In response to the problem of large errors and poor real-time performance in career path decision-making for college students, this article has adopted the Deep Q-Network (DQN) model to study the optimization of career paths for college students. Firstly, the background data of college students is preprocessed. For the DQN model, experience replay was used to store and reuse historical interactive data to reduce errors. Then, a target network was introduced to estimate the target Q value to improve the efficiency of sparse data utilization, and appropriate reward functions were designed to quantify the quality of each decision step. Finally, by comparing the performance of different path optimization strategies, the experimental results showed that the decision efficiency of the DQN model reached 98.2%, which was 3.3% higher than that of the convolutional neural network (CNN), and its response speed was only 42.5ms to meet real-time requirements, with the root mean squared error of only 0.02. This has achieved precise decision-making suggestions and good real-time performance, which has a good guiding effect on college students’ career planning.