<p>As reputation systems (RepSys) and recommendation systems (RecSys) have the ability to guide customers in making purchasing decisions, an increasing number of attacking groups cause great harm to these platforms. Conventionally, researches about group attacks aim at simulating many attackers aggregate together to attack certain products. Existing simulated group attacks are oriented to static data, lack of human-driven and organization collaboration modeling. To solve these limitations, we propose a novel strategy called OGAS-DDS, which introduces an organized group attack in dynamic data streams. Our OGAS-DDS involves an adaptive structural constraint strategy and an adaptive behavioral constraint strategy. The adaptive structural constraint strategy uses linear regression to fit degree distribution of the dataset, generating degree sequences of attacking groups that mimic real users. The adaptive behavioral constraint strategy selects camouflage products via NFS metrics, uses LLM to generate realistic reviews adapted to emotions, and utilizes the Monte Carlo method to construct a uniform attack sequence. Experiments show that our attack strategy is feasible and effective in enhancing product reputation and recommendations and is virtually undetectable by state-of-the-art detection models. The labeled attacking groups generated by the attack can be used as the ground-truth for further promoting the design and performance of the group attacking detection algorithms.</p>