Sentiment analysis is historically one of the hardest problems to solve in data science. From the early days of rudimentary "bag of words" counting to grammatical state machines, the industry has long struggled with the "context problem." The fact that language is nuanced, sarcastic, and constantly evolving. However, the social listening industry has stagnated in this space. Even with new machine learning and AI techniques, many of our competitors have not attempted to improve sentiment and emotion in 10+ years!
It’s for these reasons that Infegy never stops improving our models and platform. We are happy to announce our largest improvement to our sentiment model in years: Infegy IQ 4.0!
Growing far beyond its initial scope, Infegy IQ 4.0 is a massive system update designed to address long-standing feature requests, conquer difficult context challenges, and bring unprecedented accuracy to audience intelligence. Here is everything you need to know about the new era of Infegy IQ.
Infegy IQ 4.0 represents a definitive shift away from difficult-to-maintain, dictionary-based models. Instead, it relies on a purely AI-driven machine-learning system that learns from its mistakes and adapts without hand-coded rules.
The biggest driver of this upgrade is our transition to our own Infegy API for training data. The same one available to you! By utilizing our own massive social dataset and API, we've increased our sentiment training data volume by a staggering 21x. Infegy API's advanced query capabilities allow us to pull highly accurate neutral text samples and utilize random sampling to ensure our models learn from a diverse, unbiased, and truly representative cross-section of internet data, including spelling errors, slang, and everything in between.
Thanks to this massive influx of high-quality data, our new models are more than six times as large and boast an exponentially larger vocabulary. We've also integrated numerous model enhancements, such as a new attention mechanism, allowing the AI to dynamically focus on what matters most in a sentence, separating signal from noise. We’ve also incorporated reinforcement learning, enabling the models to continuously learn and improve from their mistakes.
The numbers speak for themselves. Infegy IQ 4.0 demonstrates spectacular performance improvements over our previous production models:
We are proud to announce that Infegy IQ 4.0 now fully supports sentiment and emotion analysis in Hindi and Urdu. This development was driven by our ability to train models exclusively on social data using Starscape refinements. Indexing for Hindi search is already underway, and with Starscape already supporting topics, narratives, and AI summaries for Hindi, these additions push our total collected data language coverage to 90%.
With Infegy IQ 4.0, we are also introducing a new Universal Adult Content Detection Filter, the first universal language model created to detect NSFW content across all languages simultaneously.
Trained on content linked to the world's top adult sites, this model boasts a 94% detection rate and an incredibly low false positive rate of just 1.6%.
Rather than quietly deleting this data, we are providing it as a query filter. This gives users the power to either exclude adult content from their feeds entirely or specifically filter for it. Allowing brands to accurately measure what percentage of their brand conversation is associated with NSFW topics.
We couldn't be more excited to get Infegy IQ 4.0 into your hands. This isn't just an update; it's a fundamental reimagining of what AI can do for social listening and audience intelligence. Get ready to see your data clearer, faster, and more accurately than ever before!