Abstract: 【Purpose/significance】With the development of big data technology, an increasing number of internet users express their opinions and views through online platforms. Therefore, deeply mining online public opinion and understanding individual emotional attitudes has become particularly important【. Method/process】This paper proposes a Deep Topic Autoencoder Model (DTAM) that inte⁃ grates autoencoders, Generative Adversarial Networks (GANs), and the Latent Dirichlet Allocation (LDA) model. By combining the fea⁃ ture extraction capabilities of autoencoders, the generative power of GANs, and the topic discovery functionality of LDA, the DTAM model can more accurately classify thematic words in movie reviews【. Result/conclusion】Experimental results indicate that the gener⁃ ated topic classifications effectively reflect the main viewpoints and emotional tendencies in the reviews. This not only helps research⁃ ers to comprehensively understand audience movie-watching experiences and emotional feedback but also provides valuable reference data for movie production and marketing【. Innovation/limitation】This paper expands the application of the DTAM model in movie re⁃ view data, improving the effectiveness of topic classification for movie reviews and providing new insights for movie theme analysis. However, this study is limited to sentiment analysis of short movie reviews and does not consider time factors.