With dongba culture researches increasing year by year, there needs to be a highly efficient classification method to classify research achievements creating conditions for further study. Aiming at the shortcomings of the traditional mutual information method, giving full consideration to the factors such as word frequency, concentration and dispersion, and using the difference between the maximum and the second large value as a global evaluation function, GMI feature selection algorithm is proposed. Use this algorithm to choose text feature after one dimension reduction, and then get classification feature combined with the literature feature on secondary dimension reduction, and finally utilize the SVM to classify dongba literature. The experimental results show that the average accuracy rate and recall rate in all categories are 83% and 82% respectively. Experimental results show the proposed method is feasible in the dongba literature classification.
展开▼