<p>Large-scale multi-objective optimization problems (LSMOPs) face dual challenges in practical applications: improving performance and reducing computational complexity. While directional sampling strategies have shown promise in large-scale optimization, they encounter two major difficulties: determination of sampling directions and computational resource allocation. Inspired by these issues, a large-scale multi-objective competitive swarm optimization algorithm based on multi-directional sampling strategy (referred to as LSCSO-MDS) is proposed, aiming to enhance the efficiency of solving LSMOPs. First, to overcome the difficulty of ineffective search directions in vast decision spaces, a multi-directional sampling strategy is designed to generate promising guiding individuals, which effectively steer the population toward the true Pareto front. Second, the critical issue of computational resource allocation is addressed through the integration of an offspring improvement strategy and a historical knowledge storage framework. Rather than performing exhaustive evaluations, a stage-dependent activation mechanism is employed alongside periodic sampling operations guided by historical diversity assessments. Consequently, redundant computational expenditures are minimized, and the utilization of fitness evaluations is strictly concentrated on the most promising evolutionary phases. To validate the efficiency of LSCSO-MDS, the proposed method is compared with six state-of-the-art large-scale multi-objective evolutionary algorithms on a widely used LSMOP test suite, with decision variable dimensions ranging from 300, 500, 1000, to 1500. Experimental results demonstrate that the proposed algorithm achieves a balanced approximation set in terms of convergence and diversity, exhibiting superior overall performance in most test instances.</p>