Sentiment Analysis on Amazon reviews helps to understand the opinions of users about various products listed on that platform. Various existing techniques were applied to analyze the sentiment of users about the products and existing methods have the limitation of overfitting and imbalance data problems. This research proposes the Attention Weighting – Long Short-Term Memory (AW-LSTM) model to focus on features in the network and improve the performance of sentiment analysis. The AW-LSTM model is tested on the Amazon review dataset and compared with existing techniques in sentiment analysis. In attention weighting, word embedding, global pooling, the importance of word order, and short sentence processing were performed. Global pooling technique helps to sample the input data that helps to handle imbalance data problems. The importance of word order helps to analyze the meaning of the sentence and Named Entity Recognition (NER) based short sentence process extract information from a short sentence. The advantages of the AW-LSTM model improve the performance of the sentiment analysis on Amazon review dataset. The AW-LSTM model has accuracy of 93.28 % and the existing BERT has 88.48 % accuracy in sentiment analysis of the Amazon review dataset.

Amazon review, Attention Weighting, Long Short-Term Memory, Named Entity Recognition, and Sentiment Analysis.