Research on Word Sentimental Classification Based on Transductive Learning

At present, word sentimental orientation identification researches mainly fall into two categories: Machine learning and semantic comprehension, machine learning seems to work in specific-field words, but cannot handle general-field words effectively, and semantic comprehension also cannot get ideal scores at precision and recall, therefore, we put forward a fusion of transductive learning and semantic comprehension for determining words´ sentimental orientation. In this paper, we firstly modify Hownet knowledge, on the basis of exist primitives, the fifth primitive -- "sentimental primitive" was proposed, and integrate it into Hownet manually, secondly, we propose a new word sentimental similarity calculation method to compute words´ sentimental value, at last, combine this way with transductive learning for judgment words´ sentimental orientation. The performance of experiments show that, compared with SVM and traditional semantic comprehension, it can get better results.