I agree if a person enjoys bullying, they are unlikely to self-correct. But an "interaction sentiment tool" makes it easier for the community to spot these people, and look more closely into what they are doing. Then try to get them to change, and <rinse and repeat> until such time as they ban them.

My comment about self-correcting behaviour is about people who don't intend to be a bully but behave abrasively without realising it. We have a lot of battle-weary editors out there who have just seen one too many vandalism, one too many blatant self-promotional article, etc and they become inclined to just shoot down "yet another" with increasing reluctance to check out the merits of the specific case, or to be terse and unhelpful in a Talk message etc. We've probably all had those moments of finding some new user's contribution that needs so much work to improve and thought "I'm just too busy, I don't have time to educate yet another one who probably won't stick around anyway, I'll just delete it and move on". I believe that most of our community does not intend to be a "bully" but may not be aware that is how they might seem to others at times. Letting people be aware that their interaction style is exhibiting higher than average "negative sentiment" *is* likely to change the behaviour of that group.

Obviously if we were to put such a tool out there, I'd suggest adding some general advice about what you might do if your score is "pretty negative", e.g.

* think about the choice of words you use, don't use words like ..., instead use ...

* are you terse or just point to a policy without being specific about your concerns

* could you have suggested a solution rather than just pointing out a problem?

* is it time for a wiki-break to recharge your batteries?

The sentiment score is likely to be generated from assessment of a number of elements of the observed interactions, so, for an individual looking at their score, it might be possible to make specific suggestions based on specific component scores, e.g. pointing out specific “abrasive” words being used regularly and suggesting alternatives.

Here’s a suggestion for something a lot simpler than the “international sentiment tool”. Just produce some word clouds for:

· a user’s edit summaries

· a user’s edits on article Talk pages

· a user’s edits on other people’s User Talk pages

· a user’s edits on their own User Talk page

What does that show us about people?

Kerry