Structural Feature Transfer Crossing Categories for Online Comments Based on Public Feature Space

Labeling tasks rely upon classification models. These models have to be trained and adjusted repeatedly before they perform a adequate precision and recall. The modeling is expensive in time and expertise knowledge and dominating only in a particular domain or category. This paper presents a method that, given some corresponding features in a source domain and a target domain, transfers the existing classifier from source domain to target domain. The proposed method is based on ADMI measurement, Word2Vec vectorization and PFS (public feature space) construction. The resulting scheme FTAC accomplishes this transfer when the similarity of source and target domain is higher than a threshold and is capable of successfully introducing the classifier for the source domain to the target domain that are acquired from considerably different online shopping platforms. Experimental results are presented, which demonstrate that the performance of the proposed method compares favorably to that of former approaches.