A hybrid chatbot that uses contextual sensors to influence responses.
Summary
AI chatbots are currently booming, and they have made great strides in the last couple of years in overtaking more traditional chatbot techniques. Nonetheless, AI chatbots are still lacking some functionalities that are uncomplicated for rule-based chatbots. This paper proposes a hybrid chatbot architecture that combines an AI and a rule-based chatbot to form a single chatbot that unites their respective strengths to cancel out some of their weaknesses. An AI chatbot is generally not able to permanently update its believes unless the model is retrained. On the contrary, rule-based chatbots can do this. Meanwhile, rule-based chatbots cannot appropriately respond to messages for which no appropriate rule is present, in contrast to AI chatbots. The hybrid chatbot can update its beliefs during conversations and can respond to messages for which no rules are known. Additionally, this ability to change beliefs during and between conversations allows for contextual sensors to influence the hybrid chatbot and provide accurate and current information to the user. The evaluation of the hybrid chatbot is done using two methods. Firstly, an experiment in which twelve participants talked to the hybrid chatbot. Secondly, an automated experiment to evaluate the effect of the contextual sensor data. The results show that the hybrid chatbot performs better on the metrics of responsiveness, repetitiveness, and conversational depth when compared to the default version of the AI chatbot. Additionally, it is shown that the information provided by the contextual sensors influence the responses generated by the chatbot.