Understanding the Impact of Products on Customers
A large online marketplace required labeled training data from diverse text sources like listings and tweets. Achieving a quality standard of 98% to train their in-house NLP engine on such a large scale was challenging.
We built a custom annotation tool for Named Entity Extraction and QC, which can distribute thousands of tasks daily to our team. The seamless interface for Data Loading, Task Management, Annotation, QC and Delivery was supported by intensive and ongoing training as the team scaled up. We have processed over 150,000 tasks to date and continue to do more.
The team has become a true extension of the client’s workforce. This enhanced the functionality of their own NLP platform by a large factor. Over multiple iterations the customer worked with our Solutions Architect to refine output classes and merge confusion classes to derive higher levels of accuracy. Based on the success, they also initiated Spanish language work with our bi-lingual team in New Orleans.