Aspect-Based-Sentiment-Analysis

Why we included this project

This package answers a more specific question than plain sentiment analysis: instead of one overall verdict, it gives you a sentiment label for each aspect mentioned in a text. You pass in the text and a list of aspects, and the transformer-based pipeline returns separate judgments, so a long review about Slack can yield distinct results for 'slack' and 'price'. The explanation layer is what makes it stand out. An auxiliary 'professor' component inspects the model's attention and internal states, can flag suspicious predictions, and surfaces the token patterns behind each decision, which matters for teams that need to audit or trust sentiment output rather than treat the model as a black box. The pipeline is modular too, so you can swap in your own fine-tuned model or adjust how documents are split into sentences when your data differs from the bundled checkpoints.