We will present a mechanism underlying the collective emotional behavior of the users on Blogs and similar Web sites using an agent-based model with parameters inferred from the empirical data. Emotional agents interact trough posts and comments creating growing bipartite network. The state of the agent is described with two emotional variables, arousal and valence, which fluctuates in time due to events on posts. In the moment of agent's action its state is transferred to a selected post. Communication of the emotions in the model rules, combined with action-delay time and circadian rhythm characteristic for human behavior, can explain the genesis of emotional burst by users obtained in the analysis of empirical data. By varying different model parameters we discover their influence on course of the dynamics.