Kawlo: Spam filtering with deep learning

Spam emails are a persistent threat to users' privacy and security. They are unwanted messages that are sent to a large number of recipients, typically for advertising purposes. As online learning platforms have become increasingly popular, they have also become a target for spammers.

Kawlo is a learning platform that has been the victim of spam on online learning platforms, and UlifeAi joined kawlo in the fight again Spam detection online.

Spam messages have been a nuisance for users and have caused significant harm to their learning experience. Some messages try to tick students to pay for fake courses or contact them for items and trick them into giving the spammers money or payment details.

Students on the platform need to be able to focus on their studies without being interrupted by unwanted messages. To combat this problem, UlifeAi has developed a powerful spam filtering system that uses deep learning technology to clearly define the line between legit and spam messages.

Kawlo

kawlo is an online learning platform with a cumulative user base of hundred of thousand of monthly users. It is a forum for students’ questions and gives them all the resources they need to succeed in their learning journey. It principally works in Africa, Uk, and India.

The solution and challenges

Using deep learning technologies, we have crafted an evolutionary algorithm for spam detection fully customized for the platform use cases.

The data collection challenge.

One of the big challenges in crafting a solution for kawlo was the data. Since 2016 the platform has been deleting most of the spam comments on their different platforms to preserve users’ safeness. But for a machine learning model. that is not good news, unfortunately. Because the model needs data to define the line between good and bad content clearly. So the first thing we did, was to prepare a proper data collection scheme and get all the necessary information from the existing model to browse and check for solutions. We then check the data balance and confirm the unbalanced hypothesis. We have then built an amazing data augmentation algorithm to simulate the writing behaviours of spammers and write more text that actually looks like what a scammer could have done. And with that method, we have been able to balance good and bad messages.

crafting the solution

• We have created a model that can be easily integrated into the existing Nodejs and Mobile system, and that is continuously assessing the data.
• We have to build a fast and easy-to-integrate API for the platform, so they can solely focus on their core business.
• We are providing continuous support in terms of model amelioration and integration.