PLStream Capabilities Analysis 📊

This project explored the linguistic capabilities of PLStream, a novel unsupervised framework for fast polarity labeling of massive data streams. In the project we analyzed how PLStream compares against state-of-the-art supervised models using the CheckList framework. The framework assessed linguistic performance across various capabilities, such as negations, robustness to typos, and invariance tests.

The results revealed PLStream's strong suitability for handling high-velocity data streams with minimal latency and high throughput. However, limitations included its inability to process contextual or temporal information and the presence of training data biases, leading to reduced performance in fairness tests.

You can access the project's repository on GitHub.