An apparatus and a method for predicting the pleasantness-unpleasantness index of words are disclosed. The disclosed apparatus includes: a computing unit configured to compute an emotion correlation between a word and one or more comparison word, compute emotion correlations between multiple reference words included in a reference word set and the one or more comparison word, compute multiple first absolute emotion similarity values between the word and the multiple reference words, and compute at least one second absolute emotion similarity value between a reference word and another reference word for all of the reference words included in the reference word set; and a prediction unit configured to predict the pleasantness-unpleasantness index of the word by using the multiple number of first absolute emotion similarity values, the at least one second absolute emotion similarity value, and a preset pleasantness-unpleasantness index of the multiple number of reference words.