An Intelligent Unsupervised Approach for Handling Context-Dependent Words in Urdu Sentiment Analysis

Abstract

The characteristic of context dependency in Urdu words needs to be handled carefully while performing Urdu sentiment analysis. In this research, an already constructed Urdu sentiment lexicon of positive and negative words is further expanded by the addition of context-dependent words. These context-dependent words are used with or without conjunctions. Rules are formulated for assigning polarities to those context-dependent words that are surrounded by the positive or negative words. These rules were incorporated in the Urdu sentiment analyzer. Fusion of these rules for handling context-dependent words and the expanded Urdu sentiment lexicon resulted in increasing the accuracy of the Urdu sentiment analyzer from 83.43% to 89.03% with 0.8655 precision, 0.9053 recall, and 0.8799 F-measure, which is a statistically significant improvement.