Interactive Question Answering Using Frame-Based Knowledge Representation

A framework for building interactive question answering (IQA) systems is proposed based on a frame representation in dialogue systems. This method uses semantic role labeling for building a frame-based knowledge base (KB) from a dataset of question-answer pairs. A frame representation of a question contains slot-value pairs presenting the question's semantic. A frame-based representation enables a question answering system to engage in dialogue interactions with a user. The IQA systems goal is capturing the user's intention and finding the best matching answer from the KB. A procedure for extracting slots (attributes) and their values for representing questions in the KB is proposed. Our framework was evaluated on datasets in the domain of car manuals. Our evaluation results show the effectiveness of the proposed framework for KB generation and IQA compared to different baseline methods. Moreover, the deployment of this method effectively reduced the manual effort for KB generation.