Just Rank: Rethinking Evaluation with Word and Sentence Similarities
This addresses the problem of outdated and misleading evaluation methods for embedding models in NLP, which is incremental as it refines existing evaluation practices rather than introducing a new paradigm.
The paper identifies that using semantic similarity as the gold standard for evaluating word and sentence embeddings leads to overfitting and hinders model development, and proposes EvalRank, a new intrinsic evaluation method that shows a stronger correlation with downstream tasks based on experiments with over 60 models and datasets.
Word and sentence embeddings are useful feature representations in natural language processing. However, intrinsic evaluation for embeddings lags far behind, and there has been no significant update since the past decade. Word and sentence similarity tasks have become the de facto evaluation method. It leads models to overfit to such evaluations, negatively impacting embedding models' development. This paper first points out the problems using semantic similarity as the gold standard for word and sentence embedding evaluations. Further, we propose a new intrinsic evaluation method called EvalRank, which shows a much stronger correlation with downstream tasks. Extensive experiments are conducted based on 60+ models and popular datasets to certify our judgments. Finally, the practical evaluation toolkit is released for future benchmarking purposes.