I was recently working on my first Chatbot project and sent it to friends to to test and I did a bit of analysis of the log data and details below:
Out of 62 questions that were asked, 35 were had correct responses from the chatbot, 0 had wrong answer from bot while actual answer is in the database, 10 had wrong answers because these questions were not in the database, 1 wrong answer due spelling error from uaer, 9 wrong answers because they used synonyms of the words in the database, 1 wrong due to bug and 5 questions did not make sense.
Link to Chatbot http://covid19chatbot.info/
Am testing it for covid-19 but i want to use it for other user cases later if the algorithm works well.
I use the Naive bayes algorthm
Any advise or comment?