From a wider perspective, there is more value in viewing this as a judgmental taste issue than really any kind of moral outrage. The real question becomes why is it so offensive.

Like has already been said, if you view the user’s tastes as being highly valuable information, the censoring aspect could be mapped against what the user has chosen before. The process could be automated to a degree so that essentially your site can remember not only what the user finds offensive (“Hide this”) but also what they find extremely interesting.

There really isn’t a need to break it up into separate communities if the user is already being isolated by their own taste choices.

If you were really looking to create new communities I would recommend either building new web sites with different appeal to users OR licensing your code to other sites to focus on that sub-group of users.

This is most likely old news to you. It seemed like a good idea to write at the time

Thanks for the illumination about what the user’s think about this issue and what your company has done about it. Very interesting reading.

Thanks for making me think about this some more. I think something like that might be workable:

(1) we provide a set of ‘knobs’ that users can turn to indicate their ‘level of tolerance’ for various categories. For this discussion, let’s call this set of values a user’s “Profile.”

(2) we provide a default set of values (or two) so that casual users don’t have to know about or deal with setting up their own Profile.

(3) we continue to have a single “report as offensive” function. But instead of interpreting “report as offensive” as a scalar vote for offensiveness, we use the reporting user’s Profile to adjust the “Offensiveness Profile” of the content. (The “Offensiveness Profile” would be just like the user’s Profile, but attached to and describing a piece of content).

So, a “report as offensive” from a user whose Profile indicates they (+) do not find nudity offensive but (+) find hate-speech highly offensive would affect only the “hate-speech” value of the item’s Offensiveness Profile. E.g., the system guesses that the reason it is offensive to this user is because it contains hate-speech.

(3a) When a user does “report as offensive,” and the item in question already has an Offensiveness Profile built for it, the item’s offensiveness profile can be used to tweak the User Profile. E.g., even if the user never explicitly set their Profile values, the system can learn to adjust them based on the things they have reported as offensive — even though no one ever explicitly set values for the ‘offensiveness profile’ either.

(4) To filter content, one simply compares the Profile values of the user with the Offensiveness Profile of the item in question. If any of the values in the Offensiveness Profile exceed the value in the User’s Profile, you hide it.

This is getting a little detailed for the comment section, but is really interesting. Maybe I’ll find some time to write a new post describing this scheme to subject it to more analysis. But it actually seems like there might be a chance to make this effective.

Thanks for the prodding on this!

Myk

Wouldn’t both the “thresholds” and the personal offensiveness categories emerge from the use of CF?

E.G., if I consistently rate nudity as a-ok and hate speech as very offensive, am I not giving you both my threshold and an explicit indication of what categories of content I consider offensive?

Even though the algorithm might not know whether a piece of content is nudity or hate speech, it knows whether to show it to me based on how it was rated by users whose ratings a similar to mine. Or similar to other users whose ratings are similar to mine. Or similar to other users whose ratings are similar to other users whose ratings are similar to mine…

I would be interested in hearing your thoughts on this.

adk

I really like the idea of making the filtering collaborative and dynamic like this. We’ve explored things like giving people a “knob” to indicate how offensive something is (like a scale from 1 to 10), and then letting them set their own “threshold” on that same scale.

You could even go so far as to let people classify why something is offensive, i.e., nudity, profanity, hate speech, etc. Internally, we could track these independent variables and allow our users to (actively or passively) let us know what their various thresholds are.

The problem with everything we’ve thought of along these lines is how to keep it simple enough so that people can understand it and will use it.

We _do_ already collect statistics about which pieces of content a user has rated, and how high they rated it. So we could do basically the same thing for a scale of “offensiveness.” Then, as you suggest, the user could be building their own “filtering profile” that helps the system determine what it should or shouldn’t show to them.

Thanks for the suggestions. We’ll definitely continue to think about this subject for a long time.

Myk

This would require keeping track of the content a user has been exposed to and their response to it. I.E., did they flag it as inappropriate, or perhapse, how inappropriate did they indicate it was.

adk