BERTopic is a topic model that extends the clustering approach by using a class-based variation of TF-IDF to generate coherent topic representations. It first creates document embeddings using pre-trained language models, then reduces the dimensionality of these embeddings before clustering them. Finally, it generates topic representations using a class-based TF-IDF procedure. BERTopic generates coherent topics and remains competitive across various benchmarks involving classical models and recent clustering approaches. The paper introduces BERTopic, which leverages clustering techniques and a class-based TF-IDF to generate coherent topic representations. It first creates document embeddings using a pre-trained language model, then reduces the dimensionality of these embeddings before clustering them. Finally, it generates topic representations using a class-based TF-IDF procedure. This three-step process allows for a flexible topic model that can be used in various applications, including dynamic topic modeling. BERTopic uses document embeddings generated by a pre-trained language model to create representations in vector space that can be compared semantically. These embeddings are used to cluster semantically similar documents. The dimensionality of the embeddings is reduced using UMAP, and the clusters are generated using HDBSCAN. The topic representations are then extracted using a class-based variation of TF-IDF, which allows for a more accurate representation of topics. BERTopic is compared with other topic models such as LDA, NMF, CTM, and Top2Vec. It is shown to perform well in terms of topic coherence and diversity. Additionally, BERTopic is used for dynamic topic modeling, where it can model how topics might have evolved over time. The results show that BERTopic performs well in both static and dynamic topic modeling scenarios. The paper also discusses the strengths and weaknesses of BERTopic. It is noted that BERTopic assumes each document contains a single topic, which may not always be the case. Additionally, the topic representation itself does not directly account for the contextual nature of the documents. However, BERTopic is flexible and can be used with various language models, making it a versatile tool for topic modeling.BERTopic is a topic model that extends the clustering approach by using a class-based variation of TF-IDF to generate coherent topic representations. It first creates document embeddings using pre-trained language models, then reduces the dimensionality of these embeddings before clustering them. Finally, it generates topic representations using a class-based TF-IDF procedure. BERTopic generates coherent topics and remains competitive across various benchmarks involving classical models and recent clustering approaches. The paper introduces BERTopic, which leverages clustering techniques and a class-based TF-IDF to generate coherent topic representations. It first creates document embeddings using a pre-trained language model, then reduces the dimensionality of these embeddings before clustering them. Finally, it generates topic representations using a class-based TF-IDF procedure. This three-step process allows for a flexible topic model that can be used in various applications, including dynamic topic modeling. BERTopic uses document embeddings generated by a pre-trained language model to create representations in vector space that can be compared semantically. These embeddings are used to cluster semantically similar documents. The dimensionality of the embeddings is reduced using UMAP, and the clusters are generated using HDBSCAN. The topic representations are then extracted using a class-based variation of TF-IDF, which allows for a more accurate representation of topics. BERTopic is compared with other topic models such as LDA, NMF, CTM, and Top2Vec. It is shown to perform well in terms of topic coherence and diversity. Additionally, BERTopic is used for dynamic topic modeling, where it can model how topics might have evolved over time. The results show that BERTopic performs well in both static and dynamic topic modeling scenarios. The paper also discusses the strengths and weaknesses of BERTopic. It is noted that BERTopic assumes each document contains a single topic, which may not always be the case. Additionally, the topic representation itself does not directly account for the contextual nature of the documents. However, BERTopic is flexible and can be used with various language models, making it a versatile tool for topic modeling.