Traditional influence maximization (IM) approaches often overlook critical aspects of real-world social networks, including topic-driven user interests, temporal activity patterns, and hybrid information diffusion mechanisms. To address these limitations, we propose TOTEM (Topic-aware Temporal Influence Maximization), a unified framework that systematically integrates topic modeling, temporal dynamics, and dual-interaction diffusion processes. Our work makes three key contributions: (1) a novel dual-interaction propagation model capturing both direct social interactions and recommendation-mediated diffusion through adaptive topic-temporal representations; (2) the formal definition and computational complexity analysis of the Dual-Interaction Temporal Influence Maximization (D-TIM) problem, with proofs establishing the submodularity of our influence function, and (3) two efficient heuristic algorithms: E-TOTEM employing temporal-discounted heuristic optimization for fast solutions, and S-TOTEM utilizing snapshot-based graph compression to achieve high quality performance. Extensive evaluations on real-world social networks demonstrate TOTEM’s superior performance in both influence spread and computational efficiency compared to state-of-the-art baselines. The theoretical soundness and empirical effectiveness of the framework suggest immediate applicability to practical social media marketing platforms.
