Developing Lexicons for Enhanced Sentiment Analysis in Software Engineering: An Innovative Multilingual Approach for Social Media Reviews

Sentiment analysis is becoming increasingly important in today’s digital age, with social media being a significant source of user-generated content. The development of sentiment lexicons that can support languages other than English is a challenging task, especially for analyzing sentiment analysis in social media reviews. Most existing sentiment analysis systems focus on English, leaving a significant research gap in other languages due to limited resources and tools. This research aims to address this gap by building a sentiment lexicon for local languages, which is then used with a machine learning algorithm for efficient sentiment analysis. In the first step, a lexicon is developed that includes five languages: Urdu, Roman Urdu, Pashto, Roman Pashto, and English. The sentiment scores from SentiWordNet are associated with each word in the lexicon to produce an effective sentiment score. In the second step, a naive Bayesian algorithm is applied to the developed lexicon for efficient sentiment analysis of Roman Pashto. Both the sentiment lexicon and sentiment analysis steps were evaluated using information retrieval metrics, with an accuracy score of 0.89 for the sentiment lexicon and 0.83 for the sentiment analysis. The results showcase the potential for improving software engineering tasks related to user feedback analysis and product development.