Tisane API - Detect hate speech, cyberbullying, and more in 27 languages

@owenfar1 Thanks! We might be a small country, but I do get the feeling that we do find our way around here!😀😇

@owenfar1 Yes, indeed, that is how we are😉Happy hunting and upvoting😎

@ana_topoljski Hi Ana! Sorry, no Slovene yet. Having said that, the effort of adding a new language is within a scope of a small to mid-size project, so we'll be happy to start a conversation and evaluate your needs / effort. You can go via our Contact Form or directly to me (vadim dot berman at tisane dot ai).

@vadimberman Thanks for the information. And kudos for the launch.😊

@stewartjarod Yes, of course, it's an inevitable (and a very good) question. But note that this is an API, a tool. Pardon the gruesome analogy, but it is as close to censorship as a kitchen knife to murder. However, we do have responsibility to encourage good practices and prevent misuse. We need to define solid criteria what the hate speech means, and some people would find ours lax, but very few people would find it disputable. Here is our criteria: http://tisane.ai/knowledgebase/b... . Note that an attack on a majority demographic group is also an attack. Note that we do not strive to detect even "mostly universally offensive opinions" (e.g. "Hitler did nothing wrong"). It's both technically challenging and bound to generate arguments. We might create that capability in the future for customised installations but that will not make it into the core, publicly exposed API. In general, if you're curious about the subject, you might want to read the article we posted about the advanced types of sentiment analysis: https://medium.com/tisanelabs/ev... (the last part, about politics). We were also asked about how to handle culture-specific exceptions (e.g. in a country with shariah laws, "women should not be allowed to drive" might not be considered bigotry); the simple answer is, the calling application receives annotations, but it may ignore specific cases (plus, again, the system comes with extensive customisation capabilities). The personal attacks are easier. "Your opinion sucks" is not an attack; "you suck" is an attack. As for the technical implementation, at this point, we'd rather not talk too much about it. Let's just say the history of the R&D is far longer than the history of Tisane Labs.

@vadimberman I understand immigrants being protected, but with the line being blurred between a legal and illegal immigrant what’s the definition in this case?

@leachy114 An interesting nuance, but no distinction. On the technical level, there is no way we can go over every statute and law all over the world and configure the system to distinguish "good" and "bad" immigrants. Then some people might say, that visa makes immigrants not cool and this visa is OK (e.g. "trailing spouses", refugees, etc.). There is no way we'll want to or be able to get into that. (Again, I'm going to send you to the article in Medium which explains that in the political sentiment analysis, "good" and "bad" is relative to the reviewer's set of beliefs.) On the practical level, I don't think it'll make a huge difference. If someone says, "all illegal immigrants are thieves / stupid / etc." that means assigning a negative attribute to an entire group, so while it might be slightly outside of the intended scope, I'll be fine if it's tagged as hate speech. An utterance like "we shouldn't let illegal immigrants in" will definitely **not** be tagged as "hate speech" (and so "we shouldn't let legal immigrants in"), so I think that should be OK with everyone. As the Dude put it, "Yeah, that's, like, your opinion, man" :) . Regardless of the technical obstacles - which are not trivial, may I say - we don't view our mission as making people to adhere to some kind of views or beliefs (and given the fact that we handle so many languages, it's plain impossible). We merely want the Internet to stop being Wild West and start being a modern city where people don't attack each other on sight.

@vadimberman thank you for the thoughtful reply. 👍I am definitely gonna dig into your blog a bit. 😄 Essentially, APIs don't kill free speech, API integrators do.

@stewartjarod That's a nice one 😉 .

@rajagopalilango Thanks and unfortunately, no. We only deal with text, but I'm sure there are services to detect nudity and offensive gestures.

@rajagopalilango @vadimberman Not Hotdog. 🤭

@recon Thanks, Sam! There is, of course, lots of room for improvement, and like any other product, it will improve over the initial level. The general idea with the abuse detection though is to pick enough to deter potential offenders; just like the police does not aim to catch every criminal, there will be always someone to slip through the cracks. We don't aim to reach the "clean room" level, more like to help make the discourse more human.

@nooruzzonline Thanks and yes, of course. But we're in the API business, so we simply collaborate with the existing comment engines (some reached out to us after this ProductHunt posting!). Just like the functionality relevant in the marketing space: Synup where you work can integrate the opinion analysis.

@vadimberman True. We're planning to build a tool internally and I guess this will make my life a whole lot more easier.

@nooruzzonline Not going to hard-sell, but you're more than welcome to explore, and we can direct you in the proper direction. With the generous free plan, it will cost you nothing to experiment!

Hi David, thank you for your comment. We're sorry it did not work for you; we fixed your issue and deployed the update, like promised. We apologise for the delay and the fact that Singapore and Denmark are so far apart in terms of time zones. As for the automatic language detection, this is in the roadmap but it is rather dangerous to use on very short messages. Think of a word like "status"; it's the same word in English, Spanish, and many other languages. If you have this sole word as an input, what language would you put there?

@geeblack_ sign up to the API, then simply use it on the content your users post. As for the business logic, this is up to you. You can either invoke it before posting (e.g. warning the user if they transgressed your policies), or forward the suspicious ones as alerts.