G-Loss: Graph-Guided Fine-Tuning of Language Models

TL;DR

G-Loss introduces a graph-guided loss function for fine-tuning language models, leveraging global semantic relationships to improve embedding quality and classification accuracy.

Contribution

It proposes a novel semi-supervised, graph-based loss that captures global semantic structure, enhancing fine-tuning of pre-trained language models.

Findings

G-Loss converges faster than traditional loss functions.

It yields higher classification accuracy across multiple datasets.

G-Loss produces more semantically coherent embeddings.

Abstract

Traditional loss functions, including cross-entropy, contrastive, triplet, and su pervised contrastive losses, used for fine-tuning pre-trained language models such as BERT, operate only within local neighborhoods and fail to account for the global semantic structure. We present G-Loss, a graph-guided loss function that incorporates semi-supervised label propagation to use structural relationships within the embedding manifold. G-Loss builds a document-similarity graph that captures global semantic relationships, thereby guiding the model to learn more discriminative and robust embeddings. We evaluate G-Loss on five benchmark datasets covering key downstream classification tasks: MR (sentiment analysis), R8 and R52 (topic categorization), Ohsumed (medical document classification), and 20NG (news categorization). In the majority of experimental setups, G-Loss converges faster and produces semantically coherent embedding spaces, resulting in higher classification accuracy than models fine-tuned with traditional loss functions.