Big Data Based Transfer Learning for Sentiment Classification with Multiple Source Domains

Abstract

Sentiment classification, served as a curial technology in natural language processing and computational linguistics, has drawn a lot of attentions from researchers. However, due to the high cost of manual labeling in the era of big data, conventional methods of sentiment classification are unqualified to be employed in a new domain directly. Hence, in this paper, we explore big data based transfer learning for sentiment classification with multiple source domains. To solve the problem of inherent domain gap, we propose a novel framework Adversarial crossdomain sentiment classification with weighted domain-dependent fEAture learning dubbed AdEA. Specifically, AdEA involves an individual domain-invariant feature extractor and several domain-dependent feature extractors. To obtain the domain-invariant feature, we use a reversed discriminator loss for these extractors. Furthermore, we propose a weighted learning module to reinforce the relationship of domain-dependent features between source and target domains. Integrated with these two domain-related features, AdEA is able to achieve better capability of cross-domain sentiment classification. Experimental results show the effectiveness of our proposed method.

Cross-Domain Sentiment Analysis Based on Feature Projection and Multi-Source Attention in IoT

“… 2ST-UDA [ 13 ]: further utilizes the pseudo labels of the target domain to train a target private extractor on the basis of WS-UDA. AdEA [ 23 ]: utilizes a weighted learning module to strengthen the relationship between domain features. …”

“…AdEA [ 23 ]: utilizes a weighted learning module to strengthen the relationship between domain features.…”

“…Dai et al [ 13 ] directly obtained the classification results of the target instances through the source domain classifier, and the weight assigned to each source domain was acquired by the domain discriminator. On this basis, Li et al [ 23 ] used weighted private features from each source domain to strengthen the learning of private features in the target domain. Furthermore, Zhang et al [ 24 ] found that the more similar domain features are, the more relevant instances are.…”