In the final quarter of 2017, Facebook removed 1.6 million pieces of content containing hate speech. In the second quarter of 2021, that number jumped to over 31 million. And that’s just one website during a single 3-month period.

Needless to say, hate speech and disinformation (referred to as “fake news”) are prevalent, and they are not going away any time soon. These comments exist in all communities, from social media networks and news sites to forums and blogs.

Wherever people are allowed to openly speak their minds, negative and hateful content can be found. 

The Problem of Hate Speech

In many countries, hate speech is a growing problem and one that creates unique challenges for media organisations. As we all know, publishers are responsible not only for their original content but also for the content their users generate on their websites.

Imagine, for instance, that you run a news website with a comments section that updates in real-time.

You post a news story about migrants or minorities and it goes viral. Great, you now have a story that could be seen by millions. But you will also have thousands of comments to deal with. Those comments could flood your page at a rate of 100s per hour.

But why can this be a problem? Comments are an essential ingredient to a published piece. They can easily shift the original message of the text and make the reader misinterpret the primary meaning. Large amounts of comments can change the narrative using expletives and hate speech. And let’s not forget that, you, as a publisher, are responsible for each comment and its content.

Most oftenly comments are used to predefine the narrative. Human moderators simply cannot read and verify every single comment real-time. This is a vulnerability that is exploited by interested parties.

It’s not feasible to check them manually. And even if you could do that, how would you prevent hateful comments from being posted in the first place? If you’re doing it manually, it means you’ll need to hire numerous people to work around the clock manually checking the comments.

It’s expensive, it’s time-consuming, and it’s impractical.

The result is that your viral news story becomes riddled with hateful comments. Not only will these offend your readers, but they could also harm the reputation of your brand.

After all, who wants to devote themselves to a community that is constantly flooded with hateful comments? Who wants to spend time on a website that fills with negativity every time a new story is published?

We’re not just talking about comments that can be perceived to be mildly inappropriate. They certainly exist, but “hate speech” typically refers to vile and obscene messages filled with hate and vitriol. No one wants to be greeted with that when they’re trying to catch up with the day’s news.

The Limited Efficacy of Existing Solutions

As every SEO and branding expert can attest, comment sections are essential for promoting a brand’s organic reach, building a strong community, and ensuring that engagement stays high. So, the solution isn’t to simply remove the comments section.

You just need to find a way of dealing with it.

Unfortunately, there are few existing systems in place for Bulgarian media outlets. It means that many businesses resort to ineffective methods.

Human intervention is one such option. But it’s simply not sustainable and few media outlets will consider it a viable use of their resources.

It’s also fallible. Unless those human moderators have been briefed on all slang terms and phrases, they can’t catch everything and some hateful comments will slip through the net.

There are automated solutions built into existing forums and comment sections. However, these are designed to detect certain words and phrases. They are effective if you want to keep obscenities out of your community, but they don’t work for hate speech.

A commenter doesn’t need to resort to using bad language to say something hurtful and insidious. By the same token, bad language can also be seen in many harmless comments.

The obvious solution is to use a model that blends automation and human interaction. And that’s where we come in.

The Identrics Solution

At Identrics, we use a human-in-the-loop hate speech detection model.

Our software checks the comments as they are posted and flags the ones that may contain hate speech. These comments are then sent for human moderation.

The human moderators are directed to the exact words that may contain hate speech, thus allowing them to make sound decisions.

It means that communities can benefit from the ease, simplicity, and speed of automation while still utilising the expertise that only human interaction can bring.

And that’s not all.

Our model is constantly learning and improving. If the human moderator determines that the flagged comment is perfectly harmless, they can send it back to the model.

The model learns why this comment should not have been flagged and knows not to flag such a comment in the future.

The longer the model remains active, the more comments it reviews and the more it learns.

As it grows, it becomes more effective at making these decisions and ensures that fewer false readings are sent for moderation.

What are the Benefits of Hate Speech Detection and Removal?

Chapter Three of the Bulgarian Criminal Code, “Crimes against the Rights of Citizens”, addresses the prevalence of hate speech and the need to eradicate it. As more of our time is invested in digital communities and more weight is placed on the importance of keeping them safe, AI will become essential.

Our hate detection model can keep communities safe, clean, and accessible. They combat the spread of hateful comments in otherwise friendly communities. Just as importantly, it can also ensure that companies remain compliant with regard to current and future laws.

After all, lawmakers rarely concern themselves with how the content gets there or if the site/platform can do anything about it. They want it gone at all costs, and may punish the platforms that fail to remove it. 

The Identrics hate speech detection model is easy to implement and will adapt to the community. It learns where those comments come from, what kind of content they contain, and whether or not they need to be removed.

Summary: Hate Speech Solutions

As noted in an article on ethics and responsible AI, Bulgaria plays a major role in the evolution of AI technologies. At Identrics, we take our work very seriously and are helping businesses across the country to utilise AI in ways that can improve their efficiency and bottom line.

Check out our hate speech detection models to learn more about how we’re changing the game.