Extracting Abstractive Summaries Through Generative AI Models
Rimsha Khan, Shanu Sharma, Divya Upadhyay · 2025
Text summarization is an advanced domain in the natural language processing (NLP) field, to shorten a large amount of text into meaningful condensed summaries to get more enhanced information. While generating summaries, maintaining the intact of the original text is very important. Furthermore, to generate the original summaries from the text, i.e. for abstractive summarization, the inclusion of neural networks and generative AI is highly required. Considering the importance of text summarization in different fields, and the advancements in generative AI models, this paper seeks to analyze the capabilities of generative AI models i.e., Mistral-7B, LLaMA3-8B, and BART-large CNN for text summarization. With tools such as LangChain and Hugging Face pipelines, the said models have been integrated to produce abstractive summaries. This study uses the DialogSum dataset, along with Groq & Google Colab experimental frameworks. Well-known evaluation metrics like ROUGE, BLEU, and BERT Score further strengthen the comprehensiveness of their evaluations of these models. The findings suggest that Mistral-7B was found to have the best performance among others. This paper also presents the capacity of different models and their ability to meet the desired output of summarization, which further facilitates text summarization-related tasks in different industries.