With the development of multi-modal analysis and intelligent recommendation systems, the trend of precise combination with Normal University is emerging. This study proposes a user behavior model construction method based on multi-modal data and a personalized advertising recommendation algorithm. From a theoretical perspective, first of all, although the user’s Internet behavior is affected by uncertain factors, this study integrates discrete instance data to construct a multi-modal data processing framework including text analysis, visual reasoning and statistical models. On this basis, this study designed a personalized advertising recommendation algorithm. The algorithm averages and maximizes the interactive behavior of users during video playback. This modeling can be used to recommend advertising directions. Experimental results show that this method improves the personalized recommendation accuracy by 5.2% compared with traditional similar algorithms. In addition, the model accumulates the influencing factors of user behavior values through deep learning, improves the model's accuracy in distinguishing user intentions, and provides an algorithm optimization solution for the filtering mechanism.