Phrase Aspect Polarity Lexicon for Enhanced Multi-Aspect Polarity Extraction in Multi-Layer Form

Abstrak

Feedback is indispensable for informed management decisions, but analyzing its can be challenging. That’s where Aspect-Base Sentiment Analysis (ABSA) is very effective in finding valuable insights from text data. Polarity extraction is critical step in ABSA, which is performed using lexicon. However, current lexicon considers single word (unigram) often missing the context of phrase or compound term. This study propose a Multi-Layer Phrase Aspect Polarity Extraction using Lexicon (MLPAPEL), an n-gram-based lexicon designed to improve aspect-level sentiment analysis. MLPAPEL is inserted between pre-processing and polarity extraction in the ABSA framework. Plus, it compatible with popular public lexicons like SentiWordNet, AFINN, TextBlob, VADER and InSet. This study applies MLPAPEL performance to evaluate higher education feedback focusing in on areas like pedagogy, professionalsm, personality, socaial and facilities. They classified sentiment as positive, neutral or negative. The results found that MLPAPEL boosted F1-score across the board: 9.33% with SentiWordNet, 12.33% for both AFINN and VADER, 12.67% with TextBlob, and 9.33% for InSet. Additionally, MLPAPEL worked faster too, bumping up both precision and recall for aspect-level sentiment analysis. Ultimately, MLPAPEL isn’t just another rule-based tool, but a new outomated approach to optimizing sentiment labeling in complex text, pushing ABSA forward both theory and practice.

Identifier

• DOI: https://doi.org/10.62527/joiv.10.2.3677
• OpenAlex: https://openalex.org/W7154508136