This paper investigates the problem of adaptive event-triggered data-driven predictive control for unknown cyber-physical systems subject to DoS attacks. To address the deficiency where traditional event-triggered mechanisms (ETM) are prone to losing critical data during DoS attacks, an adaptive ETM with a dynamic compensation factor is proposed. This mechanism quantifies the impact of attack duration on the triggering threshold via exponential decay characteristics, thereby achieving a balance between communication resource consumption and control performance. Furthermore, by combining Willems' fundamental lemma, a predictive controller is constructed directly using offline data. The prediction sequence is obtained by solving an optimization problem to actively compensate for the adverse effects of DoS attacks. Theoretically, the recursive feasibility of the multi-step control without terminal constraints under the proposed adaptive ETM is proven, and the sufficient conditions for system stability are derived. Finally, using an open-loop unstable reactor as a benchmark, the effectiveness and robustness of the proposed method are rigorously verified in an environment where DoS attacks and noise disturbances coexist.