Research On Text Classification Algorithms Based On Transfer Learning

Transfer-learning technique has become an important means of cross-cutting text classification and research focus, for its migration ability of knowledge, skills and experience. In this paper, by summarizing the application and development of transfer learning in text classification, the discussion and research are conducted for the problems and difficulties exist in the field, and several new transfer-learning algorithms are promoted. For the dimensionality disaster and undefined meaning of feature word exist in common text classification and easily lead to low classification accuracy and over-fitting problems, a feature dimension reduction algorithm HLK. based on feature selection and extraction was promoted. For the amount and similarity feature between texts of the original field and the target areas, two instance transfer learning methods CGTL and IDRTAT were proposed; for the significant difference in distribution between the data set of the original field and the target field, a feature representation transfer learning algorithm based on the feature was promoted, which was called BFRTL, and the feasibility of all algorithms were verified by experiments.