In the prior version, we found some minor problems which have been corrected. The revision notes are as follows: 1. Section 4.3: "бн to learn a classifier from target-domain documents and then use it to classify source-domain documents" -> "бн to learn a classifier from source-domain documents and then use it to classify target-domain documents"; 2. Section 5.1: "бн we show empirically that by optimizing Eq. 4.4" -> Eq. 1; 3. Section 6.1: "the second view corresponds to the latent topics obtained by applying бн counts information." -> "the second view corresponds to the topic-based features (i.e. document-topic distributions) obtained by applying probabilistic Latent Semantic Analysis (pLSA)^4 on the term counts information, where the topic number was set to 200."; 4. Proof of Lemma 4.1: "Thus the mutual information between class label and words (or links) remains unchanged during бн" -> "Thus the mutual information between class label and word (or link) clusters remains unchanged during бн"; 5. Add the period to some footnotes missing a period at the end of the sentence.