Retrieval Augmentation: A Promising Solution for Enhancing Chatbot Performance

The impact of conversational agents on human-computer interaction is deemed paramount, especially considering the increasing presence of chatbots in customer service, education, and healthcare domains. Even though there still is a more classical way of designing chatbots that has been using pre-trained language models only, such approaches often cannot guarantee factual correctness and contextual relevance, mainly in those knowledge-intensive scenarios. The newly presented Retrieval-Augmented Generation (RAG) models seem to emerge as an attractive alternative, combining external knowledge sources with response generation. This review covers the developments that have taken place in RAG-based chatbots, their methodologies, applications, metrics to evaluate performance, and ongoing challenges. Based on synthesizing key research insights, this paper lays its focus on how RAG models unlock the development of informative, accurate, and contextually aware conversational agents.