Evolutionary approach for building efficient Paraphrase Recognizers

Paraphrasing involves the restatement of a given text to convey the same intent. Paraphrase Recognition systems typically rely on lexical, syntactic and semantic features extracted from the candidate texts to identify equivalence. Though several Paraphrase Recognition systems exist, the performance of these systems has scope for further improvement. This paper reports the work done in designing an efficient Paraphrase Recognition system by using a Support Vector Machine Classifier coupled with Genetic Algorithm based Feature Selection. The developed paraphrase recognizer has exhibited comparable accuracy to the original approach by using only half the number of features.