Recent Awards

Recommendation systems are vital to keeping users engaged and satisfied with personalized recommendations in the age of information explosion. Ranking is at the core of recommendation systems by presenting users a personalized ranked list of recommendations. On the other hand, most practical recommendation systems deal with large amounts of natural language data and hence an effective ranking system requires a deep understanding of text semantics. In this project, we will propose novel deep learning based models to integrate natural language information with traditional dense features. We will utilize contextual embedding such as BERT to enhance contextual modeling. In addition, we will address the scalability of the proposed deep learning models as practical systems are often involved with a huge number of users and items. Knowledge distillation and model pre-training will be exploited. We will also explore few-shot learning to tackle the cold-start problems in recommendations.