Extraction and combination of multiple views of texts to improve text classification
Analyze information from multiple perspectives can aid data understanding and knowledge extraction, allowing one to explore a broader range of features, principally through textual data. To further understand the whole aspects of document content, it is fundamental to observe it from multiple perspectives to have an objective investigation. In this work, we are proposing a technique capable of generating multiple document-views to learn their characteristics and incorporate their knowledge into one document representation worthy of generalizing an improved predictive model to enhance text classification. We are presenting two separate strategies to combine our multi-document-views. Each approach has a few variations to further flexibility. We evaluated our methods through the document classification task over two perspectives. Initially, we compared them against single-view methods, and later, we compared them against other state-of-the-art multi-view strategies using four classifiers over six different scenarios. Our techniques outperformed the baselines on almost all experiments showing its capacity to learn from multiple views of texts.