mikem@tekcae.UUCP (Michael E. Meyer) (05/17/85)
While at a university a few years back, our group developed an automated consulting system. This system allowed the user to browse through a database (via keywords) of frequently asked questions and responses to see if the area of concern had already been discussed with a previous user. If the concern had already been discussed (50-70% of the time) then usually that ended the situation. If it was not, the user could then submit a question to a panel of "experts" electronically, one of whom would answer the question. A copy of the question/response was stored in the database. Of course, the database took some time to become really useful, but after a few months we realized a SIGNIFICANT reduction in the face-to-face consulting load. I am not very pleased with the UNIX manuals system, particularly with the lack of examples. Other attempts at "help" systems, for instance a menu driven category-oriented envelope around the "man" command, have helped but we still have the basic problem. Am I the only person on the net that feels this way? Have others implemented an automated consulting system? If there is a significant response, I will be happy to post to the net... Thank you, Michael E. Meyer ________________________________________ USENET: ...!tektronix!tekcae!mikem ARPA: tekcae!mikem.tek@CSNET-RELAY USnail: TEKTRONIX, Inc. P.O. Box 500, MS 19-075 Beaverton, OR 97077 USA (503) 627-2628
mikem@tekcae.UUCP (Michael E. Meyer) (05/22/85)
Here are the responses I have received to date regarding the automated consulting system. I would like to apologize to those of you that were expecting some software...particularly the software that I mentioned. That package was developed on a Honeywell mainframe running GCOS 3 over 14 years ago. My posting was prompted by the thought that some wizard on the net had thought of this problem and had it solved. It turns out that this is a very real problem and it has not been solved, at least in a comprehensive manner. Before coming to Tek, I worked with Multics. One of the nice features of Multics is the Forum system which allows local newsgroups (called meetings) to be formed. I'm told by our guru's that this is not possible on UNIX with news. This might solve my problem. Perhaps I did not explain myself clearly in the first message. All I need is a a "file" to collect notes on questions asked by users and responses given by "experts" that could be searched by something like "pick". Notes, which was supported earlier might have been useful. Hope all of this has been useful... Mike (Michael E. Meyer) ________________________________________ USENET: ...!tektronix!tekcae!mikem ARPA: tekcae!mikem.tek@CSNET-RELAY USnail: TEKTRONIX, Inc. P.O. Box 500, MS 19-075 Beaverton, OR 97077 USA (503) 627-2628 <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tektronix!allegra!packard!harvard!sasaki Organization: Harvard Science Center Your idea sounds like a good one. I've been trying to figure out how to organize such a data-base system for help topics, policy notes and other things about computer systems. I've been basically limited because of time. One thing that I am going to do for the next semester's (maybe this summer's) expert system course is to suggest an expert system for help. This seems like it would be an ideal project, simple and small enough for a project, and very useful. The ideal system would be a combination of an expert system and your system. <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tektronix!ihnp4!uiucdcs!liberte I am also very unhappy with Unix manuals. I try to just keep them accurate. I would be interested to learn more about the automated help system you developed. Which university? Are they willing to distribute it? Dan LaLiberte <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tektronix!watmath!watdcsu!herbie Organization: U of Waterloo the machine that i'm posting this from's primary purpose is to provide an environment for text processing to the university community. it is available to anyone with a billing account. all other machines on campus are tied to specific research grants or a dedicated to student usage for courses. the user community consists mostly of people who have never used Unix before, and sometime, no other computers before. we have a terrible time with the provided system documentation and so have cooked up some specific manuals for text processing to help these people. unix system documentation is appalling for the first time user. as i mentioned before in net.unix (and i wasn't kidding), unix man documentation is for a unix guru with a bad memory. at waterloo, many of the man pages have been rewritten for clarity, but they still suck. a lot of the /usr/doc stuff is even worse because there is no reference documentation that says "this is how xyz is supposed to work". even as one of the three user consultants in the department who is considered a unix expert, i find that i have to continually refer people with troff questions to the people who debug and enhance troff locally. the standard troff user guides and references cover some material to death and completely ignore other aspects. it is assumed in the documentation that if something isn't covered, you are supposed to try it and see if it works. this is fine if you don't have to pay for your computing. the worst part about it is the cavalier attitude of unix programs to bugs in general, failing with cryptic error messages that assume that the source is available to find out what the error really means. we have tightly controlled access to system