A big player from the e-commerce field whose name is under NDA needed to provide their customers with perfect support and feedback along with to cut the support expenses and to relieve the overloaded support department. This platform had requested us for assistance. Do you know people who do not use to be happy on little and want to get all at once?
Well, we know now. The existing decisions were unappropriating so we had decided to develop our own system that will meet all the requirements. Check what we have done, we are really proud of ourselves after long months of hard work and the perfect result.
According to the purpose we tried to create a system of tasks that will meet all the requirements. Here it is:
- To create a text summarization system that would process a text and extract the issue.
- To develop a text classification system that would categorize a text simplifying internal processes of the support department.
- To create a text generation system that would answer the most common requests just as if it were a human.
- To build a system for semantic analysis.
It looked a bit daunting but we were born as troubleshooters. Going forward let’s say that it was quite difficult still amusing and fascinating to dive deep into this project.
We used deep learning and classical NLP methods in the process of development. We had to learn a lot of new interesting and now feel like we are ready to conquer the world with all that information we had known but we like what we do so we continued with Python and TensorFlow. So after four months of hard work, we were ready to present our brainchildren:
- We have created the text summarization system with a high-quality result. Right now we do not have data on how much time support teams save with this system but at least guys can have an extra break now what is already good.
- We have developed the text classification system. It processes a text and after a quick analysis distinguishes what kind of message it was. Whenever it is a complain, request for help, thank-you note or a routine question, the system will categorize these texts in the order a user wants with 91% accuracy.
- We have built the text generation system that can create texts according to the purposes and style.
- We had also built the semantic analysis system which is 92% precise. It automatizes the process of analyzing feedback and conversations. Marketers and supporters do not have to read out lots of texts any more as for now they can the summary just within a few seconds.
We created a system of tools that automatizes routine processed like monitoring reviews and feedbacks and categorizing requests to support department. On the one hand, end-users of a big e-commerce platform (sorry but its name is still under NDA) get improved service. On the other hand, support and marketing departments have a bit more free time to spend it for something really inspiring.
If combined, these two features help the platform save money on automatized processes and earn more as their customers get better service. And your business also can do the same, you just need to ask us how.