A vector space model for subjectivity classification in urdu aided by co-training

The goal of this work is to produce a classifier that can distinguish subjective sentences from objective sentences for the Urdu language. The amount of labeled data required for training automatic classifiers can be highly imbalanced especially in the multilingual paradigm as generating annotations is an expensive task. In this work, we propose a cotraining approach for subjectivity analysis in the Urdu language that augments the positive set (subjective set) and generates a negative set (objective set) devoid of all samples close to the positive ones. Using the data set thus generated for training, we conduct experiments based on SVM and VSM algorithms, and show that our modified VSM based approach works remarkably well as a sentence level subjectivity classifier.