset FINANCE over the next 16 years. This data is calculated initially using a default 5 percent annual growth rate, and after page load, it features a sliding scale component where the user can then further adjust the growth rate to their own positive or negative projections. The maximum positive adjustable growth rate is 100 percent, and the minimum adjustable growth rate is -100 percent.

A. SUMMARY This dataset contains data from financial statements of campaign committees that file with the San Francisco Ethics Commission and (1) contribute to or (2) receive funds from a San Francisco committee which was Primarily Formed for a local election, or (3) filed a Late Reporting Period statement with the SFEC. Financial statements are included for a committee if they meet any of the three criteria for each election included in the search parameters and are not primarily formed for the election.

The search period for financial statements begins two years before an election and runs through the next semi-annual filing deadline. The dataset currently filters by the elections of 2024-03-05 and 2024-11-05.

B. HOW THE DATASET IS CREATED During an election period an automated script runs nightly to examine filings by Primarily Formed San Francisco committees. If a primarily formed committee reports accepting money from or giving money to a second committee, that second committee's ID number is added to a filter list. If a committee electronically files a late reporting period form with the San Francisco Ethics Commission, the committee's ID number is also included in the filter list. The filter list is used in a second step that looks for filings by committees that file with the San Francisco Ethics Commission or the California Secretary of State.

This dataset shows the output of the second step for committees that file with the San Francisco Ethics Commission. The data comes from a nightly search of the Ethics Commission campaign database. A second dataset includes committees that file with the Secretary of State.

C. UPDATE PROCESS This dataset is rewritten nightly and is based on data derived from campaign filings. The update script runs automatically on a timer during the 90 days before an election. Refer to the "Data Last Updated" date in the section "About This Dataset" on the landing page to see when the script last ran successfully.

D. HOW TO USE THIS DATASET Transactions from all FPPC Form 460 schedules are presented together, refer to the Form Type to differentiate.

Transactions from FPPC Form 461 and Form 465 filings are presented together, refer to the Form Type to differentiate.

Transactions with a Form Type of D, E, F, G, H, F461P5, F465P3, F496, or F497P2 represent expenditures, or money spent by the committee. Transactions with Form Type A, B1, C, I, F496P3, and F497P1 represent receipts, or money taken in by the committee. Refer to the instructions for Forms 460, 496, and 497 for more details.

Transactions on Form 460 Schedules D, F, G, and H are also reported on Schedule E. When doing summary statistics use care not to double count expenditures.

Transactions from FPPC Form 496 and Form 497 filings are presented in this dataset. Transactions that were reported on these forms are also reported on the Form 460 at the next filing deadline. If a 460 filing deadline has passed and the committee has filed a campaign statement, transactions on 496/497 filings from the late reporting period should be disregarded.

This dataset only shows transactions from the most recent filing version. Committee amendments overwrite filings which come before in sequence.

Campaign Committees are required to file statements according to a schedule set out by the "https://www.fppc.ca.gov/learn/campaign-rules/where-and-when-to-file-campaign-statements/when-to-file-campaign-statements-state-local-filing-schedules.html">C