Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!samsung!usc!elroy.jpl.nasa.gov!decwrl!nsc!pyramid!thirdi!peter From: peter@thirdi.UUCP (Peter Rowell) Newsgroups: comp.databases Subject: Re: >Fault-tolerant Information Recall Message-ID: <505@thirdi.UUCP> Date: 14 Apr 90 19:07:27 GMT References: <1990Apr3.200220.9513@sctc.com> <1990Apr13.191658.10233@sctc.com> Reply-To: peter@lloyd.UUCP (Peter Rowell) Followup-To: comp.databases Distribution: na Organization: Third Eye Software, Menlo Park, CA Lines: 41 Although I appreciate Michael's plug for Third Eye, I would like to clarify a few things. In article <1990Apr13.191658.10233@sctc.com> endrizzi@sctc.com (Michael Endrizzi ) writes: >ThirdEye software does text-retrieval and classification >for large firms like Dialog and Nexus. We do not have any business relationship with either Knight-Ridder's Dialog group or Mead Data Central (Nexis/Lexis). (Which is not to say we wouldn't like to!) >Peter said that ThirdEye talked to a major RDBMS firm >about combining technologies, and the RDBMS firm went >critical when ThirdEye explained the text-retrieval >process. Actually, they had no problems with straight "keyword and boolean", because it is "precise" (i.e. there is an "exactly correct" set of documents which will be returned for any given query). What they were not too wild about was our "content similarity" method which is based on Salton's vector-space-model involving weighted attribute vectors. RDBMS people are not happy when you say that a document "might" be retrieved. >Peter said that text-retrieval is considered >an orphan from Computer Science perhaps because it is >not an exact science, so it found a home in Library Science. Actually, it isn't really in the library schools either. It mostly isn't anywhere (there are a few exceptions, like Cornell, Virginia Tech and some others). The problem seems to be that text retrieval algorithms do not submit well to classic academic analysis and thus do not make good thesis material. Most of the existing algorithms have been arrived at through almost purely empirical means - probably not what an advisor wants to see. ---------------------------------------------------------------------- Peter Rowell peter@thirdi.uucp Third Eye Software, Inc. ...!{apple,pyramid,sun}!peter 535 Middlefield ROad, Suite 170 (415) 321-0967 Menlo Park, CA 94025