General Discussion & Announcements

Is the search to identify customer happiness by sentiment on comments really possible?

  • 1.  Is the search to identify customer happiness by sentiment on comments really possible?

    ROCKETEER
    Posted 06-01-2021 11:28
    Every year at Rocket we get the opportunity to pitch an idea for a technical project and do a bit of inventive hacking. We call this fun event Rocket Build 🚀.
    This year I wanted to find out if there is any value in using a Machine Learning sentiment algorithm to scan our huge mass of support comments to identify whether contacts (clients/users etc..) were happy (aka satisfied) with their support experience or not, so we could take remedial action. (We already do this of course - but only if we can identify it)
    On average we get about 15% survey responses on closed support issues (cases), which is a pretty good result. So we have a significant sample of data where the contact personally identifies whether they were satisfied or not with the experience AND we could use this sample data to train an sentiment algorithm on the remaining 85% of cases. - but would it really work?
    It seems that some people like replying on case surveys, but most don't. So a lot of those survey could be bias by the recurrent 'friendly' persons or by the few very angry persons which could skew the results.
    Or we could just use a generic natural language sentiment algorithm which has been trained on billions of tweets or similar, and hope that the language used in support comments will be a suitable fit. But I doubt it - have you seen all the code snippets in the support comments :) 
    Any thoughts?

    ------------------------------
    Mark Woodall
    Rocket Software (CS) -
    "In teaching others, we teach ourselves."
    ------------------------------