Large Language Models Meet Knowledge Graphs to Answer Factoid Questions
Mikhail Salnikov1,2 *{}^{*}start_FLOATSUPERSCRIPT * end_FLOATSUPERSCRIPT, Hai Le1 *{}^{*}start_FLOATSUPERSCRIPT * end_FLOATSUPERSCRIPT, Prateek Rajput1, Irina Nikishina5, Pavel Braslavski3, Valentin Malykh4, and Alexander Panchenko1,2 1Skolkovo Institute of Science and Technology, 2Artificial Intelligence Research Institute, 3Nazarbayev University, 4ISP RAS Research Center for Trusted AI, 5Universität Hamburg {mikhail.salnikov, hai.le,
Abstract
Recently, it has been shown that the incorporation of structured knowledge into Large Language Models significantly improves the results for a variety of NLP tasks. In this paper, we propose a method for exploring pre-trained Text-to-Text Language Models enriched with additional information from Knowledge Graphs for answering factoid questions. More specifically, we propose an algorithm for subgraphs extraction from a Knowledge Graph based on question entities and answer candidates. Then, we procure easily interpreted information with Transformer-based models through the linearization of the extracted subgraphs. Final re-ranking of the answer candidates with the extracted information boosts Hits@1 scores of the pre-trained text-to-text language models by $4-6\%$ .
中文速览
针对知识图谱问答（KGQA）这一难题，研究者提出了一套结合大语言模型（LLM）与知识图谱（Knowledge Graph）子图的答案重排序方法：先用预训练的文本生成模型（T5）通过多样性束搜索生成多个候选答案，再从 Wikidata 中抽取连接问题实体与每个候选答案的子图，然后将子图线性化为自然语言序列，交给 BERT 类模型（MPNet、DistilBERT）或图Transformer（Graphormer）打分并重新排序。在复杂问答数据集 Mintaka 上的实验表明，该方法将原始语言模型的 Hits@1 准确率提升了 4–6 个百分点。这项工作的意义在于，它提供了一种可解释、易扩展的方式，让语言模型借助结构化知识"按图索骥"地验证候选答案，有效弥补了大模型在复杂事实性问答上的短板。
原文 arXiv:2310.02166;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2310.02166v1