Deep Learning, WordNet, and spaCy based Hybrid Method for Detection of Implicit Aspects for Sentiment Analysis
Piyush Soni, Radha Krishna Rambola · 2021 International Conference on Intelligent Technologies (CONIT) · 2021
In the past decade, sentiment analysis/opinion mining has become a very enticing research domain. Sentiment analysis is the task of identification, extraction, and aggregation of sentiments present in the given text. Sentiment analysis may be performed at different granularity. Aspect-level sentiment analysis being the most fine-grained, and sentiments are aggregated for different aspects of an entity. Most of the current research work focuses on detecting explicit aspects, and a handful of attempts were made to detect implicit aspects that the context may imply.Most of the existing work for implicit aspect detection is based on calculating collocation from a corpus. The emergence of deep learning and its suitability for tasks similar to aspect detection has motivated us to use it for implicit aspect detection. Based on the study of existing research, we have also noticed that using an existing linguistic knowledge base like WordNet, results in significant improvement in efficiency. Hence, we have proposed a hybrid method incorporating a Recurrent Neural Network(RNN), which is trained on a dataset prepared by us and similarity calculations based on WordNet and similarity function from spaCy to detect implicit aspects. When tested on our dataset, it gives us reasonable results.