Using Context-to-Vector with Graph Retrofitting to Improve Word Embeddings
This work addresses the need for better static embeddings in low-resource and lightweight settings, though it appears incremental as it builds on existing Skip-gram and retrofitting methods.
The paper tackles the problem of improving traditional static word embeddings by incorporating contextual information from pre-trained models and using graph-based post-processing, achieving significant performance gains over baselines in evaluation tasks.
Although contextualized embeddings generated from large-scale pre-trained models perform well in many tasks, traditional static embeddings (e.g., Skip-gram, Word2Vec) still play an important role in low-resource and lightweight settings due to their low computational cost, ease of deployment, and stability. In this paper, we aim to improve word embeddings by 1) incorporating more contextual information from existing pre-trained models into the Skip-gram framework, which we call Context-to-Vec; 2) proposing a post-processing retrofitting method for static embeddings independent of training by employing priori synonym knowledge and weighted vector distribution. Through extrinsic and intrinsic tasks, our methods are well proven to outperform the baselines by a large margin.