Soft Reasoning on Uncertain Knowledge Graphs

The study of machine learning-based logical query-answering enables reasoningwith large-scale and incomplete knowledge graphs. This paper further advancesthis line of research by considering the uncertainty in the knowledge. Theuncertain nature of knowledge is widely observed in the real world, butdoes not align seamlessly with the first-order logic underpinningexisting studies. To bridge this gap, we study the setting of soft queries onuncertain knowledge, which is motivated by the establishment of soft constraintprogramming. We further propose an ML-based approach with both forwardinference and backward calibration to answer soft queries on large-scale,incomplete, and uncertain knowledge graphs. Theoretical discussions presentthat our methods share the same complexity as state-of-the-art inferencealgorithms for first-order queries. Empirical results justify the superiorperformance of our approach against previous ML-based methods with numberembedding extensions.