Abstract:

For determining the polarity of sentiment words, an algorithm based on PageRank technology is proposed. A graph constructed whose nodes consist of unlabeled sentiment words and a few sentiment seeds, and the weights between each two nodes based on the semantic similarity of HowNet are also gained. With the PageRank technology on those seeds the polarity of the unlabeled sentiment words can be obtained. Compared with the methods based on HowNet to judging polarity of sentiment words, the proposed algorithm of combining PageRank technology shows its effectiveness by 10% increase of the precision.

Key words: natural language processing, semantic orientation, PageRank model, HowNet