Extraction Of The Sentiment Elements From Chinese Microblogs Based On CRF And Noun Phrase Recognition

As the development of information technology, it is increasingly important to extract valuable information from mass data. Microblog, which is a new sociality platform appeared in recent years, has a huge number of users. Besides releasing information proactively, users can take part in discussions by using topics. The authors’ subjective viewpoints would be included in many discussions of valuable topics. The purpose of this paper is to extract sentiment elements of Chinese microblogs which include emotional objects and the polarities of these objects.In Chinese microblogs, One microblog or even one sentence may include more than one emotional object, so machine learning methods are difficult to give good results. In this paper, the polarities of the extracted emotional objects are determined by emotional units which are matched by the sentiment word dictionary and degree adverb dictionary.CRF model is used to extract emotional objects by using words, parts of speech, emotional words and dependency information as features. The closed experiments show good results, but in the open experiments the method based on CRF model performs poorly. One reason is that the training corpus is not large enough. Because of the high costs of manual tagging corpus, it is difficult to expand the size of training corpus.To solve this problem, a novel approach, based on the noun phrase recognition, is proposed to generate the emotional object candidates automatically. After the generation, the dependency information is utilized to filter the emotional object candidates. Experiments show that this method can obviously improve the recall, as well as the F-value, of extraction of emotional elements in the open experiments.