Recommendation Systems in the Era of LLMs
Ipsita Mohanty · 2023
Recommendation Systems (RecSys) in e-commerce play a pivotal role in enhancing the user experience by providing personalized recommendations of products specific to the user preferences. The power of RecSys lies in its ability to anticipate user needs through signals like historical purchases, tastes, preferences, similar user types, etc. These factors drive customer satisfaction, user engagement, revenue growth, etc. Since the advent of transformer-based embeddings, RecSys has been primarily driven by BERT-based architectures. However, in the era of Large Language Models (LLMs), interest in applying LLMs in the RecSys domain has gained significant attention. LLMs trained on massive amounts of data have the potential to enhance various aspects of RecSys through techniques like fine-tuning, prompt tuning, etc. However, these techniques are not without challenges for deployment in real-world applications. This talk discusses the potential use of LLMs in recommendation systems and challenges in real-world applications such as training efficiency, inference, and bias.