Improving Sentiment Analysis of Shopee Reviews with Informal Language and Slang

Abstract

Sentiment analysis (SA) is essential for businesses seeking to understand customer preferences and opinions. It aims to identify and interpret the emotions present in the text by analyzing its linguistic patterns. Along with the difficulty of comprehending formal textual content, it is also essential to consider the informal and online social media languages, which combine English with regional slang to express peoples' genuine feelings. The presence of slang and informal language in customer reviews can compromise the accuracy of SA. This study attempts to improve the accuracy of SA when informal and slang are present, specifically in the context of Shopee customer reviews. The approach employs feature extraction techniques such as N-grams, Term Frequency-Inverse Document Frequency (TF-IDF), and Bag of Words (BoW). Machine learning algorithms such as Support Vector Machine (SVM), Naive Bayes (NB), Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM) are then used to analyze the sentiments of Shopee customer reviews. The results show that the SVM classifier using the 2-gram TF-IDF features outperformed other algorithms. In addition, the study found that incorporating a slang dictionary into SA models improved the classification of informal language and slang, resulting in higher accuracy in sentiment analysis.