Tasks and Datasets
The framework currently supports the following text classification tasks:
🧠Supported Tasks
💬 Sentiment Analysis
- Determines the sentiment expressed in a text, such as positive, negative, or neutral.
- Commonly used in applications like product reviews, social media analysis, and customer feedback.
🚫 Hate Speech Detection
- Identifies and classifies text containing hate speech, offensive language, or harmful content.
- Essential for moderating online platforms and ensuring safe digital environments.
🔗 Natural Language Inference (NLI)
- Determines the logical relationship between two sentences (e.g., premise and hypothesis).
- Tasks include identifying entailment, contradiction, or neutrality between sentences.
- Useful for applications like question answering, summarization, and reasoning tasks.
📚 Dataset Support
EvalxNLP includes a representative dataset for each of the supported tasks and allows users to extend the framework with additional classification datasets. All datasets are rationale-annotated, meaning they include human-annotated rationales that highlight the most important words or sentences for a given class label. These rationales enable the evaluation of alignment between model explanations and human understanding.
🎬 MovieReviews
Task: Sentiment Analysis
Description: Contains 1,000 positive and 1,000 negative movie reviews. Each review includes phrase-level human-annotated rationales that justify the sentiment label.
📢 HateXplain
Task: Hate Speech Detection
Description: Comprises 20,000 posts from Gab and Twitter, annotated with one of three labels: hate speech, offensive, or normal.
📄 e-SNLI
Task: Natural Language Inference
Description: Contains 549,367 examples split into training, validation, and test sets. Each example includes a premise and a hypothesis, annotated with one of three labels: entailment, contradiction, or neutral.