CONTEXT-BASED ASPECT LEVEL SENTIMENT ANALYSIS FOR CROSS DOMAIN USING IMPROVED BERT

Abstract

This paper presents efficient contextualizing cross-domain aspect-based sentiment analysis by employing an extended BERT model. The aspect-based sentiment analysis task has been categorized into two different sub-tasks namely: 1) Aspect Extraction and 2) Aspect Based Sentiment Classification. For this purpose, firstly, we extract the aspects (features) from the education performance analysis dataset. Secondly, aspect sentiment classification has been obtained on selected aspects. The existing aspect-based sentiment classifiers are not efficient to extract the local context of words from users’ reviews and comments. To address this issue, a novel context base cross-domain ABSA model has been proposed for extracting local context of words in particular sentences. To achieve this, a pretrained model and transfer learning techniques have been utilized to improve the accuracy of the model. The proposed pretrained model has been validated on well-known semevall-14 and self-synthetic educational dataset respectively for University Faculty performance analysis and other domain categories. The results showed that proposed (CD-ABSA) model outperforms in terms of accuracy and f1-score, and these are 94% and 86%, 92% and 80% respectively on two separate datasets as compa red to State-of-the-Art similar research with extended BERT.