Improved Feature Representation Using Collaborative Network for Cross-Domain Sentiment Analysis
Keywords:Cross Domain Sentiment Analysis, Domain invariant features, Attention Network, Bi-LSTM, Aspect-based sentiment analysis
Sentiment Analysis task helps us to estimate the opinion of a person from his reviews or comments about a product, person, politics, etc., Cross-Domain Sentiment Analysis (CDSA) empowers the Sentiment models with the ability to forecast the opinion of a review coming from a different domain other than the domain where the model is trained. The challenge of the CDSA model relies on bridging the relationship between words in the source and target domain. Several types of research in CDSA focus on determining the domain invariant features to adapt the model to the target domain, such model shows less focus on aspect terms of the sentence. We propose CWAN (Collaborative Word Attention Network), which integrates aspects and domain invariant features of the sentences to calculate the sentiment. CWAN uses attention networks to capture the domain-independent features and aspects of the sentences. The sentence and aspect attention models are executed collaboratively to determine the sentiment of the sentence. Amazon product review dataset is used in this experiment. The performance of the CWAN model is compared with other baseline CDSA models. The results show that CWAN outperforms other baseline models.
Copyright terms are indicated in the Republic of Lithuania Law on Copyright and Related Rights, Articles 4-37.