Stance Detection in Manipuri Editorial Article Using CRF

The primary purpose of stance detection is to categorize the author’s attitude as favourable to, antagonistic towards, or neutral to a specific target. Most stance detection work available is on short text emphasizing the English Language and less on targeting long text, especially in low-resource languages like Manipuri. Based on the article “Citizenship Amendment Bill (CAB)” this experiment performs stance detection in Manipuri. Here, a model is created through using Conditional Random Field (CRF) to decide each word’s polarity, and the article’s stance is then decided. The best features recorded for identifying a word’s polarity is given in the form of Precision: 61.78%, Recall: 63.23%, and F_Score: 62.49%. The model gives the result with a mean accuracy of 56.66%.