Abstract

Sentiment analysis is a branch of study that focuses on determining how well subjective literature is written by examining people’s attitudes, views, and sentiments toward various objects. Data analysis and opinion word extraction from the data are difficult tasks, especially when they involve evaluations from entirely unrelated fields. In Aspect Based Sentiment Analysis(ABSA), Aspect word extraction and sentiment of the aspect word extraction are significant job. On these two tasks, numerous methodologies have recently advanced. Few studies, meanwhile, attempt to separate opinion objectives and words as pairs. In this research, we present ABWE (Aspect based word embedding) a sequence labelling subtask for ABSA that tries to filter the appropriate sentiment words for a aspect word. To do this, aspect-fused and context-fused sequence labelling neural network model is created. By using AFCR, the aspect word information is effectively joined into the context words using LSTM network. The global context representation is prepared using Bi-GRU. The encoder creates a final word representation, where the global context information, aspect and opinion information are merged. Using the ABSA benchmark dataset from laptop and restaurant reviews, we construct four datasets for ABWE. The results show that ABWE model provides better results than the other baseline models. We think that our approach may be useful for pair-wise opinion summarization as well as downstream sentiment analysis tasks.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call