Word Polarity Detection Incorporating Intensifiers and Negations

Due to the inexhaustible content posted online every day, performing sentiment analysis has become a prerequisite for organizations seeking deeper insights into their customer base. There are predominantly two methods to conduct sentiment analysis—machine-learning (ML)-based approaches and lexicon-based approaches. Lexicon-based approaches offer the benefits of comprehensibility and transparency over ML-based approaches. Moreover, due to a lack of resource-annotated corpora for various native languages like German, Hindi, Urdu, Turkish, etc., there is an even greater need for improvement in lexicon-based approaches to sentiment analysis. This paper presents a lexicon-based methodology to detect the polarity of a given word as positive, negative, or neutral based on its occurrences in an Amazon product reviews corpus. The effect of intensifiers and negation words present in each review body has also been evaluated. The algorithm has been implemented taking the German language as the starting point, although it is extensible to any language. An F-score of 76% is obtained which is a significant improvement over the existing lexicon-based approaches for the German language.