ProFusion

What ProFusion was, and where its results came from

ProFusion had no crawler and no index. It was a metasearch engine: one query in, dispatched in parallel to several underlying search engines, duplicates stripped out, and a single merged ranking returned. Everything it showed had been found by somebody else's index. As a structural description that is neutral — it is what the product was designed to be — but it does mean ProFusion never owned anything about the web itself.

What makes it worth a page is where it came from. ProFusion was academic research first, built at the University of Kansas and published in a peer-reviewed journal in 1996, and it is the most technically documented engine of its generation for that reason. Its second life, after Intelliseek bought it in 2000, was as a gateway to the "Invisible Web" — the specialist databases that general crawlers could not reach.

It is gone. No shutdown announcement exists, which is itself part of the story: ProFusion was still described as operating in Search Engine Watch's metasearch guide of 23 March 2005, and it disappeared at some point after that without anybody recording when. It does not operate as a search engine today, and because "Profusion" is a common commercial name, whatever sits at that domain now should be assumed to be an unrelated business until checked.

How the 1996 system worked

The original design is documented precisely in Susan Gauch, Guijun Wang and Mario Gomez, "ProFusion: Intelligent Fusion from Multiple, Distributed Search Engines", Journal of Universal Computer Science 2(9), 1996. Four mechanisms are worth stating exactly, because they are the substance of the thing:

• Six back-end engines. In the 1996 version: AltaVista, Excite, InfoSeek, Lycos, Open Text and WebCrawler.
• Engine selection in two modes. Users could choose engines by hand, or use Automatic Pick, in which ProFusion classified the incoming query into one of a 13-category taxonomy derived from Internet newsgroup hierarchies and then selected the three best-performing engines for that category, using confidence scores calibrated offline in advance.
• Weighted score merging. The merged ranking combined the query-document match score reported by the originating engine with the estimated accuracy of that engine — each engine's own relevance judgement, discounted by how reliable it had proved for that subject.
• Duplicate removal in three stages. Identical URLs; normalised variants of the same URL (http://server/ against http://server/index.html); and n-gram similarity over URL paths to catch near-duplicates. This was considerably more careful than contemporaries managed.

The paper also reports an evaluation, and it needs handling honestly. On a test of 12 user-selected queries, Automatic Pick ProFusion retrieved 170 relevant documents with only 4 duplicates, which the authors report as outperforming the individual engines and the rival metasearch services MetaCrawler and SavvySearch. Twelve queries is a very small sample, the evaluation was run by the system's own authors, and they judged relevance themselves. It is the authors' claim, not an established finding, and anyone repeating it without those caveats is overstating the evidence.

A university project, and a disputed start date

ProFusion was created in the Department of Electrical Engineering and Computer Science at the University of Kansas by Dr Susan Gauch, then an associate professor there, with graduate researchers Guijun Wang and Mario Gomez. Gauch later moved to the University of Arkansas, where she remains a listed faculty expert in information retrieval and personalised search. Whether Wang and Gomez were students or staff, and their exact roles, is not established.

The start year is disputed and both accounts are printed here. Information Today's NewsBreaks (12 February 2001) says ProFusion "originated from a 1995 University of Kansas research project". Metamend's archived engine profile says it was "originally created and co-founded in 1996 by Dr. Susan Gauch". The peer-reviewed paper describing a working system appeared in 1996, which is consistent with a 1995 start and a 1996 publication but does not settle it. The defensible statement is that ProFusion was created at the University of Kansas in 1995–96. No month or day is sourced anywhere.

There is also no standalone English Wikipedia article on ProFusion the metasearch engine — the title leads to unrelated topics, and a later Gauch-adjacent search product, Polycola, is sometimes confused with it. The 1996 JUCS paper is the reference that actually documents the system.

The Intelliseek years and the Invisible Web

ProFusion was acquired by Intelliseek, Inc. of Cincinnati, Ohio in 2000. The month is disputed: Information Today says March 2000; Search Engine Watch and Metamend both say April 2000. Terms were not disclosed in any accessible source. InternetNews headlined the deal "IntelliSeek Buys Last Independent Meta Search Firm", which is a useful indication of how the category was regarded by then — ProFusion was significant within metasearch even though it was never a mass-market destination.

Intelliseek repositioned it. Per Information Today's report of 12 February 2001, the version 2 beta offered direct access to the "Invisible Web" across over 1,000 targeted sources — TerraServer, Adobe PDF Search, Britannica.com, the New York Times and the US Patent database among the named examples — plus Search Alerts (saved queries re-run automatically with email or wireless notification) and Page Alerts (monitoring specific pages for relevant changes). A companion directory, InvisibleWeb.com, listed over 12,000 searchable sources. The beta reportedly experienced traffic overload initially, which is evidence of interest rather than a measurement of it.

