Zhang Qingliang, Xu Jian
To improve the efficiency of extracting sentiment words and building sentiment lexicon, the authors propose a method to extract a set of basic sentiment words, and then to calculate both the PMI-IR value between candidate word and the positive basic sentiment word set and the PMI-IR value between candidate words and the negative basic sentiment word set, to judge the orientation of a candidate word.Taking account of frequency, orientation, intensity and definiteness of words, computers are able to finish most of the work. It improves the efficiency and reduces cost of building sentiment lexicon. Experiment is processed on the dataset constituted with 71 061 reviews from 360buy and 1 736 reviews from Joyo. With the dataset, the method achieves a recall rate of 76.36%, a precision of 76.94%,and the precision of sentiment orientation is 62.70%.
