A path-based knowledge graph embedding method

Abstract

The invention discloses a path-based knowledge graph embedding method, which comprises the following steps of for a given single knowledge graph or a joint knowledge graph consisting of two knowledgegraphs, firstly sampling the given single knowledge graph by using a partial random walk method; next, modeling the obtained sampling path by using a new jumping cyclic network, so that the network can overcome the defect that the traditional cyclic neural network cannot identify the knowledge graph structure while effectively learning the dependency among elements in the path. For the output of the jump recurrent neural network, a type-based noise comparison estimation method is adopted to efficiently evaluate the loss of the network, and iteration is carried out in a feedback updating mode to carry out training. Compared with the prior art, the knowledge graph embedding trained by the method has the advantages of high expressibility, strong generalization ability and the like, and the accuracy and the efficiency of knowledge graph completion and entity alignment can be remarkably improved by applying the method.

Description

technical field [0001] The present invention relates to a knowledge graph (Knowledge Graph), in particular to a path-based knowledge graph embedding (Knowledge Graph Embedding) method. Background technique [0002] Knowledge graphs have become an important resource for various knowledge-driven applications, such as search engines, question answering, and recommendation systems. Knowledge graphs store a large number of real-world facts in a structured way. Among them, each fact is described by a triplet (h, r, t), where h, t, r represent the head entity, tail entity, and the relationship between them, respectively. Usually, the existing knowledge graph is difficult to support the knowledge required by various applications. Therefore, researchers in the field of knowledge graphs have proposed two main tasks to solve this problem: (1) Entity Alignment, which is dedicated to linking entities that refer to the same in two different knowledge graphs. (2) Knowledge Graph Complet...

AI Technical Summary

Problems solved by technology

However, there are two limitations in the knowledge graph embedding learning method based on triples: one is that the learned knowledge graph embedding lacks distinguishability

In the knowledge map, there are often cases where multiple entities have the same 1-hop neighbors. If only triples are used for learning, the embedding of these entities will be indistinguishable.

In addition, some entities are relatively poor in information (such as having few relations), which makes triple-based methods further suffer

Second, triplet-based methods lack efficiency in information exchange and dissemination

However, the hidden state contains all the previously input information (ie Tim Berners-Lee, country, United Kingdom), which makes the recurrent neural network unable to effectively use the head entity United Kingdom of the relationship language to predict its tail entity English, resulting in learning Reduced efficiency and performance

[0005] In addition, current methods for sampling paths are mostly based on undirected graphs or simple graphs, which are often only applicable to traditional social networks

These methods do not consider the direction and type of the relationship in the knowledge graph, so they are not suitable for the sampling of the knowledge graph

The traditional random walk (Random Walks) method uses an unbiased way to randomly sample entities in the knowledge graph, directly using high-quality paths that are often difficult to obtain.

Images