Collective understanding and public discourse in online news communities are shaped by the dynamic interplay of editorial practices and community interactions. This study introduces a computational model calibrated with data from 400 million comments across five news platforms to examine how collective semantic networks evolve and differ from societal-level semantic representations. The model captures the influence of editorial agenda-setting, including alignment, amplification, and reframing, as well as community dynamics such as membership turnover, troll activity, and counterspeech. Our findings show how editorial practices can sustain echo chambers or challenge them through diversified narratives, and how community responses either reinforce or resist these influences. The results demonstrate that collective semantic networks are sensitive to external shocks, editorial policies, and community interactions, offering insights into mechanisms shaping public discourse.