Text Classification Algorithm Based On Mahalanobis Hyper Ellipsoidal Learning Machine

Support vector machine is a new machine learning technique based on statisticallearning theory, which has been successfully applied in many fields. Due to the textwith multi-class and multi-label characteristics, there are still many unsolved problemswhen applied to text classificationFor multi-class text classification problem, a multi-class text classificationalgorithm is proposed which is based on Mahalanobis ellipsoidal learning machine. Foreach class of the training samples, train Mahalanobis ellipsoidal learning machine,making it include the samples as many as possible, at the same time, exclude the noise.For the samples to be classified, calculate Mahalanobis distance first mapped to eachellipsoidal core, then according to whether the mapping is surrounded by ellipsoid ornot to identify its category. If the mapping is not surrounded by any ellipsoid or issurrounded by no less than two hyper ellipsoids, according to membership the mappingbelongs to each hyper ellipsoid, determine its category. In the Reuters21578standarddata sets, the classification experiment results show that the algorithm improves theclassification accuracy and classification speed.For the multi-label text classification problem, a multi-label text classificationalgorithm is proposed which is based on Mahalanobis ellipsoidal learning machine. Foreach training sample, use Mahalanobis ellipsoidal learning machine method to train fora hyper ellipsoid in a feature space, making it include samples as many as possible, atthe same time, exclude the noise. For the samples to be classified, calculateMahalanobis distance first mapped to each ellipsoidal core, then according to whetherthe mapping is surrounded by ellipsoid or not to identify its category. If the mapping isnot surrounded by any ellipsoid, according to membership the mapping belongs to eachhyper ellipsoid, determine its category. In the Reuters21578standard data sets, the classification experiment results show that the algorithm improves the classificationaccuracy.