Empowered by the wisdom of crowds, innovation nowadays is increasingly relying on diverse individuals' knowledge collaboration. Research on crowdsourcing and open innovation has demonstrated that through deliberate understanding and reflective thinking, members of the online crowd collectively manage their knowledge to generate innovative ideas. However, the semantic patterns of how online crowd's collective reflection ultimately leads up to innovation remains unclear. Employing semantic network approach, this study analyzed a total of 1,116 posts contributed by online crowds responding to two organization-sponsored crowdsourcing open innovation challenges. Findings show that the semantic patterns of online crowds' knowledge collaboration evolve from one phase to another in accordance with crowd members' collective reflection on their diverse knowledge. Theoretical and practical implications are discussed.