Work Experience:
2013 - now Software Development Engineer
Project Experience:
* Template recommendation
Custom Service website provides tons of email templates (eg. around 50 thousands in NA) for the associate to reply the customer’s email. It’s hard for the associate to choose the right template.
Depend on the issue code and auto annotation we used association rule to build different recommendation models. Depend on multiple recommendation models to balance the accuracy and coverage, finally give the recommended templates.
For now the top10 coverage is 46.10%, accuracy is 82.27%. After importing the user id and work category data, top1 confidence is higher than 0.8. The more accurate result will reduce the associate’s AHT on the email composer page.
* Copy&Paste detection
Sometimes the associate pastes the email templates from memos instead of inserting them from the email composer page. Without the insert action, our system could not detect templates used in the email. So we need to find a way to detect such use case.
The basic idea is using jaccard similarity and NGram to calculate the similarities between the email content and each template in our database, then the most similar template could be the one used in the email (to be accuracy, the similarity should beyond a threshold).
Languages: Java, Jquery, Ruby, Clojure, R