Catalogue Explorer

A new corpus comes on board, and the time to understand it is short. You cannot read it all, and skimming a handful of documents reveals little about the whole.

This platform reads everything and turns the corpus into a map you can browse, whether it holds three thousand documents or well beyond a hundred thousand. The map is a multi-aspect topic tree, built bottom-up from the corpus itself with no predefined taxonomy imposed on it.

What you can do

• 🔎Every topic, along any aspect, expands to its documents and the exact sentences that locate it, so nothing you see stands without evidence you can check.
• 📚Once the map exists, the whole corpus becomes searchable by keyword or theme, returning every document that carries it.
• 🔗Combine two topics to find the documents where they co-occur.
• ↪While reading a document, click any linked phrase that draws your attention, as you would on Wikipedia, to see the path its topic takes through the tree, its neighboring topics, how many documents it spans, and the documents themselves to read on.

How it works

A language model reads each document and extracts the phrases that carry its meaning. Those phrases are clustered by similarity and refined into clean topics, then organized bottom-up into groups and aspects. Because the structure grows from the evidence rather than a fixed schema, every node stays anchored to the text that produced it.

Reading it well

Topic names are generated summaries, so when one looks off, open it and read the phrases underneath. The top-level grouping is a useful first cut, and the "Other" category holds the topics too distinct to place or too small to stand alone.