quickSentiment 0.3.1

Major Changes & API Updates

• pipeline() now accepts raw atomic vectors (text_vector and sentiment_vector) instead of full dataframes, drastically improving memory efficiency.
• Renamed the prediction() function to predict_sentiment() to prevent namespace collisions with base R generic functions.

Performance & Dependency Improvements

• Removed the caret dependency entirely. Cross-validation folds and confusion matrix evaluations are now handled via lightweight custom implementations.
• The vectorization pipeline now strictly utilizes dgCMatrix sparse matrices, ensuring the package scales efficiently for large text datasets.

New Features

• Models now return explicit class probabilities alongside the final predicted categories.
• Added automated ROC/AUC calculations for binary classification. The pipeline now returns an optimized threshold, which predict_sentiment() utilizes by default.
• Users can now manually define classification thresholds during inference.
• Upgraded text preprocessing capabilities with advanced arguments for custom stop words, retained words, and default corrections for quanteda stop word dictionaries.

Bug Fixes

• Resolved an initialization bug that incorrectly prevented the use of the Naive Bayes (nb) model.
• Fixed a compatibility error that occurred when pairing TF-IDF vectorization with Random Forest (rf) models.