With science and technology as the first productive forces, the construction of science and culture has gradually become an important indicator of national strength, and the competition around the construction of science and culture has intensified. In this study, we construct an LSTM behavior recognition model with multidimensional feature fusion from a deep learning perspective, aiming to identify the propagation of the scientist’s spirit through speech behavior. We deeply delve into the role of scientists’ spirits and utilize association rules to explore the relationship between the dissemination of scientists’ spirits and the construction of scientific culture. This paper’s recognition model achieves a recognition accuracy of 95.84% for innovative spirit communication behaviors and a recognition rate of 73.56% for dedication spirit communication behaviors. The recognition accuracy for the six spiritual communication behaviors of patriotism, innovation, truth-seeking, dedication, collaboration, and education is average, with an effective recognition rate of 86.05%. In summary, this study suggests a development strategy for scientists’ spiritual communication aimed at fostering the advancement of scientific culture. This strategy serves as a scientific benchmark for enhancing the development of scientific culture.