Digital media technology plays a crucial role in reproducing historical scenes and facilitating virtual interactions with political figures in ideological education, with virtual reality (VR) serving as a crucial technological support for the reconstruction of a new ideological education model. In order to improve the teaching immersion, this paper designs a multimodal gesture recognition algorithm. The acquired gesture data are regarded as curves, weights are assigned according to the degree of gesture fluctuation, and weight-based Pearson correlation coefficient algorithms (WPCCs) are proposed to calculate gesture similarity. The data and model are trained in a distributed manner using the FEDER learning technique finally. The optimized algorithm achieves a recognition rate of 96.66% for eight common gestures, which is 37.36% less time-consuming than the DTW algorithm. After applying the algorithm to virtual reality ideology education, students experienced a strong sense of presence (mean = 4.12); the post-test learning motivation score was 0.42 points higher than the pre-test (p<0.05); and the mean scores of perceived ease of use and emotional experience evaluations were 3.97 points and 4.33 points, respectively. It indicates that the algorithm in this paper has excellent performance and application efficiency.