Abstract:

Abstract In the past few years, more and more researchers have focused on the field of natural language processing and, more specifically, on aspect-level sentiment analysis. Sentiment analysis is often used to analyse people's opinions, or feelings about different entities like products or services. Given the immense amount of data generated daily in various forms on the web, sentiment analysis has become one of the most active areas of research today. In turn, online user reviews are considered a powerful marketing tool and have attracted widespread attention from marketers and academics. This motivates the current study, which focuses on sentiment analysis using four machine learning models, namely recurrent neural networks (LSTM, GRU, Bi-LSTM), and a pre-trained language model (Roberta). The models are trained to categorize customer reviews on online platforms as positive or negative. Then the best model that shows the best results is selected by evaluating each of them based on accuracy, precision, recall, and F1 Score. Finally, a topic modeling technique is used to reveal various topics present in the data and determine what’s pushed the customer to give such a review, as well as provide suppliers with the right decision for each scenario case. Although the approach proposed in this study is applied to analyse the opinions of customers towards the products of a marketplace and reveals different topics present in it, it can be used in any field to know the writer's point of view on, e.g., government policy, individuals, brands, etc.

Keywords: