Dynamic heterogeneous network representation method based on meta-path

Abstract

The invention belongs to the field of graph network representation learning, and particularly relates to a dynamic heterogeneous network representation method based on a meta-path, which comprises the following steps: constructing a dynamic heterogeneous network, and sampling different time weighted meta-path sequences from the network according to a time weighted meta-path; preprocessing the vectors of the network nodes, and aggregating the information of the network node sequence of each element path after preprocessing through the GRU; coding the time of the node sequence by adopting a relative time coding technology; acquiring deep feature information of the information of the node sequence by adopting a Bi-GRU method, and aggregating time features and structural features; a Bi-GRU with an attention mechanism is used for interacting feature information of different sequences at different times, and final representation of nodes is obtained; the method can adapt to a node learning task in a dynamic heterogeneous network and dynamic evolution of the network, a downstream task can classify and cluster the nodes, and the learning and representation capability of the graph network nodes can be effectively improved.

Description

technical field [0001] The invention belongs to the field of graph network representation learning, in particular to a meta-path-based dynamic heterogeneous network representation method. Background technique [0002] Network data is usually unstructured data, and it is very difficult to directly use machine learning models to mine information in the network. Network representation learning is to map high-dimensional sparse graph data to low-dimensional space, while retaining the structural information in the network, so as to obtain low-dimensional dense structured vector representation. [0003] At present, the research on static network representation learning and dynamic homogeneous network representation learning is relatively mature, but the research on dynamic heterogeneous network that is closer to the actual network is in its infancy, and it is necessary to explore and study it. [0004] Network representation learning is to represent the nodes in the network as lo...

AI Technical Summary

Problems solved by technology

Therefore, only considering the static processing method does not conform to the evolution law of the actual network

Another type of dynamic homogeneous network representation learning method does not consider the difference between network nodes and links. If it is directly applied to a dynamic heterogeneous network, some semantic information will inevitably be lost.

Images