Modeling Complex Relationship Paths for Knowledge Graph Completion

Abstract

Determining the validity of knowledge triples and filling in the missing entities or relationships in the knowledge graph are the crucial tasks for large-scale knowledge graph completion. So far, the main solutions use machine learning methods to learn the low-dimensional distributed representations of entities and relationships to complete the knowledge graph. Among them, translation models obtain excellent performance. However, the proposed translation models do not adequately consider the indirect relationships among entities, affecting the precision of the representation. Based on the long short-term memory neural network and existing translation models, we propose a multiple-module hybrid neural network model called TransP. By modeling the entity paths and their relationship paths, TransP can effectively excavate the indirect relationships among the entities, and thus, improve the quality of knowledge graph completion tasks. Experimental results show that TransP outperforms state-of-the-art models in the entity prediction task, and achieves the comparable performance with previous models in the relationship prediction task.

Knowledge Graph Completion: A Review

“…ii)Knowledge graph completion based on sequence learning Recurrent Neural Network (RNN) and its improved model Long Short Term Memory (LSTM) are the main models used in knowledge graph completion based on sequence FIGURE 13: Two-layer DSKG model. [106] learning. In the traditional neural network model, data is transmitted from the input layer to the hidden layer and then to the output layer.…”

“…RNN has performed well in many natural language processing (NLP) tasks, such as language modeling and machine translation [65], [66]. Literature [106] based on RNN model puts forward Deep Sequential Model for Knowledge Graph Completion (DSKG) for the knowledge graph completion, and applies a specific sampling method for model training, solving the problems that RNN will meet in the application of knowledge graph data: i) Triple group is not a natural language. Short sequences converted from triples may not provide enough contextual information for prediction.…”

“…Short sequences converted from triples may not provide enough contextual information for prediction. And because the number of paths in largescale knowledge graph is huge, the cost and difficulty to build valuable long sequence is great [106]; ii) in triples, relationships and entities are two different types of elements that appear in a fixed order. Treating them as elements of the same type in a sequence will lose some characteristics.…”

“…The summary of the sequence-based knowledge graph completion model is shown in Table 5. [104] Used to process data with sequence characteristics LSTM [109] An improved method for the problem of gradient disappearance encountered by the RNN model in processing long sequence data Attention Mechanism [111] The attention mechanism allows the processing of variable-sized inputs and makes decisions based on the most relevant part of the input.When the mechanism is used to calculate the presentation of a single sequence, it is usually called the self-attention mechanism Improved Models DSKG [106] DSKG uses different RNN units for the entity layer and the relationship layer TransP [109] Consider the indirect relationship between entities and improve the accuracy of knowledge representation learning GAT [112] Through the self-attention mechanism,the neighbor nodes are involved in the calculation of the implicit expression of each central node…”