Unequalled performance in toxic content moderation
Founded by globally-renowned NLP expert Professor Lucia Specia and her world-class team, Contex.ai is the first content moderation solution to understand written, visual and audio engagement in a fully contextual manner, emulating human understanding in detecting toxicity.
The result is transformative technology covering text, images and spoken chat, beating state-of-the-art in content moderation tools by unprecedented margins.
THE CHALLENGE
Toxic content is a major issue in our society, damaging lives and businesses. Social media companies spend billions every year deploying human moderators to filter toxicity manually. Why? Because conventional AI technology is unable to moderate content in context and fails to detect toxicity at acceptable levels of accuracy to automate the process.
Toxicity detection is a multidimensional problem - multiple factors beyond the content of the post drive human perception of toxicity.
Contex.ai uniquely addresses this challenge via a breakthrough in multi-modal Artificial General Intelligence
OUR SOLUTION
Contex.ai has developed the world's first automated technology solution to understand social media posts in the full visual and conversational context required to emulate real-world, human understanding.
The company’s multimodal contextual cooperative deep learning algorithms learn from highly heterogeneous and complex signals including the post, its context, and external cues such as emotion, age, demographics etc.
Contex.ai is also the only solution that provides explanations for model decisions - an essential enabler of transparency in content moderation.
Our suite of moderation solutions:
Contex.ai’s platform delivers unrivalled performance compared to Google, Microsoft and the other tools, as demonstrated below in benchmarks across 20k+ tests and 7 data sets:
Accuracy vs F1 (Toxic) for 7 datasets: D-Safety (Facebook), Dialogue-BOT (Facebook), GAB, Reddit, Misogynous Memes, Dota2 game (public) and conteX own. Test date Feb 2022. Accuracy = proportion of predictions that are correct and F1 = aggregated view (harmonic mean) on false positives and false negatives (for both: the higher, the better).