source because of the nature of the user population and license requirements. waterloo's primary computer system is an IBM network running VM/CMS and supports a user population of over 15,000 (that's the number of userids in the user directory). no matter what else you can say about IBM documentation, it is complete. their reference manuals are exactly that. every possible command is documented along with all the possible errors. user guides are better than any unix one i've ever seen, though they are extremely tedious for an expert wanting to learn to use a new product. the major failing of IBM manuals is their cost. i have at least two private copies of programs interfacing to the man command and i also find them less than satisfactory. a natural language help facility would work wonders, provided that it can be made as large as necessary and can be easily kept up to date. given the current state of the art, you'd probably have to dedicate a single VAX just for the help machine though. Herb Chong... <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tektronix!david@wisc-rsch.arpa we would be interested in the system you used previously, if you still have it, or access to it. We are continuously answering the same questions here at Wisconsin. david parter <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tektronix!allegra!phri!roy > a few years back, our group developed an automated consulting system. Is this software availible? Sounds interesting and useful. > Have others implemented an automated consulting system? I know of something which might be of interest available commercially. Under CCA Emacs (and TOPS-20 Emacs, and possbily others) there is a fairly sophisticated help package. At one level, emacs gives you all sorts of help in the type-?-at-any-time style. The thing that is more apropos to your question, however, is the Info program which comes with emacs. There is a heirarchical tree of documentation files with menus. With a few commands you can climb around the tree through the menus, jump back and forth between non-heirarchially related nodes via a footnote system, and do some rudimentary searching. This is rather a bit better than the "man" system, and I have had some success here with it. All of the applications software written here gets documented by adding nodes to the info tree. The user community here (scientific, but not computer people) uses the system, with some reservations. All things considered, it is an improvement on "man", but only marginally because the people here can't (or won't, which is just as bad) invest the time to develop the sophistication to take full advantage of the system's nicer features. Now, on the original work side, I did a project a while ago which never really got far enough along to go into production, but is interesting anyway. We were working with the AP newswire articles as our information data base, but it could have easily been any text. I just built a (large) hash table of all the words in all the articles. When a person wanted information about a subject, they just typed in a bunch of keywords. The keywords were de-suffixed, run past a stop list to discard "common" words and looked up in the hash table. Any article that had more that some settable threshold of hits was selected. One of the nice/cute things was that the system gave the impression of understanding natural language when it really didn't. If the user wanted to type in "Tell me about programs to check the spelling in my manuscript", fine. The sentence got "parsed" by just throwing away the common words and using the rest, order independantly. I never got much past that, but the next step would have been to list out the selected subject lines and allow the person to peruse the various articles with some sort of screen-oriented interface. Of course, you should be aware of such things as Lexis, Medlars, etc. These are more of bibliographic search systems than "automated consulting" systems, so they may not be at all what you are interested in. Well, hope this rambling has been of some help. allegra!phri!roy (Roy Smith) <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tekcrl!tektronix!decvax!seismo!aplvax!cp1!hart@maryland Organization: Chesapeake & Potomac Tel. Co., Balto. Md. I share your concern. We have many users who seem to be afraid of the standard Unix documentation. Here is my vote for posting your material. I would think you are going to be swamped with requests. Rod Hart (WA3MEZ) <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tekcrl!tektronix!uw-beaver!tikal!cholula:tim Organization: Teltone Corporation, Kirkland, WA Sounds very interesting. I'll put in a vote for posting it to the net. Tim Rosmus <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><> From: tekcrl!tektronix!decvax!mcvax!enea!luth!lka Organization: University of Lulea, Sweden >Am I the only person on the net that feels this way? No, you are not the only one. I strongly feel that the available information is insufficient. >Have others implemented an automated consulting system? Not here. We have a simple system with help menus, containing some short articles on how to do this and that. There are also a few question-answer articles. This is nothing in scope with what you are mentioning, but may show that the need really does exist. >If there is a significant response, I will be happy to post to the net... Yes, I would like to get your automated consulting system. Might it be possible that you send also your collected database? Please do! Lennart Karlsson <><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>