Session Overview

Uncovering Hidden Themes in Unstructured Data.

In this insightful session, Neha Singh takes us on a journey through the transformation of topic modeling, from traditional statistical methods to modern AI-powered intelligence. She highlights the challenge businesses face with the explosion of unstructured data—chats, social media, and surveys—which accounts for 85% of business data.

The talk explores how technologies like BERT and SBERT have revolutionized the field, enabling businesses to move beyond simple keyword matching to deep semantic understanding, extracting specific, actionable insights from mountains of text.

Key Takeaways & Concepts

Shift to AI-Powered Models: The industry is moving from traditional methods like Latent Dirichlet Allocation (LDA) to Context-Rich, Embedding-Based Models like BERTopic and Top2Vec for higher accuracy.
Handling Data Explosion: With the exponential growth of unstructured data, traditional models struggle with slang and domain shifts, necessitating advanced NLP solutions that require less preprocessing.
From Vague to Actionable: Unlike older models that produced overlapping topics (e.g., "delay, service"), LLM-based models identify specific issues (e.g., "Delayed delivery of spare parts"), enabling immediate business decisions.