Effective and Interpretable Document Classification Using Distinctly Labeled Dirichlet Process Mixture Models of von Mises-Fisher Distributions

Document Classification is essential to information retrieval and text mining. Accuracy and interpretability are two important aspects of text classifiers. This paper proposes an interpretable classification method (DLDPvMFs) by using the Dirichlet process mixture (DPM) model to discover the hidden topics distinctly within each label for classification of directional data based on the von Mises-Fisher (vMF) distribution, which arises naturally for data distributed on the unit hypersphere. We use a mean-field variational inference algorithm when developing DLDPvMFs. By using the label information of the training data explicitly and determining automatically the number of topics for each label to find the topical space, class topics are coherent, relevant and discriminative and since they help us interpret class’s label as well as distinguish classes. Our experimental results showed the advantages of our approach via significant criteria such as separability, interpretability and effectiveness in classification task of large datasets with high dimension and complex distribution. Our obtained results are highly competitive with state-of-the-art approaches.