ISEEQ: Information Seeking Question Generation using Dynamic Meta-Information Retrieval and Knowledge Graphs

Conversational Information Seeking (CIS) is a relatively new research area\nwithin conversational AI that attempts to seek information from end-users in\norder to understand and satisfy users' needs. If realized, such a system has\nfar-reaching benefits in the real world; for example, a CIS system can assist\nclinicians in pre-screening or triaging patients in healthcare. A key open\nsub-problem in CIS that remains unaddressed in the literature is generating\nInformation Seeking Questions (ISQs) based on a short initial query from the\nend-user. To address this open problem, we propose Information SEEking Question\ngenerator (ISEEQ), a novel approach for generating ISQs from just a short user\nquery, given a large text corpus relevant to the user query. Firstly, ISEEQ\nuses a knowledge graph to enrich the user query. Secondly, ISEEQ uses the\nknowledge-enriched query to retrieve relevant context passages to ask coherent\nISQs adhering to a conceptual flow. Thirdly, ISEEQ introduces a new deep\ngenerative-adversarial reinforcement learning-based approach for generating\nISQs. We show that ISEEQ can generate high-quality ISQs to promote the\ndevelopment of CIS agents. ISEEQ significantly outperforms comparable baselines\non five ISQ evaluation metrics across four datasets having user queries from\ndiverse domains. Further, we argue that ISEEQ is transferable across domains\nfor generating ISQs, as it shows the acceptable performance when trained and\ntested on different pairs of domains. The qualitative human evaluation confirms\nISEEQ-generated ISQs are comparable in quality to human-generated questions and\noutperform the best comparable baseline.\n