With the explosive growth of user-generated content on the web, sentiment analysis becomes a strenuous job. The fine-grained sentiment analysis of user reviews enables the product manufacturers and service providers to improve decision-making in business. There is a demand for sentiment analysis for mining the voluminous social content. The traditional opinion mining methods are struggling to understand the exact opinion of the people from user feedback. The deep learning algorithms are a promising alternative to a traditional opinion mining algorithm. However, current deep learning-based opinion mining methods attempt to identify the opinion of a user regardless of the aspects. Therefore, there is a demand for aspect-based sentiment analysis for performing fine-grained analysis on natural language text data. This paper emphasizes on aspect-based sentiment analysis by using the deep learning model. There is three core process involving in the proposed approach. These are finegrained aspect extraction, opinion word discovery, and classification of sentiment based on aspects and opinions. Initially, the proposed method cleans the data using preprocessing methods for transforming data into an efficient format. The next step is the fine-grained aspect extraction in which the proposed approach uses an enhanced Latent Dirichlet Allocation (LDA) model to extract a set of relevant aspects. As the second step, opinion discovery focused on extracting the opinionated terms to generate the aspect-opinion pair. The final step is deep learning-based sentiment classification. During the decision-making task, the proposed approach employs sets of seed words and negation attention mechanism for achieving promising results. Finally, the proposed approach is evaluated using the product review dataset to demonstrate the performance of the proposed approach in terms of precision, recall, and fmeasure. Accordingly, when varying the number of product reviews, the proposed approach. Accordingly, the DLASA approach obtains 22.38% of the higher precision score compared to the existing NBSA approach while increasing the number of product reviews from 500 to 2500.
Volume 12 | 04-Special Issue
Pages: 1457-1465
DOI: 10.5373/JARDCS/V12SP4/20201624