This paper investigates the mesoscopic structure of innovation networks, using patent data from AI, biotechnology, and semiconductors to compare community detection methods and analyze how collaboration patterns influence technological impact. It reveals distinct hierarchical structures in inventor and organization networks and highlights the importance of inference-based models over traditional modularity for capturing these complexities.