AutoRegressed on the Road: American Family Insurance Analytics Forum
Tim Dobbins and Chris Morgan recently had the opportunity to present their work on Chat Topic Modeling at the annual American Family Insurance Analytics Forum. The event was comprised of many wonderful speakers from across the enterprise (American Family, Homesite, and The General®) and included guest speakers Dr Noan from the University of Wisconsin and keynote speaker Dr Stephen Aglash from Stanford University.
The General® uses topic modeling to understand what our customers are chatting about and how it changes over time. Whether it be to track A/B tests within specific user experiences or as a training tool for our chat agents to improve interactions with customers, we can combine the topic modeling component with metadata within a chat session to answer numerous questions. We are actively working to include sentiment analysis, word embeddings, and dependency parsing into future iterations. Additionally, we hope to expand our work around topic modeling to proactively route chats to specialized agents to improve customer experience and engagement.
Look for a future post about hierarchical Latent Dirichlet Allocation to learn more about how we are able to categorize chats by topic.