ELVISIO/ee5438-g36-roberta-large-joint-boundary

Joint boundary scorer for Tweet Sentiment Extraction

Status: trained, evaluated, and released for reproducibility.

This repository is reserved for the EE5438 Group 36 joint scorer. It will rank up to eight valid spans proposed by the RoBERTa-large baseline. A small scoring head will use the start and end representations, their interaction, and the supplied sentiment representation. The trained encoder and candidate list will remain fixed when independent and joint scores are compared.

This scorer depends on the exact baseline encoder checkpoint and candidate-generation contract. OOF validation scores for seeds 42, 43, and 44 are 0.70486, 0.70184, and 0.70367 (mean 0.70346, sample standard deviation 0.00152). The selected seed-42 head scores 0.71120 on the held-out local test. These are local split measurements, not Kaggle leaderboard claims.

The released artifact is the selected seed-42 head. It is an add-on to the baseline encoder and requires the baseline checkpoint and candidate-generation contract. The upstream parent license and terms apply. Reported scores are local split measurements, not Kaggle leaderboard claims.

Baseline model: https://huggingface.co/ELVISIO/ee5438-g36-roberta-large-span

Project code and plan: https://github.com/ELVISIO119/ee5438-group36-tweet-sentiment-extraction