An Approach for Recommender System Based on Multilevel Sentiment Analysis Using Hybrid Deep Learning Models

From e-commerce websites to social media, Consumers produce a massive volume of product-related reviews, which has a significant potential for commercial business value of marketing objectives. In particular, the online customer reviews can have an impact on the brand image and position. In this context, sentiment analysis (SA) has been used in a recent trend of study to investigate the text reviews and classify customer opinions. Customer reviews may contain sentiment words with varying polarities. The user may have differing feelings about different features of the product, Product managers can be interested in determining customers opinion about each feature of a product for brand management. In this paper, the main contribution is the implementation of an automatic recommender system based on different levels of sentiment analysis using hybrid deep learning methods to identify the polarity-based sentiment features to calculate the sentiment score of a product. The review sentiment score and aspect-level sentiment scores are combined with price and star score to determine the global score of a product. which enable us to rank the products based on their positive features using visualization approaches such as dashboards, assisting marketing persons and customers in their decision-making process. We proved the validity of our approach using customer review (specifically smart phones), extracted from various e-commerce websites attaining acceptable and promising results. Finally, the proposed method for sentiment analysis of customer review and ranking scores are able to improve the progress of recommender systems.