关键词: 态度挖掘, 情感词识别, 极性判别, 直推式学习, 词典解释

Abstract: In recent years，opinion mining has become one of the hottest topics in the text mining field.It aims to discover the author’s opinion polarity in texts and thus to provide support for decision-making process based on the public opinion.Up to now，most of the algorithms developed for opinion mining use an opinion word dictionary to identify the opinion words occurring in a sentence or text and then determine the polarity of the sentence or text based on polarities of these words.However，it’s not realistic to construct and maintain manually a perfect opinion word dictionary.Therefore，the problem of determining the polarities of Chinese opinion words is investigated，and an algorithm based on transductive learning is proposed to solve the problem.This algorithm uses a few opinion words as seeds and performs polarity transduction from the seeds to other words based on their dictionary interpretations.Compared with supervised learning algorithms，which use the interpretations of the same seeds as the training data，the transduction-based algorithm has obtained an accuracy improvement of about 20%.

Key words: opinion mining, opinion word identification, polarity determination, transductive learning, dictionary interpretation