Sentiment Analysis of Social Media Content in Pashto Language using Deep Learning Algorithms

Sentiment Analysis (SA) has become an active research area due to introduction of social media as it provides content generation facility of its users. Thanks to social media platforms, common people can share their views, opinions and experiences. The main focus of the researchers has been to carry out sentiment analysis in the English language content. Minimal work has been done in the field of SA in content in Pashto language which is the national language of Afghanistan and widely spoken in Pakistan as well. In this research study, our aim is to perform SA in Pashto text of social media content using machine learning and state of the art deep learning algorithms. We exploit various text feature engineering techniques like Term Frequency-Inverse Document frequency, bag-of-words, n-gram, as well as deep features of word2vec, and GloVe. We perform three sets of test subjectivity analysis, binary and tertiary level sentiment classification. Being a pioneer work, we received satisfactory results on self-prepared datasets which is extracted from social media sources. The empirical analysis-based results are evaluated using standard performance evaluation measures such as accuracy, precision, recall and f-measure. Among numerous applied algorithms, Random Forest obtained better results as compared to other algorithms.