Semantic Triplet Restoration: A Novel Protocol for Hierarchical Table Understanding in Large Language Models

Researchers introduced Semantic Triplet Restoration, a new way to represent tables as <item path, feature path, value> triplets.

The method is paired with TripletQL, which routes each question to the most relevant triplets before answering.

In tests on four Chinese and English table-QA benchmarks, it matched or beat HTML-based baselines.

It also used fewer input tokens, making table understanding more compact and potentially cheaper for LLMs.