A Method Based on Contextual Reasoning for Knowledge Retrieva‎l in Question Answering Systems

10/2/2023 12:00:00 AM

Question Answering (QA) systems for information retrieva‎l have become increasingly popular with the expansion of textual information resources. Information retrieva‎l is made easier by QA systems, which can respond explicitly and immediately to incoming questions. The ability to answer multi-hop questions is one of these systems' new challenges. Answering multi-step questions is a complex process and the QA model must be able to make logical inferences and draw conclusions based on the information given to it. This requires a deep understanding of the question and the context in which it is being asked. Another critical challenge of QA systems is Explainability, which helps to understand and interpret the prediction of the artificial intelligence model. Explainability allows users to understand the reasoning behind the answer and eva‎luate the accuracy of the results. We present a multi-component method to solve the multi-hop challenge of QA systems. In this approach, first, the documents related to the question are retrieved in a two-step process. Then, to answer the question, knowledge is extracted from the documents. After the representation of the knowledge, contextual reasoning is done by combining the transformer and heterogeneous graph neural network. We obtain the supporting facts using the retrieved golden documents and the predicted response for the system describability challenge. Our approach has been tested on HotpotQA's open-domain multi-hop dataset and has been able to achieve strong performance compared to the baseline model and several related works.

پرسش و پاسخ چندگامي , استدلال محتوايي , بازيابي دانش , توضيح‌پذيري

Multi-hop Question Answering , Contextual Reasoning , Knowledge Retrieva‎l , Explainability

milad molazadeh oskuee

Dr Behrooz Minaei