关键词: 分类, 似领域, 上下文, 共享特征, 特征迁移学习

Abstract: Distortion and low efficiency are two constant problems when employing traditional context in classification problems. Inspired by the transfer learning theory, the paper regards the similar domain of the target domain as context, and constructs the low-dimensional common features between the target domain and its similar domain by structural correspondence learning method. Based on the common features between similar domains, the paper puts forward a new classification model. The experiment employs users’ personalized options of QQ-zone as context to classify users’ preferences of e-commerce web pages, the results verify the feasibility and availability of the proposed model.

Key words: classification, similar domain, context, common feature, feature-based transfer learning