Skip to Main Content
A fundamental assumption in machine learning is that the data distributions of the training and the test sets should be identical. When the assumption does not hold, the traditional machine learning algorithms might perform worse. In this paper, we tackle this transfer learning problem by implementing a general graph ranking framework for a sentiment classification task. We construct a fusion graph model by using the in-domain and the out-of-domain data. The in-domain data can help us to get pseudo labels of the out-of-domain data. The out-of-domain data can help us to update the labels and can get the convincing prediction labels. Experimental results show the significant improvements in accuracy and demonstrate the effectiveness of this algorithm.
Date of Conference: 29-31 May 2012