You are building an NLP system to route incoming support tickets to the right queue in a consumer fintech support organization. Tickets arrive from chat, email, and in-app forms, and they often contain short descriptions, account terminology, transaction references, and informal language. Historical tickets already have routing labels such as card disputes, direct deposit, account access, fraud review, and transfer issues, but the data is noisy because some tickets were reassigned after intake. You want a practical solution that can classify new tickets accurately enough to reduce manual triage while still allowing uncertain cases to fall back to human review.
How would you build a model to route support tickets using NLP?
Text classification framing for support routingTokenizer and preprocessing choices for noisy ticket textTF-IDF baseline design versus transformer fine-tuningF1-driven evaluation for imbalanced queues