Implicit Aspect Analysis Approach for Sentiment Classification

Abstract

Implicit aspect-based sentiment analysis (ABSA) is a challenging task that involves identifying sentiments toward aspects that are not explicitly mentioned in text. Unlike explicit ABSA, where aspect terms are directly stated, implicit ABSA requires contextual and linguistic understanding to infer the hidden relationships between words and their associated aspects. This complexity makes it a crucial yet underexplored area in sentiment analysis. This study aims to develop an effective framework for implicit aspect identification and sentiment classification, focusing on hotel reviews as a case study. To achieve this, we propose a two-stage approach: (1) constructing an implicit aspect corpus using Word2Vec Skip-gram and dependency parsing, and (2) developing a binary aspect-based sentiment classifier that associates sentiments with specific hotel aspects, including staff service, cleanliness, value for money, and location convenience. The dataset, sourced from TripAdvisor, consists of 2,000 English-written hotel reviews, carefully annotated by linguistic experts to ensure high-quality aspect and sentiment labeling. The feature representation combines TF-IDF and Word2Vec embeddings, capturing both statistical and semantic relationships within the text. The experimental results demonstrate that deep learning models significantly outperform traditional machine learning methods in implicit aspect identification and sentiment classification. Among all models, LSTM achieved the highest accuracy (88.2%) and F1-score (90.4%) for implicit aspect identification, while SVM with a Linear Kernel achieved the best sentiment classification accuracy (93.0%). A comparative analysis with existing studies highlights that our method achieves higher accuracy and better generalization than previous approaches that relied on dictionary-based, rule-based, and probabilistic techniques. Unlike earlier models, which were often domain-dependent, our approach effectively captures hidden aspect relationships using deep learning techniques. These findings validate the effectiveness of deep learning in implicit aspect-based sentiment analysis, providing a scalable and adaptable solution for sentiment classification tasks. The study contributes to advancing sentiment analysis, implicit aspect detection, and opinion mining, making it highly valuable for businesses and researchers analyzing customer feedback.

-