3-7, August 20
Kiva
Kevin O'Brien
kevin@kiva.org
Towards the end of my time at Kiva, I began a new project that aimed to cut down on the amount of time Kiva's community support team has to spend responding to common and tedious questions. For instance, despite entries on their FAQ page, Kiva gets many email requests with questions about how to change passwords and email addresses, which are all answered in a very similar way.
I had written a program that attempted to identify the most likely existing solutions for new questions. The program could identify the top 3 most likely solutions, and testing showed that the correct solution would likely be among those 3 more than 50% of the time.
I presented my findings at a meeting with the community support team and the machine learning weekly meeting. We brainstormed various ways to present the predictions: we could show them on Kiva’s website before Kiva issued a request, or we could include them in the confirmation email they receive after they submit a request. Hopefully in the future, these predictions will allow the community support team to focus on more important cases.
After the meeting, I worked with Melissa, a software engineer, to set up a new server environment that anyone at Kiva could use to do simple data analysis. We ended up staying in the office until about 7 because of some tricky issues with the setup. In the end, we had it set up, and now anyone at Kiva who knows a little Python can easily do analysis on Kiva’s data.
No comments:
Post a Comment