Latent Semantic Analysis for Customer Care
The customer care department I work in handles over 21,000 text-based customer interactions each week. Though it's easy to track certain types of data associated with those interactions (what topics were tagged, how quickly did we respond, etc), the volume means a lot of other "soft" data (tone, sentiment, pain points) falls through the cracks. There simply isn't enough time to analyze all of it.
I've begin milling around the idea of applying some machine learning methods to customer interactions. Other companies are already doing it, and with a bit of customized python, it seems quite achievable!
Here's a great video I found that walks you through the steps of applying latent semantic analysis to a large database of text:
In my first test, I'll be using the model above on a group of customer interactions we already have human insights on, to see how they compare.