In the early '90s, Thinking Machines Inc. experimented with the use of massively-parallel computers to perform relevance searching. After locating a valuable text, the scholar could then ask the computer to find similar documents. The computer system compared the full text of the original text with that of all other texts in the database to find those most similar to it.-- effectively, it conducts a search based on thousands of keywords.

They found this analysis method to be highly effective, and the quality of the search improved with the length of the comparison prototype. It also demonstrated how massively-parallel architectures could result in substantially different analysis algorithms from those used with conventional computers.

Relevance searching is fundamentally different from Boolean keywords in that it seeks not to find texts that contain all of the keywords, but a similar overall pattern of word usage -- a type of analysis that is extravagantly difficult to perform without massively-parallel computing architectures.

In the mid-90's, Thinking Machines encountered major financial losses in its attempts to market parallel computers. It responded by scaling back its hardware and research programs and now limits itself to developing and marketing data-mining software based on relevance searching algorithms.