In text classification task one of the main problems is to choose which features give the best results. Various features can be used like words, n-grams, syntactic n-grams of various types (POS tags, dependency relations, mixed, etc.), or a combinations of these features can be considered. Also, algorithms for dimensionality reduction of these sets of features can be applied, like Latent Dirichlet Allocation (LDA). In this paper, we consider multi-label text classification task and apply various feature sets. We consider a subset of multi-labeled files from the Reuters-21578 corpus. We use traditional tf-IDF values of the features and tried both considering and ignoring stop words. We also tried several combinations of features, like bigrams and unigrams. We also experimented with adding LDA results into Vector Space Models as new features. These last experiments obtained the best results.
展开▼
