Learning to Rank with BERT for Argument Quality Evaluation

The task of argument quality ranking, which identifies the free text arguments, remains, to this day, a challenge. While most state-of-the-art initiatives use point-wise ranking methods and predict an absolute score for each argument, we instead focus on learning how order them by their relative convincingness, experimenting with several learning-to-rank quality. We leverage BERT's powerful ability in building representation paired approaches (point-wise, pairwise, list-wise) rank arguments according measure convincingness. also demonstrate ensemble models trained different losses often improves performance at identifying convincing list. Finally, compare BERT coupled all major datasets available task, demonstrating approach generally performs better outlining topmost arguments.