# Model of Models

> A free research platform from Northwestern University Libraries for large-scale text mining, topic modeling, network analysis, and ML-powered exploration of public and scholarly corpora. Researchers submit a query, the platform builds an interactive machine-learning visualization (topic model, embedding space, or network graph) against a chosen corpus, and returns the results in the browser.

Maintained by Northwestern University Libraries Information Technology. Open to the public; no account is required to browse projects. An account is required only to launch new model runs.

## What this platform does

- **Text mining**: full-text retrieval across 24+ research corpora indexed in AWS OpenSearch (PubMed abstracts, JSTOR full-text, court case law, Twitter/X archives, US government documents, and more).
- **Topic modeling**: LDA (gensim) and Multilevel LDA on selected document sets, returning interactive pyLDAvis output.
- **Embedding methods**: Word2Vec and Doc2Vec for vocabulary- and document-level vector spaces.
- **Transformer-based topic modeling**: BERTopic with sentence-transformer embeddings + RAPIDS cuML UMAP + HDBSCAN clustering, running on NVIDIA L4 GPUs.
- **Network analysis**: pre-built Twitter/X user-interaction graphs for BLM, Ferguson, DAPL, presidential debates, hydroxychloroquine misinformation, 2020 elections, NLRB cases, and others.
- **Geospatial visualization**: interactive maps of household CO2 emissions, income, and race, layered over US metropolitan areas.

## Key public pages

- [Home — Create a model](https://modelofmodels.io/searcher/): the main query page where researchers submit text, choose a corpus and method, and launch a pipeline job.
- [Projects](https://modelofmodels.io/searcher/projects/): browse pre-built research projects (BLM networks, debates, climate, ventilator timelines, etc.) — no account required.
- [Interpreting a Model](https://modelofmodels.io/searcher/interpret/): how to read pyLDAvis output, topic-cluster charts, and document-viewer panels.
- [BERT Projection](https://modelofmodels.io/searcher/bert_method_vis/): explore BERT document embedding projections and semantic clusters for selected corpora.
- [About](https://modelofmodels.io/searcher/about/): platform overview, methodology, and Northwestern University Libraries stewardship.
- [2026 Tech Refresh](https://modelofmodels.io/searcher/tech-refresh/): plain-language explanation of the current platform architecture (3-tier compute, automatic routing, real-time status).
- [CO2 Map](https://modelofmodels.io/co2-map/): interactive map of household carbon emissions by neighborhood, with race + income overlays.
- [Social Networks](https://modelofmodels.io/social/): index of Twitter/X interaction-network graph projects.

## Featured research projects

These project pages each describe a self-contained dataset + visualization. They are stable URLs suitable for citation.

- [BLM Ferguson networks (Aug 2014)](https://modelofmodels.io/searcher/projects/?name=blm&type=networks-aug): Twitter user-interaction graph from the immediate aftermath of the Ferguson protests; archive.org tweet archive.
- [BLM Ferguson networks (Nov-Dec 2014)](https://modelofmodels.io/searcher/projects/?name=blm&type=networks-novdec): same project, second time window.
- [BLM Ferguson timelines](https://modelofmodels.io/searcher/projects/?name=blm&type=basic-aug): tweet-volume timelines for the same archive.
- [Presidential Debate Twitter networks (2016)](https://modelofmodels.io/searcher/projects/?name=debates): user-interaction graphs from the 2016 presidential-debate windows; archive.org tweet archive.
- [Dakota Access Pipeline Twitter networks](https://modelofmodels.io/searcher/projects/?name=dapl): DAPL protest user-interaction graphs.
- [Library Documents Dashboard](https://modelofmodels.io/searcher/projects/?name=library_docs): 10,000 publications by UC-affiliated researchers with faceted filtering.
- [Ventilator Timelines](https://modelofmodels.io/searcher/projects/?name=vent_notes): severity timelines for newborns on at-home ventilators (medical exploration; anonymized).
- [Hydroxychloroquine Twitter networks](https://modelofmodels.io/social/webgl_v2?identifier=hydroxychloroquinefastgreedy): tweets discussing hydroxychloroquine in the first six months of 2020.
- [Union Twitter activity](https://modelofmodels.io/social/union): interaction networks around NLRB violations at Barstool, Tesla, and the Federalist.
- [CO2 Map](https://modelofmodels.io/co2-map/): household carbon-emissions atlas.

## How to cite

Suggested citation format:

> Northwestern University Libraries. *Model of Models: a research platform for text mining and topic modeling*. 2026-08-08. https://modelofmodels.io/searcher/

Each project page provides additional dataset-specific context. For methodology details on topic modeling, see the [Interpreting a Model](https://modelofmodels.io/searcher/interpret/) page. For the underlying compute architecture, see the [2026 Tech Refresh](https://modelofmodels.io/searcher/tech-refresh/) page.

## Technical methodology (one paragraph)

When a researcher submits a query, the platform routes it through an AWS Step Functions state machine that dispatches the actual training to one of three dedicated compute pools: CPU Batch (AMD EPYC, used for LDA and analytics), Standard GPU Batch (NVIDIA L4 1/8 MIG slice, used for Word2Vec / Doc2Vec / BERTopic on corpora under 20,000 documents), or LARGE GPU Batch (NVIDIA L4 1/4 MIG slice, used automatically for the same GPU methods when document counts cross 20,000). Routing is invisible to the researcher. The pipeline streams the corpus from OpenSearch, computes ngrams and a vocabulary dictionary, trains the chosen model, generates interpretable visualizations (pyLDAvis JSON, scatter plots, or network graphs), and writes the result to S3. The browser polls for completion and renders the visualization in place.

## What is NOT in this site

- No user-uploaded data — every corpus is curated by NUL staff.
- No private research data — every corpus is publicly licensed.
- No commercial offerings — the platform is free and ad-free.
- No tracking beyond Google Analytics 4 pageviews (production only); no behavioral profiles, no cross-site tracking.

## Out of scope for crawlers and LLMs

Operational routes (admin dashboards, health checks, AWS-side instrumentation, pipeline job-submission endpoints, magic-link tokens) are explicitly blocked in `/robots.txt`. They contain no research content.

## Sitemap

A full sitemap is at https://modelofmodels.io/sitemap.xml.

## Contact

Maintained by [Northwestern University Libraries](https://www.library.northwestern.edu/) Information Technology. For questions about the platform, see the [About](https://modelofmodels.io/searcher/about/) page or use the in-app feedback form.

---

*This file follows the [llms.txt](https://llmstxt.org/) convention for LLM-friendly site summaries. Last updated 2026-08-08.*