Chatbot implementation for the Cambridge restaurant system
In this report, the implementation of the Cambridge Restaurant System chatbot is described and evaluated. The chatbot has the goal of suggesting a restaurant to the user, using the food, area and price preference of the user. A dataset of dialogs between a user and a chatbot is used to make a state transition diagram representing the user's utterance and the response given by the chatbot. The user's utterances are classified by type of dialog act. The state transition diagram is implemented in a chatbot, in which the type of dialog act of the user's utterance is classified by a RandomForest and a suitable response is given by the system. The chatbot works fairly well due to the F1 score of 0.98 of the algorithm on the test data. However, the implementation of the state transition diagram and the use of word patterns to find the user preferences limits the chatbot. Another limitation is the implementation of the spellcheck. The chatbot could be mainly improved by focussing more on the framework