RFC: Keep an eye on offensive comments

[fgrehm]

follow up from telegram chat

As we all know, after serenata got a couple minutes of fame in the TV, the # of followers increased quite a lot. Along with that, I noticed quite a few offensive comments which could make deputies really angry and eventually get Rosie in trouble.

I think we should make an attempt to watch out for those comments and reply to people making it clear that this is only a suspicion and they are innocent until proven guilty.

Something along the lines of the following could be a great start:

:warning: Atenção: tenha em mente que o reembolso é apenas suspeito. Por favor, dê uma chance de resposta ao deputado :smile: :heart:

:warning: Heads up: keep in mind this is just a suspicious reimbursement. Please give the deputy a chance to reply :smile: :heart:

[miguelgraz]

Just throwing out ideas but a nice first step can always be a simple parser searching for famous big words, definitely not ideal but surely good enough to handle a nice percentage of the occurrences :smiley:

[Irio]

Great feature idea.

I'd prefer to have another Twitter account (other than @RosieDaSerenata) for doing so. Since Twitter doesn't seem to be very clear about rules for bots, I believe it's better to distribute the tweet rate between accounts. If the auto-reply account gets blocked, the original won't.

[fgrehm]

Just throwing out ideas but a nice first step can always be a simple parser searching for famous big words, definitely not ideal but surely good enough to handle a nice percentage of the occurrences

I personally think a check for "Ladrão", "Bandido" and "Corrupto" would be a great start since 99% of the time it is going to mean that the person is accusing the politician 😅 From there we could maybe have some sort of combination between "Sentiment analysis" and offensive words to trigger a reply. I'm just not sure if doing this analysis in portuguese is a problem already solved (haven't checked but there might be something out there)

I'd prefer to have another Twitter account (other than @RosieDaSerenata) for doing so. Since Twitter doesn't seem to be very clear about rules for bots, I believe it's better to distribute the tweet rate between accounts. If the auto-reply account gets blocked, the original won't.

Makes sense. I'll think about a nice name for that account so we can at least create it in advance. As food for thought, in my company's slack we have what we call a "Friendly Bot" that sends people messages when they join specific channels with additional information about that channel and the processes related to the specific teams.

Also, in order to stay "truly agile", I think we could have an MVP of this "using" a human for the work and measure the impact before automating the process. Or even a bit of both worlds and put together a simple dashboard that analyses replies to rosie and attempts to categorize them 💭

[anaschwendler]

I personally think a check for "Ladrão", "Bandido" and "Corrupto" would be a great start since 99% of the time it is going to mean that the person is accusing the politician 😅 From there we could maybe have some sort of combination between "Sentiment analysis" and offensive words to trigger a reply. I'm just not sure if doing this analysis in portuguese is a problem already solved (haven't checked but there might be something out there)

I would add "vergonha" and other shaming words, too

[rlage]

@fgrehm, I am not sure “solved”, but there are a few approaches out there. I saw IBM do sentiment analysis in the World cup, using AI and a manually labeled database to train a model