Minutes: Hybrid Text Summarizer for Online Meetings
Aaditya Mahadevan, A Ananthakrishna Pillai, Jigyassa Lamba · 2023
Online meetings have become commonplace, making it difficult to keep up. A summarizer for meetings, that captures the meeting content concisely, can prove to be beneficial. Meeting Summarization involves two key modules: Speech-to-text conversion that extracts transcripts from the meeting audio, and a hybrid text summarization model, incorporating extractive and abstractive summarization approaches. Extractive summarization was implemented using a centroid-based method that utilizes the compositional capabilities of word embeddings to capture the semantic relationships between sentences. The extracted sentences are fed to an abstractive model which uses the BART model as a general baseline. The BART model was fine-tuned on a dataset and optimized to generate better quality summaries. The T5 model has been used to generate one-line abstractive summaries. The models were evaluated using quantitative and qualitative testing. The summarizer thus assists the user to run through the important aspects of the meeting at a later stage.