The paper talks about how unstructured data like text can be converted to structured data and facilitate coherent chat.
Chappie - A Semi-automatic Intelligent Chatbot
The personal assistant, a human is inefficient and takes a long time to process a single request such as booking tickets, ordering food, getting beauty-services etc. This reduces the throughput and business performance drastically. Obviously there is an increase in the demand of chat automation because a) it removes the human factor and b) it can give a 24 hour service which will have a multiplicative effect on the revenue generation. The chatbot, Chappie was born as a business requirement to automate the personal assistant or concierge. Presently Chappie is being used as a routing agent wherein it can classify the requirement of user into one of the services provided by business based on the first few chats and then transfer it to an agent expert in that service. It uses natural language processing (nlp) to analyse chats and extracts intent of the user with a score similar to the likes of WIT. Then it uses this information and AIML(Artificial Intelligence Mark-up Language) to make a conversation with the user. This is the marked difference between Chappie and existing chatbots like ALICE, which work solely on AIML. Hence we describe it as intelligent compared to other bots. Once it understands the intent and gets other information like name and email id, it transfers the chat to a human agent. Thus it is a semi-automatic chatbot that switches to a manual mode once it is able to understand the intent of the user and the kind of service required by her. Presently we let Chappie go live once an hour everyday and we record the responses as well as test cases for classification. So far the results have been encouraging since people have not been able to distinguish Chappie from a human being. Also it has been able to route correctly on several occasions. Once we obtain a decent amount of accuracy in routing and chat coherence, it will be deployed on a 24 hour basis.
Only registered members are allowed to comment.