Transfer learning in classification based on semantic analysis

Traditional classification methods, such as supported vector machine and naive bays, rely much on high quality labeled data. However, in many real world applications, labeled data are in short supply. It often happens that obtaining labeled data in a new domain is expensive and time consuming, while there may be plenty of labeled data from related but different domains. In this paper we proposed a novel transfer learning approach to address the classification task while no labeled data available in the target domain and labeled data in related domains available, by selecting features with the similar semantic meanings in both source and target domains. The semantic meanings of words in different domains are extracted through semantic nets built using a complex network model named AF and cosine similarity is used to calculate the semantic similarity of words.