CMLM-CSE: Based on Conditional MLM Contrastive Learning for Sentence Embeddings
This work addresses sentence embedding quality for NLP applications, offering incremental improvements over existing methods.
The paper tackles the problem of sentence embeddings by addressing the neglect of individual word influence on semantics in traditional contrastive learning, proposing CMLM-CSE, an unsupervised framework based on conditional MLM that integrates sentence embedding for MLM tasks. The result shows improvements over SimCSE, with gains of 0.55 percentage points using Bertbase and 0.3 percentage points using Robertabase on textual similarity tasks.
Traditional comparative learning sentence embedding directly uses the encoder to extract sentence features, and then passes in the comparative loss function for learning. However, this method pays too much attention to the sentence body and ignores the influence of some words in the sentence on the sentence semantics. To this end, we propose CMLM-CSE, an unsupervised contrastive learning framework based on conditional MLM. On the basis of traditional contrastive learning, an additional auxiliary network is added to integrate sentence embedding to perform MLM tasks, forcing sentence embedding to learn more masked word information. Finally, when Bertbase was used as the pretraining language model, we exceeded SimCSE by 0.55 percentage points on average in textual similarity tasks, and when Robertabase was used as the pretraining language model, we exceeded SimCSE by 0.3 percentage points on average in textual similarity tasks.