Temel Kavramlar
MasonTigers presented an ensemble approach for Semantic Textual Relatedness at SemEval-2024, achieving notable rankings across different languages and tracks.
Özet
1. Introduction
Understanding semantic relatedness is crucial in NLP.
Various applications benefit from modeling semantic relatedness.
Research on semantic relatedness in multiple languages is increasing.
2. SemEval-2024 Task 1
Tracks A, B, and C focus on supervised, unsupervised, and cross-lingual approaches.
Evaluation involves Spearman Correlation between predicted scores and human annotations.
3. Experiments
Supervised Track:
Utilized Term Frequency - Inverse Document Frequency (TF-IDF) and language-specific BERT models.
Achieved rankings ranging from 11th to 21st in Track A.
Unsupervised Track:
Used TF-IDF and language-specific BERT models.
Rankings ranged from 1st to 8th in Track B.
Cross-Lingual Track:
Employed TF-IDF and unrelated language-specific BERT models.
Rankings varied from 5th to 12th in Track C.
4. Results
Ensembling predictions improved Spearman Correlation Coefficient across all tracks.
Best performance seen with specific model combinations for each language.
Error Analysis
Challenges faced due to dataset sizes, model limitations, and linguistic diversity.
İstatistikler
この論文では、ランキングや相関係数などの重要な数値データは提供されていません。
Alıntılar
"Utilized ensemble of statistical machine learning approaches combined with language-specific BERT based models."
"Our best performing approaches achieved rankings ranging from 11th to 21st in Track A."