Convolution embedding representation reasoning method based on fragmented knowledge

Abstract

The invention relates to a convolution embedding representation reasoning method based on fragmented knowledge. The method comprises the following steps: obtaining a fact triple, judging whether entities and relationships in the fact triple are trained or not, removing the head entity or the tail entity to destroy the complete fact triple and form a missing fact triple, putting the missing fact triple into a CE-RCF model to calculate to obtain an evaluation result, and if the evaluation result is greater than a set threshold, marking the fact triple as a trained fact triple, judging whether the number of the untrained fact triples is greater than a threshold value or not, if yes, putting all the fact triples into the CE-RCF model for parameter training, and otherwise, marking the current fact triple as an untrained fact triple, taking out and combining the untrained fact triple and the trained fact triple together, inputting the untrained fact triple and the trained fact triple into the CE-RCF model for training or retraining, marking all the merged fact triples as trained fact triples, and storing the trained fact triples to obtain a perfected fact triple.

Description

technical field [0001] The present invention relates to the field of massive data storage and reasoning under knowledge graphs, and in particular to a method for convolution embedded representation reasoning based on fragmented knowledge. Background technique [0002] Currently, convolutional embeddings represent inference algorithms for existing relational graph convolutional networks R-GCN and two-dimensional convolutional knowledge graph embeddings ConvE. The former uses a convolution operator to capture local information in the graph, which can adopt the same aggregation scheme when computing the convolution of each node. The R-GCN model mainly extends GCNs (Graph Neural Networks) from the local graph domain to handle large-scale knowledge graphs. The graph neural network can be understood as a special case of a simple and differentiable message passing framework. The ConvE model first connects the head entity embedding vector and the relationship embedding vector and f...

AI Technical Summary

Problems solved by technology

[0003] The R-GCN model can only deal with undirected graphs, and cannot solve the problem of directed knowledge graphs in real life; the ConvE model works well, but it needs to connect the head entity embedding vector and the relationship embedding vector before entering the convolution, so the model The effect of is related to the two-dimensional shapes of the two. Different connection methods of shapes may lead to different effects, and the interaction between the head entity and the relationship only occurs at the connection, wasting a lot of dimensional information between the two

In addition, the ConvE model only considers the information of the fact triple itself, ignoring the influence of other entities in the knowledge base on the fact triple

Due to the arrival of knowledge fragments, the knowledge base will change dynamically, and the existing embedded representation reasoning methods cannot solve the problem of dynamic fragmented reasoning.

Images