SKILL: Structured Knowledge Infusion for Large Language Models
This work addresses the challenge of efficiently integrating structured knowledge into LLMs for improved performance on knowledge-intensive tasks, particularly useful for industry-scale applications where alignment between graphs and text is not required.
The authors tackled the problem of whether large language models can better internalize knowledge from structured data like knowledge graphs compared to text, by proposing a method to infuse structured knowledge into LLMs through direct training on factual triples. They showed that models pre-trained on Wikidata KG outperformed T5 baselines on tasks like FreebaseQA and WikiHop, with a 3x improvement in exact match score on MetaQA when trained on a smaller KG.
Large language models (LLMs) have demonstrated human-level performance on a vast spectrum of natural language tasks. However, it is largely unexplored whether they can better internalize knowledge from a structured data, such as a knowledge graph, or from text. In this work, we propose a method to infuse structured knowledge into LLMs, by directly training T5 models on factual triples of knowledge graphs (KGs). We show that models pre-trained on Wikidata KG with our method outperform the T5 baselines on FreebaseQA and WikiHop, as well as the Wikidata-answerable subset of TriviaQA and NaturalQuestions. The models pre-trained on factual triples compare competitively with the ones on natural language sentences that contain the same knowledge. Trained on a smaller size KG, WikiMovies, we saw 3x improvement of exact match score on MetaQA task compared to T5 baseline. The proposed method has an advantage that no alignment between the knowledge graph and text corpus is required in curating training data. This makes our method particularly useful when working with industry-scale knowledge graphs.