Vote-Flip

Vote Flip Algorithm is a method proposed by Choi and Cardie in their paper Adapting a Polarity Lexicon using Integer Linear Programming for Domain-Specific Sentiment Classification.

The original algorithm from above paper is shown below. It is a simple algorithm that works by counting the number of negators in the expression to determine the overall polarity.

The flag fFlipPolarity determines whether the overall majority polarity based on the number of negators in the given expression should be flipped. If the positive (or negative) polarity words dominate the given expression, and if there is no need to flip the majority polarity, then the positive (or negative) polarity is taken as the overall polarity. If the positive (or negative) polarity words dominate the given expression, and if we need to flip the majority polarity, then we take the negative (or neutral) polarity as the overall polarity. Notice that the result of flipping the negative polarity is neutral, not positive. which was found to work better than flipping the negative polarity to positive. Finally, if the number of positive words and the negative words tie, and there is any neutral word, then neutral polarity is assigned. Negators are neglected as flipping a neutral polarity would still result in a neutral polarity. If none of above condition is met, than the polarity defaults to the most prominent polarity of the data.