Bleepr

Flagging hate speech and propaganda on Twitter using AI

1 follower

Flagging hate speech and propaganda on Twitter using AI

1 follower

•

Bleepr uses natural language processing (NLP) to automatically flag hate speech published on social media. In our first release, we scan Tweets of users with more than 10k followers, automatically label for 7 dimensions of unsafe language, and update daily.

Launch tags:Developer Tools•Artificial Intelligence

Launch Team

AWS Activate CreditsCredits that fuel your startup’s next milestone

Promoted

Bleepr

Maker

We launched Bleepr to hold social media platforms accountable to the removal of harmful, hateful and propaganda content. We uncover the bleeps posted on social media - many of which are about the current U.S presidential election, with the help of AI. The public is told that the platforms are doing a lot of work to take down abuse, but have few ways of knowing if it is working or not apart from what the platforms self-report. The internet also mainly focuses on the terms of service violations that make the headlines. The purpose of Bleepr is to focus more on the less popular cases of abuse which still exist but are not being highlighted by anyone. How it works We scan tweets in bulk every day. We look at around 500 tweets in more detail - and only those which were posted by twitter profiles with more than 10K+ followers. We run our ML algorithms on those tweets. Every tweet which has at least one of the automated labels ((sexism, racism, hate speech, obscenity, political bias, toxicity, insults and threat level)) with a score above a certain threshold is then listed on bleepr.ai. We highlight these labels in red. Our AI is trained using expert annotators from communities like the International League against Racism and Anti-Semitism (LICRA) and Jugenschutz in Germany, along with other specialist anti-abuse advocacy groups. Known issues - Sexism detection flags up higher than it should sometimes for words like ‘bi**h’ and ‘motherf****r’ even in contexts which are not hateful. - There are a few bleeps which contain swear words but are random and obscure and not truly hate speech - Sometimes we flag news stories about controversial topics like homophobia, paedophile or sexual abuse, rather than original hate speech - We sometimes flag Tweets don’t exist anymore, which means Twitter has luckily already taken them down or authors have removed them. - Sometimes we flag Tweets that talk about political entities in a fairly coherent way, but use the words idiot a lot Future releases include using human curation of Bleeps using a small curated QA community, adding more social media platforms to Bleepr, improving the models, and having live refresh of likes/shares/retweets. Help us make social media a safer space. You can do this by clicking on the AGREE/DISAGREE WITH AI buttons which are shown under every bleep on bleepr.ai. We will record your anonymous feedback and use it to make our algorithms better. If you would like to contribute to our algorithms or join our community to take Bleepr to take it further, please contact us on info@factmata.com. We are just getting started.

Report

6yr ago

@factmata @dghulatifactmata sounds cool tho . i am more curious about this serving price

Report

6yr ago

Bleepr

Maker

@factmata @dghulatifactmata @pondechan this is free for now, but we do charge for the API underlying it for enterprise customers at around $0.05 per post going down to $0.01 for high volumes.

Report

6yr ago

@factmata @dghulatifactmata em that's lota cost

Report

6yr ago

@factmata @dghulatifactmata How can u say with a straight face that the word "motherf***er" is used in contexts that are not hateful?

Report

6yr ago

Bleepr

Maker

@factmata @dghulatifactmata @jay_bienvenu For example, what about a caption "Check out this little motherf***er! So cute" for an image of a pet dog stealing a treat ? Or a rap lyric? Or a film transcript?

Report

6yr ago

Any time I see something or someone claiming to flag "propaganda" or "misinformation," I always have one question: What are you doing to make sure those flags do not perpetuate political bias and bigotry? For example, you chose to highlight a tweet demonstrating bigotry against two "liberal" individuals, but you don't have one highlighting bigotry against a "conservative" individual. I don't tolerate the former any more than the latter but as a "right-winger" I'm not going to trust a product that can't demonstrate and ensure impartiality.

Report

6yr ago

Bleepr

Maker

@jay_bienvenu the technology does not distinguish between who is writing the Tweet. Other AI in research in this area does use user signals and metadata to train the AI. We have never liked this approach for the reason you correctly highlight. No matter who you are, if the content you post fits the pattern of figurative phrase and semantic structure our ML has learned to flag, it will be picked up. We several other examples of anti-conservative bigotry being flagged, feel free to scroll through old Bleeps found.

Report

6yr ago

Quite sad how many people are in favor of this and other similarly censorious (borderline authoritarian, even...especially once someone decides to implement anything decided via AI algorithms with automated instantaneous banning/removal) so-called "products" on ProductHunt lately. Seems to me like it's less an actual product most of the time, even less often a viable product/service with a realistic/legitimate business model, and instead often just a means to signal to others that you align with the political beliefs or social opinions being expressed in the promo materials or used as examples; sometimes tacitly, other times rather blatantly. The main issue I have with this however, isn't really with whether or not it's censorious (if people want to censor themselves/block their own access to information that others choose to retain their freedom to read/stumble across, that's totally fine by me) but with the whole idea of an AI system/ML-trained algorithm/bot becoming the arbiter of what's deemed 'misinformation', 'hate speech' or 'propaganda' -- all three of these being just a few such nebulous terms with completely subjective definitions depending upon factors like who it is that might be writing the material, the context that it's being written in, etc. What should be most alarming however are the claims of being able to flag items as containing 'misinformation' or 'propaganda', a claim which on it's face should be considered specious at best, and one which I will preemptively consider to be totally bogus until person(s)/company making said claim can definitively prove otherwise. There is no chance that with the state of current ML technology - which we know suffers from huge issues that stem from the inherent biases of the data sets provided for training the algorithms as well as the individual team members who work on refining them who all bring their own personal biases to the table. The innumerable examples of so-called 'fact-checker' websites, 'debunking experts' and so on+so forth getting stories completely wrong, almost always when it's got a political angle involved (Snopes is a great example of this, same with Politifact) shows that this already wrought with problems. Even more nefarious than the (frankly, quite dumb/childish - IMHO anyway) 'hate speech' issue, this is something that could have very real and serious impacts on things like elections, financial markets, even matters of life/death. Besides, I don't understand the potential customer for this anyway. Those API rates are absurd. Surely it's not going to be individual users paying for this, and legacy media is dying faster than...something that dies quickly, I dunno, use your imagination. Point being, I don't see where any demand for this product really exists. It's a sad state of affairs when you have a small group of people who seem to be very gung-ho on establishing themselves the judge of what is 'hateful', 'toxic', 'offensive' or construes propaganda, misinformation, etc. when it seems to me that the platforms like twitter themselves are already rather overzealous in their content moderation policies. Twitter has introduced a number of features that essentially enable a user to pretty much reduce the chances of coming into contact with alleged 'hate speech' or whatnot should they choose to configure their settings to do so. But I have a real problem with the idea that the internet needs to become a de facto 'policed' state where organizations like the ones who donate to Bleepr's development can exert influence over what we are allowed or not allowed to say or read. Best of luck to you anyways, I just wanted to share my thoughts on the matter and hopefully this is taken as constructive criticism and not as an insult because it is not meant that way, though I think this is clear by the tone of my post.

Report

6yr ago