News Title Classification Based on Sentence-LDA Model and Word Embedding
Fan Zhang, Wang Gao, Yuan Fang · 2019
Due to the severe data sparsity problem, conventional text classification methods are difficult to achieve good results in news title classification. In this paper, we design a novel news title classification method, referred as Word-Embedding-based Sentence-LDA (WESL) model. WESL employs the Sentence-LDA model to distill topic vectors from news titles, and FastText is used to learn word distributed representations. Furthermore, the proposed model expands the features of news titles by combining word and topic vectors. Finally, we utilize a Support Vector Machine (SVM) to classify news titles, and the classification results are evaluated using the precision, recall and F1-value. The experimental results illustrate that our method can significantly enhance classification performance.