Sentiment Analysis of Code-mixed Roman Urdu-English Social Media Text

Abstract

Evaluation of digital and social landscape continues in unexpected ways. The internet continuously grows as individuals discover new means to consume information. Internet platforms led the charge for Internet usage.  Almost 60% of the world’s population are active social media users who come to rely on social media platforms to satisfy their daily data needs. Every day, billions of text messages are produced on social media sites like Facebook, Instagram, and Twitter, creating a substantial amount of data from people all around the world. Businesses and organizations use text analysis tool in social media monitoring to learn how the public views their brand, goods, and services.   Sentiment analysis is one of the most popular applications of text analysis in social media monitoring. This method classifies text as neutral, negative, or positive using natural language processing (NLP) techniques. Sentiment analysis can be used to monitor changes over time and spot trends in customer sentiment. This can be helpful for tracking the success of marketing activities and for spotting possible problems or opportunities for a business.

The use of code-mixed text on social media platforms serves as a key communication method for multilingual users. It enables the integration of more than one language within a single message, which often mirrors cultural identity, conveys nuanced meanings, and promotes effective communication among diverse linguistic groups, especially in areas where multiple languages are prevalent. However, due to the intricate interactions between multiple languages, such as word meaning ambiguity, grammatical variations, and the scarcity of easily accessible labeled datasets, code-mixed text sentiment analysis presents numerous challenges that make it challenging to accurately assess the sentiment of a text.  This paper focuses on leveraging Long Short-Term Memory (LSTM) networks for sentiment analysis of Urdu-English code-mixed text. By addressing the inherent challenges and optimizing LSTM models, this study contributes to developing effective sentiment classification frameworks for this underrepresented language pair.