This paper explores the potential of large language models (LLMs) as an alternative to traditional topic modelling methods. Traditional approaches like LDA have limitations, including a lack of semantic understanding and overlapping topics. The authors propose a new approach using LLMs to extract topics from extensive text corpora. They introduce a framework that prompts LLMs to generate topics and establish evaluation protocols to assess the effectiveness of LLMs in topic extraction. Their findings show that LLMs with appropriate prompts can generate relevant topic titles and adhere to human guidelines for refining and merging topics. Through experiments, they summarize the advantages and constraints of using LLMs in topic extraction. The paper discusses the limitations of traditional topic modelling and close-set topic classification, highlighting issues such as the need for predefined topics and the inability to capture unseen topics. LLMs, on the other hand, offer the ability to understand context, nuances, and subtle thematic undertones, allowing for more detailed topic categorization. They also adapt to evolving language trends and emerging topics, ensuring that topic modelling remains relevant and up-to-date. The authors evaluate the performance of two LLMs, GPT-3.5 and LLaMA-2-7B, against traditional methods like LDA and BERTopic. They conduct experiments to assess the effectiveness of LLMs in topic extraction, including the use of seed topics and summarization techniques to refine the generated topics. The results show that LLMs can produce high-quality topics with appropriate prompts and manual guidelines. They also introduce evaluation metrics to assess the quality of topics generated by LLMs, suitable for both labelled and unlabelled datasets. The paper also presents a case study on the temporal analysis of vaccine hesitancy, demonstrating the ability of LLMs to extract and summarize topics over time. The results show that LLMs can effectively extract topics and generate explanations for analysing dynamic datasets. The authors conclude that LLMs offer a viable and adaptable method for topic extraction and summarization, with potential for further development in handling larger datasets and improving evaluation protocols.This paper explores the potential of large language models (LLMs) as an alternative to traditional topic modelling methods. Traditional approaches like LDA have limitations, including a lack of semantic understanding and overlapping topics. The authors propose a new approach using LLMs to extract topics from extensive text corpora. They introduce a framework that prompts LLMs to generate topics and establish evaluation protocols to assess the effectiveness of LLMs in topic extraction. Their findings show that LLMs with appropriate prompts can generate relevant topic titles and adhere to human guidelines for refining and merging topics. Through experiments, they summarize the advantages and constraints of using LLMs in topic extraction. The paper discusses the limitations of traditional topic modelling and close-set topic classification, highlighting issues such as the need for predefined topics and the inability to capture unseen topics. LLMs, on the other hand, offer the ability to understand context, nuances, and subtle thematic undertones, allowing for more detailed topic categorization. They also adapt to evolving language trends and emerging topics, ensuring that topic modelling remains relevant and up-to-date. The authors evaluate the performance of two LLMs, GPT-3.5 and LLaMA-2-7B, against traditional methods like LDA and BERTopic. They conduct experiments to assess the effectiveness of LLMs in topic extraction, including the use of seed topics and summarization techniques to refine the generated topics. The results show that LLMs can produce high-quality topics with appropriate prompts and manual guidelines. They also introduce evaluation metrics to assess the quality of topics generated by LLMs, suitable for both labelled and unlabelled datasets. The paper also presents a case study on the temporal analysis of vaccine hesitancy, demonstrating the ability of LLMs to extract and summarize topics over time. The results show that LLMs can effectively extract topics and generate explanations for analysing dynamic datasets. The authors conclude that LLMs offer a viable and adaptable method for topic extraction and summarization, with potential for further development in handling larger datasets and improving evaluation protocols.