Twitter is fast evolving as servicing channel, though in a nascent stage. Twitter helps provide near real-time customer support, making brands more reachable and developing perception of customer-focused brand. Tweets as servicing request are pretty manageable in early days, but as user engagement increases the volume of tweets increase thereby bringing in plethora of problems. Classification of Tweets as serviceable becomes huge challenge, which provides data science an opportunity to play a prominent role.
These problems tend to complicate and multiply many folds especially in Indirect Tweets. We have lately built a solution with goal of solving Serviceable Indirect Tweet classification, Process Optimization on near real time platform.
We used Machine Learning Algorithm to classify incoming tweets in various category binaries. These binaries in turn act as Independent variables to classification algorithm which assigns appropriate class to tweets. The solutions keeps spirit intact by being near real-time.
Lib – Scikit Machine Learning Libraries
Example of some Binaries created by ML Algo
- Credit Card/Acceptance
- Credit Card/Upgrade
- Travel/XYZ Airline
We created more than 500 binaries/Indexes, basis ML algorithm to classify. These binaries became independent variable for serviceability classification Model. The Model is built using SVM classifier with customized kernel.
Machine Learning helped us reduce cost of servicing per tweet by 29%, as we were able to identify more serviceable Tweets within same sample. This also opens up a new interesting area where we are working to develop New products.