Skip to main content

View Diary: High Impact Diary Lists: Current Methods (25 comments)

Comment Preferences

  •  great idea (3+ / 0-)
    Recommended by:
    knowthings, Eddie Haskell, va dare

    I was hoping you would be up for that after the move.  But you're still in boxes?

    Do you need data on a couple of individual diaries, or aggregate data over many diaries?

    •  We move tomorrow... (1+ / 0-)
      Recommended by:
      knowthings

      The computers all went off-net on Friday, as part of box-everything-up day. Hopefully, we'll have some semblance of being online in a few days, though reconnecting computers may not be the highest items on the priority list. If anyone desperately needs those graphs, I can put them up on imageshack or something; the actual source files are on my laptop.

      In terms of the analysis, if possible, I'd like to look at the raw data; basically a table listing the posting times of all comments in as many diaries as is convenient. An entire day's worth would be best, but that's a lot of data. What I was thinking of doing (assuming I get the chance) is a second generation of my Monte Carlo model incorporating diary visibility. Visibility starts out high, and then decays as subsequent diaries are added to the list; diaries that pass a certain threshold in comments gain visibility. In addition to the attraction parameter (due to the number of comments already placed), the chance of a diary getting comments depends on its visibility coeffiicient. Bascially, trying to model the behavior of the Recent and Recommended lists.

      -dms

      •  I already have a full days comments in hand (1+ / 0-)
        Recommended by:
        knowthings

        let me know when you've time to look at them.  I know it must be around here somewhere.  They seem to follow a saturation curve (logistic).  Did we talk about this already?

        Modeling recent and recommended list as a combined visibility factor seems a good idea.  But I'm not sure it is comments you should count since showing up on the recommended list doesn't pay any attention to comments.

        I don't have a full days recommendations laid out by time, I've never looked at that before.

        •  How big is the file? (1+ / 0-)
          Recommended by:
          knowthings

          If it's a manageable size (a couple of megabytes at most), you could email the sucker to me, and I'll look at it when I have a chance. dmsilev (at) uchicago.edu. Otherwise, stick it up on a web server or anonymous FTP server, and send me the URL.

          We did talk about the observed saturation effect; I remember you showing the 4-parameter and 6-parameter curves that did a pretty reasonable job of matching most of the data, based around diary visibility.

          I use comments as a proxy for recommends just because it's easier to track a single variable, and there's a reasonable (though certainly not perfect) correlation between the two. A really good model would certainly want to track both independently; that would also allow for real modeling of the Recommended list.

          Another thing that implementing diary visibility would allow is relatively straightforward incorporation of diarist reputation; the hotlist-effect can be modeled by implementing a rollover effect. Highly-attractive diaries generate a bonus to visibility for that diarist in the next round of simulations.

          -dms

          •  It's small (1+ / 0-)
            Recommended by:
            knowthings

            bzipped, all comments on diaries posted one day from 10 AM to 10 PM is under 100 kb.  I'll mail it.  

            I restricted the time to get around the drastic diurnal variation.  Maybe not such a good idea, but it's a start.

Subscribe or Donate to support Daily Kos.

Click here for the mobile view of the site