baidu/Senta

Implements SKEP (Sentiment Knowledge Enhanced Pre-training), which augments transformer models with unsupervised sentiment knowledge to boost performance across sentence-level classification, aspect-level sentiment analysis, and opinion extraction tasks in both Chinese and English. Built on PaddlePaddle, it provides pre-trained models initialized from ERNIE and RoBERTa with a unified API enabling few-line implementations for multiple sentiment analysis tasks. The framework includes configuration-driven training scripts and production-ready inference tools supporting GPU acceleration.

2,006 stars and 69 monthly downloads. No commits in the last 6 months. Available on PyPI.