The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Topic Modeling for Customer Insights at USAA with William Fehlman - TWIML Talk #276
Jun 20, 2019
William Fehlman, Director of Data Science at USAA, shares his expertise in topic modeling and natural language processing. He explains how USAA leverages these techniques to enhance customer service by analyzing both structured and unstructured data from communication channels. Fehlman delves into methodologies like latent semantic indexing and non-negative matrix factorization, discussing their effectiveness in uncovering customer insights and optimizing operational strategies. Insights on coherence scoring and term frequency provide a deeper understanding of the topic modeling process.
AI Snips
Chapters
Transcript
Episode notes
From Computer Vision to NLP
- William Fehlman's background in computer vision prepared him for his current work in NLP.
- He uses similar mathematical principles in both fields, treating images and text as matrices.
Topic Modeling for Customer Insights
- USAA uses topic modeling to analyze large volumes of unstructured and structured data from various customer interaction channels.
- This helps them gain insights to better serve their members and uncover emerging topics.
Proactive Disaster Response
- USAA uses topic modeling to identify emerging topics in chat conversations, such as those about natural disasters.
- This allows them to proactively address member concerns and prepare resources before call volumes peak.