Investment chatbot: applying natural language processing to Q&A
As trust companies expand, they face increasing challenges with communicating with their trust owners. Trust owners may find themselves needing a quick answer to a question and waiting for their trust officer to reply. Trust officers, responsible for managing clients’ accounts and communicating with the clients, may find themselves responding to questions that do not require industry expertise. In this thesis, I explore natural language processing to build a chatbot that could improve the efficiency of communication between trust owners and trust officers. My chatbot uses machine learning to predict the intent of each message from a user following the tutorials in - and example code in . Then, the chatbot either generates a response for the user or passes the conversation off to the trust officer. Sometimes, a user will send a message to the chatbot like “Hello” or “Thank you”. The chatbot can use a predefined, static response such as “Hello there” or “You are welcome” in these cases. These static responses are chosen by matching the predicted intent of the user’s message, then selecting an appropriate response from the responses.json file as suggested in -. However, if the chatbot needs to retrieve data or perform a computation for the user, the chatbot’s response cannot be simply chosen from a file of predefined responses. The response is generated dynamically through a topic handler class that retrieves the information the user requests and formats the information into a response for the user. The chatbot could provide quick responses to common questions while freeing trust officers to respond only to those inquiries that require the trust officer’s expertise.