Natural language processing for sentiment analysis

Abstract

In today’s business and technology world, data analytics has become indispensable. Technologies such as sentiment analysis allow companies and corporations to gain key insights about consumer perceptions and opinions that can help shape effective strategic and marketing decisions. Advancements in the fields of natural language processing and machine learning have resulted in sentiment analysis algorithms that are capable of extracting sentiment orientation in relation to a specific topic from consumer generated text inputs. The aim of this project is to implement a sentiment analysis algorithm capable of classifying consumer generated tweets as having positive, neutral or negative sentiment orientation with respect to the subject of interest, and exploring its real-world applicability. In this dissertation, two training datasets are used related to the topics of climate change and technology products respectively. However, the proposed sentiment analysis algorithm can be trained on and applied to any topic domain. The aim of this project is achieved by developing a theoretical framework based on contemporary literature and selecting an appropriate machine learning approach and architecture. The performance of various deep learning models is then evaluated and compared to benchmark and commercially available algorithms. Finally, the stand-out deep learning model is implemented on a cloud platform. Results show that the performance of the proposed deep learning sentiment analysis algorithm compares favourably with both commercially available software and contemporary literature. The discrepancy in performance of the sentiment analysis algorithm when applied to the different datasets suggests that contemporary approaches to sentiment analysis cannot effectively extract sentiment from more complex topic domains such as climate change.