Transformer based multi-head attention network for aspect-based sentiment classification

Aspect-based <span>sentiment classification is vital in helping manufacturers identify the pros and cons of their products features. In latest days, there has been a tremendous surge interest aspect-based sentiment (ABSC). Since it predicts an aspect term polarity sentence rather than whole sentence. Most existing methods have used recurrent neural networks attention mechanisms which fail to capture global dependencies input sequence leads some information loss models for this task, but training these bit tedious. Here, we propose multi-head transformation (MHAT) network MHAT utilizes transformer encoder order minimize time ABSC tasks. First, pre-trained Global vectors word representation (GloVe) embeddings. Second, part-of-speech (POS) features are fused with extract grammatical aspects Whereas most neglected this. Using SemEval 2014 dataset, proposed model consistently outperforms state-of-the-art on tasks.</span>