With the growing popularity of online platforms, purchasing products online has become a more convenient way of shopping. In this area, feedback given by users regarding the products purchased needs to be structured and analyzed thoroughly to extract useful and meaningful information. This analysis helps both customers to decide on which product to buy and also the manufacturers to improve their product. Thus, Item Rating prediction has become an interesting research topic that has a wide range of applications. The traditional lexicon-based sentiment analysis considers polarity scores of words but ignores the differences among aspects. There are some other existing techniques for Item Rating prediction methods (i.e., Sentiment strength-based methods, Matrix Factorization, Deep neural network-based methods ) which achieved some noticeable and useful results. These methods need extensive training with manual intervention and effort and also they suffer from cold start problems. In order to overcome these limitations, This is an unsupervised framework for Item Rating Prediction.
In this framework, we use LDA to extract topics from reviews and label each topic with aspect name. Later, we estimated aspect ratings and aspect weights of each product to predict the overall rating of a product. The overall rating of a product is the linear summation of the product of aspect ratings and their corresponding aspect importance. We made extensive experiments on Amazon unlocked mobile phones dataset containing around 4,00,000 Reviews and demonstrated the effectiveness of the unsupervised framework.
The major objectives of this research work are the following : 1.To determine the salient aspects in the dataset. 2.o detect the users’ sentiment towards a product in the dataset. 3.To identify the importance of the aspects with respect to users’ feedback. 4.To calculate the overall rating of the product of the test dataset. 5.To evaluate the performance of the proposed approach.