We use AI to save time for content creator to create great short videos and less worry about the manual work.
Combined state of the art speech recognition and unique data characteristics of the education industry and short video, we utilize data augmentation, grammar correction and text alignment technology to develop an end-to-end video speech to text algorithm.
The company has developed our IPs for scene recognition and tagging engine. By integrating UGC (user-generated content, eg. video, voice and text data), we have improved a 10 percent improvement of understanding video scenes. This helps us to build user interest maps, and automatically assign tags and categorize videos into educational themes so we can recommend videos based on user’s English language level and user interests.
Through user interest graph, we use natural language processing to build a mining engine to match a user with topics they might like. This provides a solid foundation for our recommendation system.
Compared with the industry's conventional recommendation system, the recommendation engine combines multi-dimensional data such as audio and video and text, user interest maps, and real-time interest analysis based on user flow. The recommendation engine achieved more accurate real-time interest matching and better content recommendation.