Knowledge graph representation learning model method fusing entity description and path information

The invention discloses a knowledge graph representation learning model method fusing entity description and path information, which comprises the following steps of: firstly, extracting two subsets from a large knowledge graph Freebase as training sets, inputting the two subsets into a finely-tuned and strongly-optimized pre-training language representation model RoBERT by combining entity description made by a Ruobbing Xie artificial training set, and constructing a knowledge graph representation learning model; training and learning are carried out through a self-attention mechanism and a feedforward neural network comprising four full-connection layers and an activation function Relu, and an entity described by a fusion entity is output in the last layer, and a relation represents a learning vector; then converting the knowledge graph into a vector fused with ordered relation path information, and outputting energy values EP of all training set triples; then, vector representation of the triple is optimiz