Mixed feature expansion-based short text classification

To cope with the characteristics taken on short text classification, feature sparse and lack of background features in the field, a mixed feature expansion-based text Categorization is proposed, incorporating the Latent Dirichlet Allocation (LDA) and Frequent Pattern growth (FP-Growth). First and foremost, the theme words were extracted by LDA topic model as features and thereupon added to the original feature to first phase feature expansion. Secondly, given that the frequent probability index varies with the classes of word sets, different categories of basic confidence are set for the text. Furthermore, the high-quality and scalable features are filtered out for second phase feature expansion whereby the FP-Growth algorithm and chi-square test. Experimental results prove the mixed feature expansion method is able to effectively increase the accuracy of short text categorization in news text categorization.