A chatbot by combining finite state machine, information retrieval, and bot-initiative strategy

In this work, we designed a conversational system by combining a finite state method, a retrieval model and a machine-initiative dialogue strategy. By using the machine-initiative strategy, most of the user queries are handled by the finite state machine which models predefined dialogues. If the user’s utterance is out of the range of modeled dialogues, the retrieval model processes the input. On comparative experiments, our retrieval model showed a better user satisfaction than the neural-network-based model. The experiments also showed that our context handling method and keyword emphasizing are key components of a retrieval model for the superior performance. In addition, dialogue strategies to retain initiative to machine was better to make a sustaining conversational system than to be strictly constrained to input sentences.