Evolutionary Multitask Optimization enables simultaneous optimization of multiple tasks through knowledge transfer. However, existing approaches primarily focus on knowledge acquisition and knowledge transfer, with limited attention given to investigating the impact of transfer target selection. To address these issues, we design a new evolutionary multitasking approach that adaptively selects transfer targets via external sample matrix, called MTEA-ATESM. Our approach integrates an ε-greedy adaptive transfer strategy with Softmax dynamic weighting, optimizing knowledge transfer across tasks. Additionally, an external sample matrix stores high-quality solutions, enhancing transfer precision and search efficiency. Extensive experiments on CEC17-MTSO, WCCI20-MTSO, and the real-world PKACP problem demonstrate that MTEA-ATESM consistently achieves better results than five advanced multitask optimization methods. Ablation and sensitivity analyses confirm the effectiveness and stability of the proposed strategies.