Task-specific dependency-based word embedding methods

While most traditional word embedding methods target generic tasks, two task-specific dependency-based are proposed for better performance in text classification tasks this work. First, we exploit the dependency parsing tree structure to capture structural information of a sentence, and develop method called (DWE). It finds keywords neighbor words as contexts via parsing. Next, leverage word-class co-occurrence statistics model class distributional incorporate it into learning process. This leads class-enhanced (CEDWE) method. Task-specific corpora matrix-factorization-based framework used train DWE CEDWE. Seven datasets evaluate CEDWE, experimental results show that they outperform several state-of-the-art methods.