Classification Coverage Biomimetic Pattern Recognition Algorithm Based On Super Ball Text

Since Google CEO Eric Schmidt first proposed the concept of “cloud computing”,it has been becoming an important growth point in the international IT industry now.With the cloud computing era coming, types of Internet applications emerge inendlessly, and various types of data show explosive growth. To the massive text datastored in the cloud database, how to access, manage and use them quickly andefficiently for users has become an important problem, that need to be addressedurgently. In which text classification is an effective data processing method for thetext processing.How to construct the classifier is extremely important in text classificationsystems. Traditional text classification algorithm or traditional pattern recognition isbased on the optimal partition, while bionic pattern recognition (BPR) emphasize“matter cognition” instead of “matter classification”--it is more in line with thenature of human understanding the world. The mathematical method of BRP is usinghigh-dimensional complex space geometric shapes to cover the sample in the featurespace, then we can achieve pattern recognition. BPR has achieved good recognitionresults in speech recognition, face recognition and ground physical recognition. Thepractice shows that BPR has many advantages, that it can reduce the error rate of therecognition effectively, and the efficiency of recognition superior to the traditionalpattern recognition methods.As a starting point, this article first briefly introduced the concepts and keytechnologies of text classification, then we studied the theoretical basis and neuralnetwork algorithm achievement of BPR. Second, from the point view of geometricaltheory, we proposed a BPR algorithm, which use hyper-ellipsoidal ashigh-dimensional space homology class template, it owns to our laboratory. Went onthis basis, targeted at the high-dimensional of text feature space and the sparsenessof text vector, we proposed an improved BPR algorithm, that is based onhyper-sphere model. This algorithm can cover the sample points in the feature spaceflexibly, and overcome the sparseness of text feature. The experimental results show that, on the Chinese data, the classification performance of BPR classificationalgorithms which is based on hyper-sphere coverage is better than that of thetraditional classification algorithms. Although the former need to be improved in thetraining time, it can achieve a faster rate in the classification time.