Evaluating Cross-Domain Sentiment Analysis using Convolutional Neural Network for Amazon Dataset

Abstract

Sentiment Analysis (SA) has garnered extensive research attention over the past decades as a means to comprehend users' attitudes and opinions in various domains. With the proliferation of online communities and the rapid generation of social media content, understanding sentiments has become crucial for decision-makers and stakeholders. Cross-Domain Sentiment Analysis (CSDA) is the process of analysing and interpreting sentiments in text data across different subject areas or contexts, accounting for the varying nuances and contextual differences in sentiment expression. The problem of CDSA poses a significant challenge in the field of Natural Language Processing (NLP), as the sentiment polarity of words and expressions can vary drastically across different domains. For instance, a word like "unpredictable" can convey positive sentiment in the context of a movie review but may signify negative sentiment when referring to the performance of a computer system. Deep Learning (DL), a subfield of machine learning, has shown promising results in various domains since its emergence in 2006, especially in complex problem-solving involving vast datasets. This paper aims to evaluate CDSA performance using Convolutional Neural Network (CNN) on the Amazon dataset. The study builds upon our previous research that highlighted the limitations of classical Machine Learning (ML) approaches for CDSA. The result demonstrates that the DL model is the state-of-the-art in machine learning classification tasks even though with a limited features engineering task. In conclusion, understanding people's opinions across different subjects on the internet is crucial but complex and using advanced Deep Learning methods like the Convolutional Neural Network can help address these challenges effectively.