Re: [unrev-II] Knowledge Repository War Stories...

From: Eric Armstrong (
Date: Thu Mar 09 2000 - 15:10:46 PST

  • Next message: Jon Winters: "Re: [unrev-II] Knowledge Repository War Stories..."

    From: Eric Armstrong <>

    Valuable info, indeed!

    You are quite right that "horror stories" tracks important information
    what can go wrong, and how. The fellow from Citigroup who spoke in
    Session 9 mentioned several important Murphyisms to be alert for.

    QUESTION: What is HyperNews? Do you have a link to it? It may be
    very close to what I am proposing this evening. If so, I could have used

    a DKR much earlier! (Typically, I find that things that sound promising
    don't turn out to be as good as hoped, but there are always

    The issue you mentioned where people ask a question that has already
    been answered is very, very important. It creates some interesting
    requirements for the system. There are a couple of related issues along
    the same lines, as well. First, the question you raised:

    Use Case: Person asks a question that has already been answered.
      System Action:
         1) It is important that the question is *not* entered into the
             until it has been refined sufficiently to be a unique
    question.(Or it
             should be possible to delete it later, as well as revise it.)

         2) Ideally, the system will be "friendly" enough for people to find

             things. Most NewsGroups aren't that friendly, because of the
             large volume of frequently redudant material (see point #1) and

             the inability to group things together. So the ability to
             the organization of the data by adding categories and moving or

             copying items to them is paramount.

         3) Even with the best system, a new person will frequently ask a
             question "in the wrong way" -- using the wrong terminology,
             That makes automated detection a difficult task. What's needed
             is a way that makes it tres simple for another user of the
             to respond "answered in X", where X is a link to the
             node. That answer might have the effect of adding the question
             an "alternative form" to the original, or simply prevent it
    from being
             added to the database, unless and until it is redefined and
             as a new question that no one responds to in that way.

    I recently sent out a query to the company wide mailing list: Anyone
    any good optometrists? I got back about 10 replies. One of them was from

    a fellow who saved the 10 replies *he* got when he asked the same
    question a few months earlier. To continue his good work, I merged my
    responses with his. The results were instructive:

    There was a 30% overlap between the two lists. Three people were kind
    enough to repeat the same recommendation they had given earlier. The
    other 7 were new -- from new people in the organization or because they
    had come across new information.

    That raised some interesting questions:
       a) How are questions updated with new answers?
           It is repeated queries that spark new additions. What should
           ideally, is that new answers are linked to the original question,
           than to the duplicate (which should not be entered). Ideally,
           are eliminated there, as well. But again, that needs to be a
    human process,
           since Jeffrey Archer in one response might be J. Archer in
    another, and
           that might be hard to distinquish from his brother, K. Archer, in
    a third.

       b) How are answers updated?
           If J. Archer changes his address from Cupertino to Pleasanton,
           important information, and the correction should replace the

       c) How are invalid and out of date responses removed?
           If J. Archer retired, how/when is that response removed from the

    All of these issues apply to a how a system is designed, or how it is
    as well, since change is the norm, rather than the exception.

    They're important issues to solve. Even after we figure out what to do,
    figuring out how to do it is hard...

    Jon Winters wrote:

    > From: Jon Winters <>
    > Having learned long ago that we should learn from our mistakes I would
    > be
    > interested in hearing any stories of knowledge repositories gone
    > horribly
    > wrong.
    > In the past I have set up a couple of different web based discussion
    > forums and I currently maintain a private NNTP server. The NNTP
    > server
    > limits the audience to those geeky enough to know what a newsreader is
    > and
    > how to use it. Depending on your target audience this is a good thing
    > or
    > a bad thing. In our case switching to NNTP gave us a more technical
    > audience and seeded out a lot of the off topic drivel.
    > The web based systems reached different audience. Hypernews was
    > working
    > great for a photographers forum that I was running. It was very easy
    > to
    > use and extremely stable once it was in place. Many of the
    > photographers
    > who used it were not computer people and they appreciated the ease of
    > use.
    > Unfortunately I had it on a server owned by my employer and they shut
    > down
    > the office pulling the plug. One day I would like to re-install
    > Hypernews
    > on my own server.
    > Lastly I've noticed that new users will always ask the same questions
    > even
    > if they have been answered many times before and there are tools to
    > search
    > the previous posts.
    > Is it human nature to want to ask someone instead of searching a
    > knowledge
    > repository? If so then an agent or bot might be in order. The bot
    > acts
    > just like a visitor to the site. The bot hangs out and listens for
    > frequently asked questions and when it gets a hit it searches the
    > repository, old chat logs, etc and replies with an intelligent answer.
    > (in chat or email or whatever system the visitor was using when they
    > asked)
    > Now the bot should identify itself as a bot and it should be smart
    > enough
    > not to send out duplicate replies if the visitor did not find what
    > they
    > were looking for the first time.
    > Bots should not behave like those dreaded phone menu systems. (if at
    > all
    > possible)
    > Being an admin Its also rough to decide what to keep and what to throw
    > away. Currently my NNTP server is set to expire articles every 100
    > days.
    > I would like a system that would expire by default after 100 days
    > unless a
    > post were flagged as a keeper. Keepers get moved to archival storage.
    > The old hypernews system I was using kept everything and over time the
    > signal to noise ratio degraded.
    > Post your stories if you have them... change the names to protect the
    > innocent if you need to.
    > Thanks
    > --
    > Jon Winters
    > OpenVerse
    > -----------------------------------------------------------------------
    > -----------------------------------------------------------------------
    > Community email addresses:
    > Post message:
    > Subscribe:
    > Unsubscribe:
    > List owner:
    > Shortcut URL to this page:

    GET A NEXTCARD VISA, in 30 seconds! Get rates as low as 2.9%
    Intro or 9.9% Fixed APR and no hidden fees. Apply NOW!

    Community email addresses:
      Post message:
      List owner:

    Shortcut URL to this page:

    This archive was generated by hypermail 2b29 : Thu Mar 09 2000 - 15:17:48 PST