Academic paper fetcher and summariser for Fintech related topics powered by BART (HuggingFace Transformers library).
📘 Access the FinAcNews Notebook:
Open Version 1.5 - updated list of reviews and websites
Open Version 1.4
Requirements: Check requirements.txt
FinAcNews is a modular Python pipeline that fetches, filters, and summarises academic FinTech research from leading journals and archives.
Designed for analysts, researchers, and enthusiasts, it transforms dense abstracts into digestible insights—complete with keyword highlighting, interactive tables, and full-source links.
Not working in FinTech? No problem! You can easily adapt this script for your own field by:
- Defining your own key-word list
- Replacing the media links by others related to your field (just make sure to get the link of a valid RSS feed from their website)
Here is how FinAcNews works:
- Step 1: Configuration
In this part, the script defines customisable aspects of the research, such as the number of days of articles it should fetch and the keywords used for filtering articles.
-
Step 2: Summariser setup
After checking whether the system has a d-GPU, this part defines the function that will be used to summarise the articles using the BART model. Keep in mind: the script doesn't scrape anything, it just fetches the article academic summary and summarises it to make it more accessible. As a result, only the academic summary - and not the complete article- are summarised. However, this is more than often enough to give an overview to the user and let him decide wether or not he should check out the full article.This project uses the Transformers library by Hugging Face to perform text summarization via the facebook/bart-large-cnn model. The model is licensed under the MIT License, and the implementation is made possible thanks to the Hugging Face team and contributors.
Model reference: Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A., Levy, O., Stoyanov, V., & Zettlemoyer, L. (2020). BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. arXiv:1910.13461
- Step 3: Helpers
A series of functions that:- Make sure the articles match the specified time window
- Highlight in the summary the keywords having triggered the keyword-filter, mainly for debugging but also to make more visible the topics.
- A way to extend the keyword list (eg: crypto -> cryptocurrency, cryptography, cryptos,...)
- Adapt titles/ text to make them more readable on display (2 functions)
- Filter the articles with the series of keywords prepared beforehand.
-
Step 4: Fetchers
This part fetches the articles from relevant fintech research websites, such as:- Springer Open Financial Innovation journal
- Cryptology ePrint Archive
- The Journal of Fintech (worldscientific.com)
- ARXIV (q-fin and CS CR tipics)
- IMF (Working papers, World Economic Outlook, International economic and financial papers). Not always relevant, but from time to time it's interesting to see what the IMF discusses about FinTech.
- Journal of Finance - Wiley
- Review of Financial Studies - University of Oxford Press
- Review of Finance - European Finance Association
- Journal of Corporate Finance - ScienceDirect
- European Journal of Information Systems - Taylor and Francis
- Journal of Information Technology - SAGE
- Industrial and Corporate Change - University of Oxford Press
Note: This script does not scrape the data from the websites, and only fetches the publicly available data through RSS feeds. Full articles remain the intellectual property of their respective publishers and authors.
- Step 5: Display
A function to display the articles by descending date in a customisable and interactive Plotly table.
- Step 6: Main
Where everything finally happens: The block where the functions are called and the output displayed.
Although this project has been undertaken as a personal endeavour, I have been fortunate to benefit from the support of my professors. I wish to thank in particular Dr. A. Ballis, whose guidance in selecting reviews and websites employed in this script proved invaluable.
Code: © G.RUQUILLA - MIT License