Recent advances in conversational agents, such as popular chatbots, have driven a fundamental shift in how people search for the information they need. People increasingly use modes resembling spoken conversations instead of relying solely on classic text queries. This shift in how people interact with conversational systems presents unique challenges, particularly in understanding the context of individual sentences within a conversation.

This innovative study reflects on the importance of adopting a knowledge-based perspective to enrich conversations and improve conversational search systems. Using pre-trained language models and commonsense knowledge bases offers a promising approach to improving the understanding of conversations and optimizing the performance of conversational search systems.

Many studies have explored methods to effectively convey the context of a discussion to each utterance, thus improving the retrieval performance of conversational search systems. Strategies range from identifying previously mentioned terms in the conversation to using sophisticated neural models to rewrite utterances.

This study proposes a comprehensive framework for managing multi-turn conversations by integrating common-sense knowledge. Their innovative approach uses pre-trained large language models and common-sense knowledge bases to enrich search queries with relevant concepts. The framework includes an entity generator, which extracts candidate entities based on the context of the conversation, and a selector that determines the most appropriate one that can be used to improve the current utterance and optimise retrieval effectiveness.

In addition to these advances, researchers have drawn on external knowledge bases to enrich expressions further and improve the search system’s overall performance. By incorporating common-sense knowledge, the researchers aim to overcome existing methods that focus primarily on enriching expressions at the term level. Their experiments demonstrate cases where relevant concepts, not explicitly mentioned in the conversation, can be inferred from the context, thus providing valuable information and improving understanding of the dialogue.

Furthermore, researchers have extended their approach by using broader common-sense concepts, which represent general phenomena or actions without requiring specific terms. This broader conceptual framework promises to improve the retrieval performance of conversational search systems.

This research significantly impacts people’s everyday lives by improving and streamlining their interactions with conversational and other information retrieval systems. Such advances can make a significant difference in today’s complex Internet landscape, where intuitive and efficient conversational interactions are paramount.