Processing email responses for intents and deciding best fit intent from multiple intents

We have a dataset of generic sales email responses of on average 20 word length. They are classified into the following intents at the high level - unsubscribeRequest, notInterested, alreadyInTouch, leadDoesNotHaveBuyingAuthority, busyWillRevertLater, interestedAndPossiblyMeetingRequest, connectingtoDecisionMaker, usingCompetitor, happyWithExistingSolution, nurture (notInterestedRightNowButMightBeInterestedLater)

I have already worked on sentence level intent recognition and basic heuristic based best fit intent and its giving pretty good accuracy. I also extract entities and use it in decision making to some extent. Although I explored entire email level classification, considering how overlapping the above classes can be, the accuracy is not that great.

I am looking for someone who can work with me for a longer period of time - 4-6 weeks, on the above problem to improve precision for the cases above. I am open to potential employment but I definitely want someone with availability right now as hiring for this role will take more time.

1) the work will involve improving the word2vec to better the intent recognition (similar phrase matching)
2) sequence classification on the above classified intents to find best fit intents
3) improving heuristic by understanding data to better decision making using both intent and entity signals
About the recuiter
Member since Aug 11, 2017
Pollyanna
from Rhone-Alpes, France

---

---

Looking for help? Checkout our video tutorial
How to search and apply for jobs

How to apply? Do you have more questions about the Job?
See frequently asked questions