Target Sentiment Classification Based On Deep Memory Networks

Target sentiment classification identifies the sentiment polarity that a sentence expressed to the giving target.The first step of target sentiment classification is to identify the context expressed sentiment to the target.Deep memory network based on attention mechanism can solve it effectively.However,one research found that memory network only focusing on attention mechanism cannot tackle this situation where the context sentiment relies on the concrete target.What's more,typical deep memory networks for target sentiment classification cannot synthesize the context semantic information when generating the context expression vector.For detecting the context of one target,this paper proposed a attention-based deep memory network model,which computed every context word weight with attention mechanism.In order to address problem of identifying target-dependent context sentiment,this paper proposed a target-sensitive deep memory network model based on deep memory network with attention mechanism.This model exploited attention mechanism to capture the sentiment information of context corresponding to a given target.Then the interaction module squashed context sentiment expression and interaction information between context and target to the classification features.Finally,classify the features to get sentiment polarity of target.Then deep memory networks combined the TDLSTM and attention mechanism to adaptively generate the context vector of the given target,which aim to solve the problem of expressing context in typical deep memory networks.This paper conducted experiments on the two datasets from SemEval 2014 task4.Attention-based deep memory network achieves better F1-Macro scores up to 5.2%than typical models.Target-sensitive deep memory network outperforms the attentionbased deep memory network by 5.49% at most.TDLSTM-based deep memory network outperforms other models.According to the experimental results,attention mechanism is effective for targeted-context detection,considering the interaction between context and target is useful to identify the context sentiment relies on the concrete target,TDLSTM can generate better context expression vector.