If millions of people are quietly burning paths through the Web every day, the simplest possible system is one that just remembers the path. Capture the query, capture the clicks, and let the trail itself become the search result.
This patent - filed in 2004 while building Trexy in London - describes a way to record and re-use those trails across one or more search engines. The user keeps using whichever engine they prefer (Google, a topic-specific engine, anything in between). A small client-side detector watches for search-form submissions and quietly records the query and the pages clicked afterwards. Server-side, those steps become a re-usable trail.
The novel bit isn't the recording - it's that the trail spans any search engine the user touches, and that consecutively accessed pages are stored as ordered "trail steps" tied back to the query. That makes a trail searchable in its own right: future users typing similar queries can land on a path others have already worn smooth.
Several pieces fall out of this design. A small adapter manager handles the differences between search-engine URLs and form fields. Trails are stored cheaply in memory and flushed to disk. A trail searcher ranks them by recency, frequency and source. The trails themselves - not the documents - become the index.
It's a quiet idea, but a stubborn one. Search engines mostly rank documents by what publishers say about each other. Trails rank documents by what users actually did after asking. Authority is conferred by use, not by markup - closer to Vannevar Bush's 1945 vision of a Memex than to any crawler.
This was the part of Bush's As We May Think that stuck with me from the start, and that the patent is really an attempt to implement:
There is a new profession of trail blazers, those who find delight in the task of establishing useful trails through the enormous mass of the common record. The inheritance from the master becomes, not only his additions to the world's record, but for his disciples the entire scaffolding by which they are erected.
Vannevar Bush, "As We May Think", The Atlantic, July 1945
The trailblazer is the user. The trail recorder is the scaffolding. The next person typing a similar query inherits both. The full backstory, including the Brisbane Goat Trail that nudged this whole line of thinking along, is in Search Trails: Back to the Future.
The US grant has been cited by 32 later patent families, including filings from Microsoft, Yahoo, eBay, Oracle and Excalibur IP. Citations are an examiner's signal that the earlier work was genuinely relevant to what someone tried to claim later - a quiet but durable measure of how far an idea travels.
Citing assignees include:
Citation count and assignees as listed on Google Patents for US 8,572,100 B2. Citations span search-result trails, deeplinks, behavioural-variability search, query-related suggestions and trail-based recommendation systems.
I built the invention into Trexy, a search-trail engine that ran from London between 2003 and 2012. The technology won the British Computer Society's IT Industry Services Award in 2006. The full backstory is in Search Trails: Back to the Future.