Unveiling Bengali Text Summarization: Unsupervised Approach with BanglaBERT, Graph-based Ranking, and Interpretable Insights

Text summarization is a fundamental natural language processing task that deals with generation of condensed summaries from large textual data. This paper presents an unsupervised approach to Bengali text summarization, incorporating BanglaBERT embeddings with LexRank- a graph-based ranking algorithm. BanglaBERT captures semantic depth of sentences, which LexRank utilizes to rank sentences employing graph centrality and cosine similarity. Top-ranked sentences are extracted to form summary. LIME is used to ensure interpretability, providing human-understandable reasons for sentence selection. On a dataset of 8,000 Bengali news articles, the proposed method performs better than traditional methods such as TF-IDF and TextRank.