Experts at Haptik cater to millions of user queries belonging to multiple domains of products and services. Our primary goal to provide accurate solution over a messaging platform within four minutes , necessitates experts to work efficiently by optimizing their effort.
In order to scale, optimize and cater to such a large volume of messages, our Natural Language Processing and Machine Learning algorithms are key for our experts to find the best solution with negligible latency. In this journey of augmenting machine intelligence and human effort, the first problem we solved was training a bot to understand user queries from an ongoing conversation.
After thorough analysis we identified that most of the conversation follow the subset of below general pattern :
- A conversation starts with a casual greeting from the user
- Followed by a specific query
- Sometimes the experts ask follow up questions to get more clarity of the users problem
- After understanding the query a message with the concrete solution is sent to the user
- And the conversation ends with courtesy messages from each side
In order to dig deeper into identifying this pattern at a granular level, we used Natural Language Toolkit (NLTK 3.0) to pre-process incoming chats and designed a classifier to categorize our messages into the above categories.
Messages which are classified as a user queries and clarifications are further processed and fed into our response recommendation algorithm. This algorithm takes the data relevant to the user’s question and looks it up across an extensive knowledge base which we carefully curated over a period of time and returns the best solution to Haptik expert all within milliseconds.
Providing accurate, quick and personalized response to our users being our highest priority, our efforts will continue in this direction to make a perfect blend of machine and human intelligence.
Want to join is us in scaling our technology as a part of engineering team? Just get in touch here 🙂