To enhance resume screening and talent matching effectiveness, this study proposes an algorithm optimization method for a natural language processing-based resume screening and talent matching platform. First, we construct a CNN-Attention feature fusion model. This model is then applied to extract resume topic features through topic segmentation and classification, enabling efficient resume screening. The selected resumes are subsequently processed by the GA-LightGBM model for human-job matching. Finally, both the resume screening model and the human-job matching model undergo rigorous validation and analysis. Experimental results demonstrate that the CNN-Attention model achieves precision, recall, and F1 score of 98.42%,99.61%, and 97.37% respectively, while the GA-LightGBM model outperforms other comparison models with AUC, Acc, and MAP values of 0.83,82.53%, and 0.291 respectively. Comprehensive analysis confirms that the developed model effectively performs resume screening and job-person matching, thereby optimizing the talent matching platform design to meet practical application requirements with demonstrated effectiveness and generalizability.