Extracting Conversation Highlights using Dialog Acts
Online messaging platforms and virtual meetings have been a dominant mode of communication for many years now and more so in recent times. However, processing and condensing this data into useful snippets of information is still an on ongoing research problem. In this talk, I will introduce the concept of dialog acts to understand some semantics of these conversations and use them to highlight useful, actionable information. We will talk about dialog act datasets, models and their applications in different modes of conversation like multi-party meetings and chat messages.
*Meeting summaries are subjective and collecting annotations are expensive, but research shows participants find actionable-items to be key takeaways.
*Speech acts are used to understand the structure of a conversation and a subset of them can serve as a proxy for identifying these actionable-items.
*Pretrained models fine-tuned even on small datasets can achieve good accuracies in identifying speech acts that are useful for the task.
Varsha Embar is a Senior Machine Learning Engineer at MindMeld, Cisco, where she builds production level conversational interfaces. She works on improving the core Natural Language Processing platform, including features and algorithms for low-resource settings, and tackles challenging problems such as summarization and action item detection in noisy meeting transcripts. Prior to MindMeld, Varsha earned her Master’s degree in Machine Learning and Natural Language Processing from Carnegie Mellon University.