Enhancing Sentiment Analysis in Short Texts with POS-Embedded LSTM Models
Shaurya Gulati · 2024
This research paper focuses on sentiment analysis, aiming to classify short texts, such as tweets, into positive or negative sentiments. The authors present a comprehensive approach that starts with baseline methods, including Naive Bayes and embedding-based models. They then propose improvements incrementally, experimenting with word embeddings, principal component analysis (PCA), and part-of-speech (POS) embeddings. The study’s findings highlight the impact of embedding size on accuracy and show that the inclusion of POS information improves sentiment classification. The proposed POS-enhanced model exhibits enhanced validation accuracy and reduced loss. Although the paper includes some discrepancies in reported validation accuracies, the overall conclusion underscores the value of considering POS information to enhance sentiment analysis in short texts.