Polarity and Similarity Measures Towards Classifying an Article on Food Insecurity

Abstract

Polarity measure has been applied to text mining tasks such as sentiment analysis and text classification. At the same time similarity measure has also been applied to text mining tasks such as summarization, classification and information retrieval. Limited studies if any have used both measures as predictors in a Machine Learning task to automatically identifying articles with conversations on food insecurity in a news feed. These conversations on food insecurity can be generated into trends. Now these trends can guide stakeholders in taking appropriate action depending on food insecurity situation. The proposed strategy relates to a study that used polarity measures for tweets on food prices to predict actual food prices which could also provide proxy information on food insecurity to similarly guide the stakeholders. However our study is on blending polarity and similarity as handcrafted predictors in Machine Learning to automatically label articles on food insecurity using Machine Learning. To explore the proposed strategy, the study used articles from Monitor site (www.monitor.co.ug), a ugandan news media. Promising findings were obtained with KNN and N-BAYES classifiers which had AUC measures of 0.931 and 0.927 respectively compared to random classifier of 0.91. Future work considers blending these handcrafted features with automatic features from deep learning to explore performance improvement.