Research On The Adaptation Of Emotional Classification Field Based On Semi - Supervised Machine Learning

In recent years, the sentiment classification has become a research hotspot in the field of domestic and foreign relevant, obtained the further research of many scholars, but it still has a lot of problems need to solve.This article mainly solves the two problems in sentiment classification, one is the unbalanced sentiment classification problem, the second is domain adaptation of sentiment classification.When the training sample categories are unbalanced, the classifier will be more and more tend to a certain category, which will lead to worse results. This is called unbalanced sentiment classification problem. In this paper, on the basis of the EM algorithm an improved algorithm is proposed to solve this problem. And through the experiment we can find that the proposed algorithm is superior to ordinary EM algorithm. The experiment proves the validity and rationality of our algorithm.In a particular area (which we call the source field) marked on the sample set learning classifiers usually only get on a test sample of the same good performance in the field, for the other field (we call target areas), especially in the field of distributed source field goal difference is large, the algorithm performance will be greatly reduced. This is called domain adaptation problem. This paper presents a field of adaptation solution chosen by some representative features in the source areas to be modeled, and then using the improved algorithm to classify the samples by comparing the experimental analysis shows that the proposed article the method has significantly improved the sentiment classification performance.