Abstract:Multi-hop Knowledge Base Question Answering (KBQA) aims to predict answers that require multi-hop reasoning from the topic entity in the question over the Knowledge Base (KB). Relation extraction is a core step in KBQA, which extracts the relation path from the topic entity to the answer entity. Compared with single-hop questions, multi-hop ones have more complex syntactic structures to understand, and multi-hop relation paths lead to a larger search space, which makes it much more challenging to extract the correct relation paths. To tackle the above challenges, most existing relation extraction approaches focus on the semantic similarity between questions and relation paths. However, those approaches only consider the word semantics of the relation names but ignore the graph semantics inside the knowledge base. As a result, their generalization ability relying on the naming rules of the relations, making it more difficult to generalize over large knowledge bases. To address the current limitations and take advantage of the graph semantics of relations, we propose a novel translational embedding-based relation extractor that utilizes pretrained embeddings from TransE. In particular, we treat the multi-hop relation path as a translation from the first relation to the last one in the semantic space of TransE. Then we map the question into this space under the supervision of the path embeddings. To take full advantage of the pretrained graph semantics in TransE, we propose a KBQA framework that leverages pretrained relation semantics in relation extraction and pretrained entity semantics in answer selection. Our approach achieves state-of-the-art performance on two benchmark datasets, WebQuestionSP and MetaQA, demonstrating its effectiveness on the multi-hop KBQA task.

Faithful Embeddings for Knowledge Base Queries

Neural-Symbolic Entangled Framework for Complex Query Answering

Neural-based Mixture Probabilistic Query Embedding for Answering FOL Queries on Knowledge Graphs.

Prompt-fused framework for Inductive Logical Query Answering

Question Answering over Knowledge Base using Language Model Embeddings

Path-Enhanced Multi-Relational Question Answering with Knowledge Graph Embeddings

Leveraging Knowledge Graph Embeddings For Natural Language Question Answering

An Inference Approach To Question Answering Over Knowledge Graphs

A Two-Stage Approach towards Generalization in Knowledge Base Question Answering

Combining Inductive and Deductive Reasoning for Query Answering over Incomplete Knowledge Graphs

Complex Query Answering on Eventuality Knowledge Graph with Implicit Logical Constraints

Sequential Query Encoding For Complex Query Answering on Knowledge Graphs

Query2Triple: Unified Query Encoding for Answering Diverse Complex Queries over Knowledge Graphs

Few-shot In-context Learning for Knowledge Base Question Answering

Knowledge base graph embedding module design for Visual question answering model

SPARKLE: Enhancing SPARQL Generation with Direct KG Integration in Decoding

A Knowledge Graph Embedding Model for Answering Factoid Entity Questions

Leveraging Knowledge Graph for Open-Domain Question Answering

Translational Relation Embeddings for Multi-Hop Knowledge Base Question Answering

Knowledge Base Question Answering by Case-based Reasoning over Subgraphs

EffiQA: Efficient Question-Answering with Strategic Multi-Model Collaboration on Knowledge Graphs