PROJECT TITLE :

A Semantic Network Encoder for Associated Fact Prediction

ABSTRACT:

A network of concepts that are connected to one another by semantic relations is called a semantic network. Binary semantic network as well as multiplex semantic network are both included in this structure. The associated fact prediction is a link prediction task that aims to infer the implicitly connected facts by mining the high-level representation of the network. This is accomplished through a process known as "association mining." Previous techniques for predicting associated facts placed a large amount of importance on the topological characteristics of the network but did not make use of the information contained in semantic expressions. In this paper, we propose a Semantic Network Encoder (SemNE) that can learn a feature mapping function from binary semantic networks and then apply that function in a pre-training manner to multiplex semantic networks. This function is learned from the binary semantic networks. An embedding encoder and a prediction decoder are both part of the SemNE framework, which is a two-stage architecture. In order to enrich the network representation, it models both the semantic information and the network topology simultaneously. In order to unify the topological feature representations and the semantic feature representations, a method of word self-organization that is based on the factual boundary has been proposed. Experimental results on binary semantic networks show that SemNE achieves the state-of-the-art results in associated fact prediction. Experimental results on multiplex semantic networks show that SemNE is scalable and can effectively improve the performance of existing models. Both sets of results demonstrate that SemNE achieves the state-of-the-art results in associated fact prediction.