Using different ways of correlation, the characteristics based on the differences between knowledge points, core predicates, and discourse characters are investigated. The relevant content of sports textbooks is used to train the word2vec relationship model with the similarity between the statistical knowledge points. As a result, the features are obtained based on the noun vector along with in-depth meaning-related information. The extracted features are used to train the sorter method of support vector machine (SVM) and K-nearest neighbor (KNN) for the analysis of the relationship between knowledge points. According to the experimental data, the specific content of the physical education textbook is selected. Compared with the traditional methods, the refined method can effectively improve the F score of the correlation. Finally, the new association extraction method is used to establish the knowledge image of sports discipline. The experimental results show that this method can effectively extract the knowledge points from the physical education curriculum textbooks.