Metamend's later profile, undated but roughly a mid-2000s snapshot, lists the general-web back ends of that era as About, AltaVista, LookSmart, MSN, Teoma, Yahoo!, AllTheWeb, AOL, Lycos, Netscape and WiseNut, with up to ten engines queried at once. The architecture had shifted from "query six web engines" to "query the right specialist database", and that was a genuinely good diagnosis of where the valuable material was.

How it made money

In its academic phase it had no business model at all — it was a research prototype on university infrastructure. Under Intelliseek it was a free, advertising-supported consumer service that also functioned as a shop window for the company's enterprise business. Intelliseek's real trade was corporate information monitoring, and ProFusion.com was the public demonstration of the technology.

The alerting features give the game away: Search Alerts and Page Alerts are exactly what gets built when the paying customer is a corporate analyst tracking a market rather than a casual searcher looking something up once. No subscription, paid-inclusion or metered-API revenue for ProFusion itself is documented, and nor is any acquisition price.

Intelliseek's consumer-generated-media business merged into Nielsen BuzzMetrics around 2006. Whether ProFusion was shut down, sold or simply abandoned in that transition is not recorded anywhere accessible.

Privacy, and one genuine distinction

There is no current privacy question, because there is no current service. One historical point is worth making because it is a real architectural difference rather than a marketing claim: ProFusion's original routing model was calibrated offline, by testing engines against subject categories in advance, rather than learned from live user clicks. SavvySearch, its closest counterpart, learned routing from what users selected. That means ProFusion's core relevance mechanism did not structurally depend on profiling anybody — unusual for the period, and worth noting.

It says nothing, however, about what the commercial ProFusion.com actually logged, which is not documented. Under Intelliseek the Search Alerts feature necessarily stored users' saved queries and contact details. No privacy policy, regulatory action or independent audit was located.

What it got right, and what it got wrong

ProFusion is the most technically serious engine of its cohort, and the reason is that it had to survive peer review. Three of its ideas were right and are still right: classify the query before deciding where to send it; trust different sources for different subjects; and take duplicate detection seriously enough to do n-gram path matching. Modern federated search systems implement all three, often without knowing where they came from. Its second act was also a correct diagnosis — recognising in 2000–2001 that the most valuable material sat in databases no crawler could reach, and building a router to them, describes a problem that still is not solved.

What it was bad at were the weaknesses of a research system in a commercial market. The engine-quality calibration was offline and hand-built, so it went stale as the underlying engines changed — and between 1996 and 2001 they changed enormously: AltaVista declined, Open Text and InfoSeek vanished, Google arrived. Its published evaluation rests on twelve queries judged by its own authors. The Invisible Web pivot depended on hand-curated source lists that were expensive to maintain and impossible to keep current across a thousand-plus sources. And it shared the terminal metasearch problem: once one engine is dramatically better than all the others, there is nothing left to fuse.

Nobody can use ProFusion; it has not operated for roughly two decades and has no successor. The honest framing for a directory is that ProFusion matters as research, not as a product. Anyone building federated search, retrieval over heterogeneous sources, or routing between tools should read the 1996 paper, because the problem statement is nearly identical. Anyone looking for a search engine to use today should look elsewhere.

Common misconceptions

• "ProFusion was a search engine." It was a metasearch engine — no crawler, no index, ever. Everything came from other engines and, later, other databases.
• "It was an Intelliseek product." It was a University of Kansas research project first, and that is the interesting part. Intelliseek bought it in 2000, four or five years in.
• "It was the same kind of thing as SavvySearch." Superficially yes, architecturally no. SavvySearch learned engine selection from live usage; ProFusion classified queries into a 13-category taxonomy and applied offline per-category confidence scores. They are the two serious 1990s metasearch designs and they solved routing differently.
• "ProFusion beat MetaCrawler and SavvySearch." Its own paper claims that, on twelve queries, with relevance judged by its own authors. Cite it with those caveats or not at all.
• "ProFusion shut down on a particular date." No such date is documented. It was operating in March 2005 and gone by the time anyone looked again.
• "ProFusion invented deep-web search." It was an early and well-executed gateway to it in 2001, not the inventor of the concept.

Frequently asked questions

Sources

• jucs.org/jucs_2_9/profusion_intelligent_fusion_from/Gauch_S.html
• rsl.ku.edu/publications/documents/Gauch1996_WebNet96.pdf
• newsbreaks.onlineinc.com/Digest/Intelliseek-Announces-the-Beta-Release-of-Pr…
• searchenginewatch.com/2005/03/23/metacrawlers-and-metasearch-engines/
• metamend.com/archive/past-engines/profusion-search-engine/
• expertstest.uark.edu/experts/185/susan-gauch