แนวคิดหลัก
Legion improves Pre-trained Language Models for GitHub topic recommendation.
สถิติ
BERT의 F1 점수는 Head에서 0.409, Mid에서 0.081, Tail에서 0.0이다.
ELECTRA의 F1 점수는 Head에서 0.358, Mid에서 0.0, Tail에서 0.0이다.
คำพูด
"Legion can improve vanilla PTMs by up to 26% on recommending GitHubs topics."
"Legion employs a filter to eliminate vague recommendations, thereby improving the precision of PTMs."