RumourFlow Web Platform Prototype

RumourFlow is a prototype web-based toolkit for tracing source attribution and text reuse in historical newspapers.

Users upload a CSV/XLSX article dataset. The app returns reviewable source-cue candidates, gazetteer matches, text-reuse edges, reuse clusters, and graph-ready outputs.

Live demo: https://rumourflow-apzmwwkqv9dqezkcxadcag.streamlit.app

Status

RumourFlow currently includes a Streamlit demo interface, a source-detection and gazetteer-matching backend, text-reuse clustering, graph export, an input template, and a Moon Hoax sample dataset.

The system is suitable for local testing, poster demonstration, and method development. It produces reviewable candidates rather than definitive historical claims, and it is not yet a hardened public production service.

Input format

Required columns:

column	meaning
article_id	unique article identifier
newspaper_title	newspaper title where the article appeared
date	publication date
text	OCR/article text

Recommended columns:

place, language, url, collection, issue_id, page_id.

A template is provided in:

templates/rumourflow_input_template.xlsx

Outputs

After processing, RumourFlow creates:

file	purpose
articles_normalised.csv	cleaned input
source_candidates.csv	detected source mentions and gazetteer matches
source_candidates_for_review.xlsx	spreadsheet for manual checking
text_reuse_edges.csv	article-pair reuse links
reuse_clusters.csv	reuse families
article_cluster_membership.csv	article-to-cluster map
integrated_nodes.csv	graph nodes
integrated_edges.csv	graph edges
integrated_graph.graphml	graph export for Gephi/Cytoscape
integrated_graph.gexf	graph export
summary.json	run statistics

Run locally

Install dependencies:

pip install -r requirements.txt

Start the app:

streamlit run app.py

Then open the local URL shown by Streamlit.

Gazetteer

The prototype expects the gazetteer here:

data/gazetteer_full_three_archives.csv

The included package may contain this file if it was available during packaging. If not, place the full gazetteer CSV there manually.

Notes for public deployment

For a public website, add:

1. upload size limits;
2. temporary file deletion;
3. a privacy statement;
4. job queue for large datasets;
5. user-facing documentation;
6. clearer review dashboard;
7. stable GitHub/Zenodo release.

Historical date parsing

The prototype accepts mixed date formats, including:

• September 09, 1835
• 5/31/1900 0:00
• Thursday 02 May 1861
• Saturday 22 December 1877

If a date still cannot be parsed, the article can be processed, but text-reuse directionality may rely on input order rather than publication date.

Moon Hoax sample date standardisation

The included Moon Hoax sample was rebuilt from Combined_moon_hoax(2).xlsx and uses standard ISO dates in the date column. Two dates that appeared as corrupted values in an earlier CSV are corrected from the original XLSX:

• moon_0017: 1900-05-31 (The Wichita daily eagle. [volume])
• moon_0018: 1900-06-07 (The morning news. [volume])

The original date value is preserved in original_date, and a readable version is provided in date_display.

Website/platform framing

RumourFlow is designed as a web-based research toolkit. A user uploads a historical newspaper article dataset, and the platform returns reviewable source candidates, text-reuse clusters, and graph-ready connection files.

Suggested public-facing description:

RumourFlow is a prototype platform for tracing source attribution and text reuse in historical newspapers. It turns uploaded article datasets into reviewable maps of source cues, reuse families, and possible information flows.

Platform status

Current version:

• local Streamlit demo interface;
• backend pipeline for source detection, gazetteer matching, text reuse, clustering, and graph export;
• sample Moon Hoax dataset;
• input template;
• downloadable ZIP outputs.

Still needed before public release:

• public hosting;
• upload-size limits;
• retention/deletion policy;
• interactive manual review dashboard;
• improved title-ranking logic;
• documentation and citation page.