People’s ability to quickly convey their thoughts, or opinions, on various services or items has improved as Web 2.0 has evolved. This is to look at the public perceptions expressed in the reviews. Aspect-based sentiment analysis (ABSA) deemed to receive a set of texts (e.g., product reviews or online reviews) and identify the opinion-target (aspect) within each review. Contemporary aspect-based sentiment analysis systems, like the aspect categorization, rely predominantly on lexicon-based, or manually labelled seeds that is being incorporated into the topic models. And using either handcrafted rules or pre-labelled clues for performing implicit aspect detection. These constraints are restricted to a particular domain or language which is domain-dependent. In this work, we first propose a novel unsupervised probabilistic model Topic-seeds Latent Dirichlet Allocation (TSLDA) that leverages semantic regularities for the articulation of explicit aspect-categories. Then, based on the articulated categories, a distributed vector is used for the identification of implicit aspects. The experimental results show that our approach outperforms baseline methods for different domain-data with minimal configurations. Specifically, utilizing the RI measure, our proposed TSLDA outperformed multiple clustering and topic models by an average of 0.83% in diverse domain-data, and roughly 0.89% using the Precision metric for implicit aspect detection.
