Influence maximization research in the real world allows us to better understand, accelerate spreading processes for innovations and products, and effectively analyze, predict, and control the spread of diseases, rumors, and computer viruses. In this paper, we first put forward a new path-based node similarity measure, named the dynamic local similarity index, which can be dynamically adjusted to the optimal mode according to network topology characteristics. Compared to the Katz index with high complexity and an LP index with a limited application range, the proposed index achieves an excellent balance between complexity and precision. Second, combining the extended neighborhood coreness with the minimum distance, a novel strategy is presented for selecting initial centers of clusters, which is helpful for speeding up clustering convergence and avoiding local optimum, especially in non-connected networks. Subsequently, we present an adaptive heuristic clustering algorithm, which can find the seed set with maximum collective influence through clustering. The empirical results on four real datasets show the effectiveness and efficiency of the proposed algorithm, which compares favorably to several state-of-the-art algorithms.