- We contributed to broadening the capability of Conversational Information Seeking (CIS) agents by introducing the capability of curiosity through commonsense knowledge graphs.
- We introduced ISEEQ, a CIS system capable of providing conversational assistance by dynamically retrieving meta-information supportive in generating information-seeking questions (ISQs).
- In contrast to clarifying or followup questions, ISQs go a step further with expanding the query context by exploring relationships between entities in the query and linked entities in a knowledge graph. Thus retrieve a diverse set of passages to address a user query adequately.
- ISEEQ opens up future research directions in CIS by facilitating the automatic creation of large-scale datasets to develop and train improved CIS systems. Furthermore, crowd-workers can evaluate and augment such datasets rather than create them anew, thus improving dataset standards.
- Broadly construed, through reinforcement learning with the reward on conceptual flow and logical agreement, ISEEQ can be trained to generate questions that are safety constrained and follow a specialized knowledge processing. reference