In recent years, the topic of scientific exercise has attracted widespread attention, and people realize that participating in physical exercise requires the management of physical exercise behavior in order to reap the desired results. In this paper, we propose an exercise prescription recommendation based on the K-means clustering algorithm and a self-adjustment mechanism based on NLP sentiment analysis. We aim to provide the public with more professional and planned exercise programs. A sample of 1000 online user data from exercise software is taken for simulation experiments to verify the recommended method’s iterative effect and effectiveness. The data show that the optimal number of clusters for both male and female users is five, and the clustered body types are analyzed according to the mean of the Z-Score standard score, which shows that the five boys’ body types are obese, thin, lean, strong, and athletic, and the five girls’ body types are lean, agile, thin, obese, and fat, respectively. The confusing heatmap of the NLP affective tendency reflects its affective. The classification method is effective, and the overall correct rate of sentiment analysis reaches 94.5%, of which the classification of users with positive emotional tendencies is the best, with a correct rate of 97.8%, while the classification of users with no emotional tendencies is poor, with a correct rate of 87.6%. The model’s AUC value in the ROC curve is 0.82, which means it has a better classification effect. In addition, taking a sample in the first category of male physique as an example, the NLP sentiment analysis method derives an exercise prescription adjustment that increases the intensity of this user’s exercise by 40% through his exercise data and sentiment feedback text.