Elongated words such as “heellloo” or “heyyy” are a frequent feature of oral communication and are often used to underline or exaggerate the hidden message of the root word. Although elongated words are rarely found in written languages and dictionaries, they are common in social networks. They can be considered in the analysis of users' sentiments. In this paper, we analyze the impact of elongation on the classification of sentiments, in addition to an in-depth study at the level of lexical forms of elongation. In this work we present a method to improve the accuracy of sentiment classification based on elongations of features. Experimental results conducted on Twitter data show that our model achieves an accuracy of 0.79 in 7-fold cross-validation experiments.