Unlabelled text mining methods based on two extension models of concept lattices

Concept lattice is a useful tool for text extraction. The common text clustering method fails to generate hierarchical relationships among categories and realize soft clustering simultaneously, while the concept lattice ignores the negative correlation between an object subset and an attribute subset. Motivated by the problems, we propose unlabelled text mining methods based on fuzzy concept lattice and three-way concept lattice. Firstly, we excavate hierarchical text categories to construct a classification system based on fuzzy concept lattice, and the labelled samples are obtained by the word matching method. Then, we construct a three-way concept lattice to get positive and negative classification rules based on the labelled samples, and the classifier is constructed to predict the new samples. Finally, Sogou laboratory news corpus is used to evaluate the efficiency of text clustering and classification methods. The results demonstrate that the improved clustering method has a higher average cluster goodness than earlier procedures and the classification model based on three-way concept lattice achieves a higher accuracy.