Sentiment Analysis of English-Punjabi Code Mixed Social Media Content for Agriculture Domain

Abstract

In India, more than 70% of the population is dependent on agriculture. Since the independence of India, the people involved in agriculture mostly stay in rural areas. The government has taken numerous efforts for the improvement of conditions of farmers. Still the condition is not improved to acceptable rate. Currently, it has been easy to extract the reviews of farmers from micro-blogging websites. Since decades, a trend has been seen that multilingual speakers often switch between more than one languages to express themselves on social media networks. Multiple languages are mixed with different rules of grammars, which in itself is the challenging task. In this paper, the authors have extracted the agriculture-related comments having code-mixing property with English-Punjabi mixed content. Further, the performed language identification, normalization, and creation of English-Punjabi code-mixed dictionary. After that, we have tested various models trained on English-Punjabi code mixed data using Support Vector Machine and Naive Bayes techniques for sentiment analysis, tested the pipeline for unigram predictive model. Later experimented for n-gram and performance was found to be better in our implemented model.