User participation and idea performance are essential for the success of open innovation communities (OICs) as valuable sources of innovation. Interaction within OICs plays a crucial role in the idea generation process. This study examines the relationship between user-to-user and user-to-firm interactions and their effect on the idea generation performance. Using a transactive memory system, a research model is proposed to explore the influences of user interaction characteristics (i.e., specialization, credibility, and coordination) on idea generation performance and the mediating effect of content richness. Furthermore, we reveal whether metadiscourse in firm feedback (i.e., interactive and interactional resources) moderates the association between interaction characteristics and idea generation performance. A dataset of 7,603 ideas from the Huawei's Joint Definition Community (JDC) is analyzed to verify the research model. Results show an inverted U-shaped relationship between specialization and idea generation performance. Coordination and credibility among users have positive effects on idea generation performance. Content richness has a mediating effect on the relationship between interaction characteristics and idea generation performance. Moreover, metadiscourse enhances the influences of specialization, credibility, and coordination on idea generation performance. These findings provide rich insights into idea generation performance in open innovation and offer practical implications.