Tuesday, July 31, 2012

Surfacing good comments

Comments are supposed to make digital media more engaging and interactive. Somewhere in the crowd of readers and viewers are ideas, insight and thoughtful criticism. However, the comments found on popular internet sites like YouTube or news sites generally inspire a loss of faith in humanity. From the comfort and pseudo-anonymity of thousands of living rooms comes a stream of abuse, wingnuttery and outright stupidity that overwhelms etiquette and common sense.

Clay Shirky thinks Gawker is on to something with its attempts to surface quality comments. Gawker redesigned their comments section to serve the people reading the comments, rather than the people writing them, moving most comments off the main page of an article and enabling enhanced curation.

Manual curation is labor intensive. I wonder whether a machine learning approach might be able to do a reasonable job of identifying good comments, or at least weeding out most of the inane ones. It's not really much different than spam filtering. That might make a fun little project.

1 comment:

  1. Another approach that seems to offer some promise is the Q&A format of Stack Overflow and Quora. These sites seem to generate some quality answers because
    1. The Q&A format provides a clear structure by which to judge responses.
    2. Readers can very easily vote up / down answers, and also modify earlier answers. The sites do a reasonable job using microscopic contributions of readers to help curate meaningful content.
    3. On StackOverflow, people from the community can become moderaters for particular subsets of content (defined by tags)

    I'm not sure how much additional machine leanering or company-sponsored additional curation goes on behind the scenes, but it's clear sites need to focus on serving the community, not just the commenter to produce meaningful content.

    We have recently started exploring Stack Overflow as a mechanism for doing customer support for our open source Synapse project. The idea is to sponsor a "Sage-Synapse" tag for questions we receive, and agreesively respond and answer questions with this tag. One hoped benefit is that if questions get cross-tagged (e.g. R, Sage-Synapse) it links us into the large community of people on Stack Overflow who aren't aware of us. Also, we get questions about using Synapse that are really questions about using R.... getting non-Sage people to answer these would be a huge win.