E-ReaRev: Adaptive Reasoning for Question Answering over Incomplete Knowledge Graphs by Edge and Meaning Extensions

Multi-hop knowledge graph question answering aims at answering natural language questions by reasoning over knowledge graphs. However, insufficiency of training samples and incompleteness of knowledge graphs can lead to serious decline in model performance. Previous methods either consume excessive storage memory in introducing additional knowledge resources or require a large amount of computation power when introducing additional networks to predict missing links. In this paper, we introduce a new few-shot setting into several datasets that considers insufficient training samples as well as incomplete knowledge graph edges but complete knowledge graph nodes. Moreover, we propose a more efficient model for question answering over incomplete knowledge graphs. Concretely, we propose a novel edge-extension module to predict missing edges by similarity calculation which consumes low computational resources. Furthermore, a new meaning-extension module is introduced to incorporate semantic information of entities into node representations with lower memory consumption. Experimental results on two datasets demonstrate the effectiveness of proposed method compared with other baselines under the few-shot setting.