Exploring Ensemble of Models in Taxonomy-based Cross-Domain Sentiment Classification

Abstract

Most cross-domain sentiment classification techniques consider a domain as a whole set of opinionated instances for training. However, many online shopping websites organize their data in terms of taxonomy. With multiple domains (or, nodes) organized in a tree-structured representation, we propose a general ensemble algorithm which takes into account: 1) the model application, 2) the model weight and 3) the strategies for selecting the most related models with respect to a target node. The traditional sentiment classification technique SVM and the transfer learning algorithm Spectral Features Alignment (SFA) were applied as our model applications. In addition, the model weight takes the tree information and the similarity between domains into account. Finally, two strategies, cosine function and taxonomy-based regression model (TBRM) are proposed to select the most related models with respect to a target node. Experimental results showed both (cosine function and TBRM) proposed strategies outperform two baselines on an Amazon dataset. Three tasks of the proposed methods surpass the gold standard generated by the in-domain classifiers trained on the labeled data from the target nodes. Good results from the three tasks enable this algorithm to shed some new light on eliminating the major difficulties in transfer learning research: the distribution gap.

Cross‐domain sentiment classification using decoding‐enhanced bidirectional encoder representations from transformers with disentangled attention

“…SFA method is adopted by Lin et al and proposed two methods cosine function and taxonomy-based regression model (TBRM) for selecting the more identical model based on the target node. 39 Recently, to align the domain-specific and domain-independent features for CDSC, the author utilized the technique of SFA method and introduced modified maximum entropy along with bipartite graph clustering. 40 The limitations of these approaches are: (i) heavily dependent on the labeled dataset and (ii) manually selection of pivot features.…”

Cross‐domain opinion classification via aspect analysis and attention sharing mechanism

“…SFA technique mainly focused on mutual information between domains and features. The SFA method is utilized by Lin et al 12 and introduced two techniques to select the closest resembling model using target node named taxonomy‐based regression model (TBRM) and cosine function. Further, the technique of Blitzer et al 8 is adapted by Yu and Jiang 13 and utilized the neural network to address the cross‐domain opinion classification problem.…”