Design of Deep Learning Frameworks for Aspect Based Opinion Mining on User Generated Content in Electronic Media

Abstract

Aspect-Based Sentiment Analysis is becoming increasingly important for businesses because they want to understand how customers feel about different aspects of their products or services. This information helps them improve their competitiveness, make well-informed changes, and ultimately provide a better experience for customers. However, current business approaches struggle to recognize certain issues like losing meaning in data, understanding context-specific information, and identifying the specific things customers have positive or negative feelings about. To tackle these problems, this research proposes new deep learning methods that are designed to overcome the current limitations of aspect-based sentiment analysis. These methods aim to better analyse and understand customer sentiment in a more accurate and comprehensive way. newlineA distributed framework is developed to perform opinion mining by combining latent semantic knowledge and valid opinion information. To achieve this, data from tweets is gathered using the Twitter API. The N-LDA technique is used to capture latent semantic knowledge, while the LSTM approach identifies relevant aspects and sequences within the tasks. The effectiveness of this framework is tested on the collected dataset and two additional benchmark datasets such as health news tweets and PubMed medical abstracts. Further, various deep learning models are employed for evaluation and the obtained results demonstrate that the proposed framework outperforms existing methods by achieving accuracies of 97.84%, 88.37%, and 84.10% on the respective datasets. newlineThe Deep Refinement Network is a designed to improve the ability for extracting meaningful features. This is achieved by combining context-aware aspect knowledge with sentiment information. The network includes a multi-channel module that captures context-related aspects based on specific targets, uncovering hidden local patterns to predict sentiment polarity. newline