The objective of this study is to classify the positive and negative reviews of customers on different products. The effects of hyperparameters in the model used for classification on model performance were compared.
Data preprocessing operations were performed on the reviews. Data preprocessing includes tasks such as removal of punctuation and numbers, normalization of uppercase and lowercase letters, removal of stop words.
The data was prepared for the embedding layer. Tokenization and padding operations were done. The resulting models were trained. 7 different models were used. These models are compared with the activation function, lot size, optimization algorithm, release layer and LSTM layer number, accuracy value.
In this study, Amazon fine food reviews data set was used. Amazon is one of the largest online vendor in the world. People often gaze over the products and reviews of the product before buying the product on amazon itself. The dataset includes the reviews from Oct 1999 to Oct 2012 and has a total of 568,454 reviews on 74,258 products.