Contrastive Learning Using Graph Embeddings for Domain Adaptation of Language Models in the Process Industry
This work addresses domain adaptation for language models in the process industry, where text logs are critical, but it is incremental as it applies an existing method to a new domain.
This paper tackled the problem of adapting language models to the process industry domain by using graph embeddings for contrastive learning, resulting in a 9.8-14.3% performance improvement over a state-of-the-art model on a proprietary benchmark while reducing parameters by three times.
Recent trends in NLP utilize knowledge graphs (KGs) to enhance pretrained language models by incorporating additional knowledge from the graph structures to learn domain-specific terminology or relationships between documents that might otherwise be overlooked. This paper explores how SciNCL, a graph-aware neighborhood contrastive learning methodology originally designed for scientific publications, can be applied to the process industry domain, where text logs contain crucial information about daily operations and are often structured as sparse KGs. Our experiments demonstrate that language models fine-tuned with triplets derived from graph embeddings (GE) outperform a state-of-the-art mE5-large text encoder by 9.8-14.3% (5.45-7.96p) on the proprietary process industry text embedding benchmark (PITEB) while having 3 times fewer parameters.