RPR-KGQA: Relational Path Reasoning for Multi-hop Question Answering with Knowledge Graph

Knowledge Graphs (KG) is a type of knowledge base organized in the framework of semantic network, which has directed graph structures. As a basic task performed with KGs, the aim of Knowledge Graph Question Answering (KGQA) is to retrieval from a KG the entities and relations related with the input question. Multi-hop KGQA is a branch of KGQA tasks, whose input question contains several entities and relations between the topic entity and the answer. Owing to the lack of the inference information produced in the intermediate process and the incompleteness of the KGs, it is very challenging to achieve high performance for the task of multi-hop KGQA in weakly supervised situations. To address this problem, we propose a novel method, namely RPR-KGQA, which utilize relational path reasoning for multi-hop question answering with knowledge graph. Unlike traditional methods which acquire answers from the subgraphs extracted from the neighbourhoods, we determine the answer by computing the semantic similarity between the explicit relational statements in the input questions and the implicit relational paths stored in the structural KGs. Experiments on the benchmark datasets have validated the effectiveness and generality of our method.