Robin-WZQ/Text-sentiment-polarity-judgment

Implements three complementary approaches—rule-based scoring using sentiment/negation/degree lexicons, Naive Bayes statistical classification, and PyTorch deep learning—achieving 68.7%, 88.7%, and 91.2% accuracy respectively on hotel review corpora. The rule-based method constructs sentiment scores through lexicon lookups with negation handling and degree modifier weighting, while handling Chinese text preprocessing via jieba/HanLP tokenization and careful stopword filtering. Provides comparative analysis across multiple sentiment dictionaries (BosonNLP, HowNet) and evaluation metrics (precision, recall, F-score).

No commits in the last 6 months.