A Study On Sentiment Feature Selection In Aspect Level Sentiment Classification

Aspect Level Sentiment Classification is a fine-grained Sentiment Analysis.Different from Sentiment Analysis predicting the sentiment polarity toward the whole sentence or document,it aims to classify the sentiment polarity of a given aspect term or aspect category in a sentence.For a given aspect term,since a sentence may contain more than one aspect term,there may exist some opinions which are not the modifiers of the given aspect term.It is necessary to capture relevant opinion for a certain aspect term.Deep Learning achieves great growth in this task because of its strong ability in extracting sentiment features.However,there are shortcomings for most current models in selecting the aspect-related sentiment feature from all extracted sentiment features,and it results in selecting wrong opinion and misclassification.Relative position distance,gating mechasim and attention mechanism are most adopted in selecting the aspect-related sentiment feature.We analyse their shortcomings and propose an improved method.We model dependency relation between the given aspect term and all words in the dependency parse tree.We integrate this relationship into deep learning models to select aspect term related sentiment feature.Specifically,we propose dependency weighting mechanism to quantitatively compute such relation.We integrate this relationship into CNN and Bi-LSTM respectively to deal with this task.We conduct sufficient experiments to verify the effectiveness of dependency weighting and our proposed model achieve state-of-the-art results among neural network models.