Bhargav Rao edited this page Jan 21, 2019 · 1 revision

Some quick background

HeatDetector runs its NLP on two feeds (bad vs good); it calculates the probability that a comment matches these feeds. If the probability versus the bad feed is high enough it get reported in chat.

The feedback fp and tp (and report command, now fn) was initial designed to know which comments I should add in different feeds. The initial idea was that I would only send feedback if I wanted it to be added to a feed, fp to good feed and tp to bad feed. The ones with no feedback I would simply ignore.

However I realized over time that the feeds are of fundamental importance for the NLP system (we did different sessions to change feeds, meta questions etc) and I started to use the feedback only as indication (filters in query) to find what to add. It's simply too difficult to know in chat if a comment should actually be added to feed.

The new system has been designed to be more easy to use in chat; it will give indications about how a user sees the comment. This indications can be used both for statistical reasons and also to filter out comments that can be included in new feeds.

The feedback types

  • tp = True positive: this comment is offensive (not within the be-nice policy), I'm flagging "It contains harassment, bigotry, or abuse", you guys should consider to included it in bad feed.

By the time it took you to write that useless comment, you could have read What topics can I ask about here. Which leads me to believe that you're either a dick or you can't read, or both

  • fp = False positive: this comment is not offensive (within the be-nice policy), you guys should consider to included it in good feed

Thank you. Real stupid from me to make such a beginner fault

  • nc = Non constructive1: It's not a constructive comment and it seems also unfriendly but I'm not really sure it's offensive. I flag it as "It's unfriendly or unkind." and maybe it should be included in bad feed.

1 This is the old name of the flag

you are kind of wasting other people time by changing your mind and focus in the middle of poorly formulated question.

  • sk = Skip: I'm not sure what this is (or this is useful but we are tracking it with regex), I would not include it in any feed. (Don't be afraid to use this, it's a very useful indication, we got tons of comments and having possibility to filter away some comments is important.)

Your answer is now plagiarized from

This is spam

  • fn = False negative (through report): this is an abusive or unfriendly comment that I found and flagged either "harassment, bigotry, or abuse." or "unfriendly or unkind.", it needs to be included in the bad feed.

This loser seems to think I'm here for points. lol

Remember that comments are not like NAA posts (clear guidance). Different people will find different things offensive, so often there is not a correct nor incorrect feedback. You can not actually tell a user "That you should have indicated as..", only "I feel that it should be indicated as...". With this in mind do not try to correct other users' feedback (as long as you are confident that they understand the rules), but simply send you own. The new Hydrant system will contain both.

#The future

My idea is to bring HeatDetector out of Queen's code and create an api instead. Within this api I will introduce a tracking concept for regex like "spam","plagiarism", "kolcak". Hence comments may be reported because something is being tracked even if it is not related to "offensive", this to simplify the "Heat" concept that is hard to understand and explain. Related to this api (or Higgs) my idea is to build a web dashboard to finally manage feeds (review, included, exclude, order etc).

Clone this wiki locally
You can’t perform that action at this time.
You signed in with another tab or window. Reload to refresh your session. You signed out in another tab or window. Reload to refresh your session.
Press h to open a hovercard with more details.