NeuInfer: Knowledge Inference on N-ary Facts

Abstract

Knowledge inference on knowledge graph has attracted extensive attention, which aims to find out connotative valid facts in knowledge graph and is very helpful for improving the performance of many downstream applications. However, researchers have mainly poured attention to knowledge inference on binary facts. The studies on n-ary facts are relatively scarcer, although they are also ubiquitous in the real world. Therefore, this paper addresses knowledge inference on n-ary facts. We represent each n-ary fact as a primary triple coupled with a set of its auxiliary descriptive attribute-value pair(s). We further propose a neural network model, NeuInfer, for knowledge inference on n-ary facts. Besides handling the common task to infer an unknown element in a whole fact, NeuInfer can cope with a new type of task, flexible knowledge inference. It aims to infer an unknown element in a partial fact consisting of the primary triple coupled with any number of its auxiliary description(s). Experimental results demonstrate the remarkable superiority of NeuInfer.

Link Prediction on N-ary Relational Data Based on Relatedness Evaluation

“…However, GETD only focuses on handling n-ary relational facts of the same arities. HINGE [9] and NeuInfer [10] consider principal and subordinate structure information and represent each n-ary relational fact as a primary triple coupled with a set of its auxiliary descriptive r:v pair(s). Actually, no all n-ary relational facts have a primary triple.…”

“…The representative or state-of-the-art methods for link prediction on n-ary relational data directly are RAE [6], HypE [7], GETD [8], HINGE [9], and NeuInfer [10]. As the tensor based method GETD inherently requires all facts to have the same arities, it is not applicable to JF17K and WikiPeople with mixed arities.…”

“…where Rel + is a valid n-ary relational fact, and Relis one of its corrupted facts; R Rel + and R Rel -are the overall relatedness feature vectors of Rel + and Rel -, respectively; sgn(x) is the function that returns 1, if x > 0, otherwise, returns 0; sum(•) is the element-wise sum function. In Equation (10), the left part of the minus sign counts the number of dimensions that R Rel + is larger than R Rel -, and the right Similar to the evaluation procedure, we replace M ichael Douglas in Fact-1 and N obel P rize in P hysics in Fact-2 with all the other role-values in V , then these corrupted facts, as well as Fact-1 and Fact-2 are fed into the proposed methods to obtain the overall relatedness feature vectors. Subsequently, the distinguishability metric is computed via Equation (10).…”

“…In Equation (10), the left part of the minus sign counts the number of dimensions that R Rel + is larger than R Rel -, and the right Similar to the evaluation procedure, we replace M ichael Douglas in Fact-1 and N obel P rize in P hysics in Fact-2 with all the other role-values in V , then these corrupted facts, as well as Fact-1 and Fact-2 are fed into the proposed methods to obtain the overall relatedness feature vectors. Subsequently, the distinguishability metric is computed via Equation (10). Fig.…”

“…The above example also elucidates its necessity. Some newly proposed methods [7], [8], [9], [10] further apply position-specific convolution [7], tensor decomposition techniques [8], or structure-aware methods [9], [10]. They do not consider any type constraint.…”