Method, device, and program product for managing knowledge graphs

An electronic device, comprising:

• at least one processor;
• and
• memory coupled to the at least one processor and having instructions stored therein, wherein the instructions, when executed by the at least one processor, cause the electronic device to perform actions comprising:
• acquiring, in a processor-based machine learning system, a historical sequence of knowledge graphs, wherein the historical sequence comprises a plurality of historical knowledge graphs respectively associated with a plurality of historical time points, and a historical knowledge graph in the plurality of historical knowledge graphs includes a plurality of nodes and at least one edge between the plurality of nodes;
• applying the historical knowledge graphs to an attention-based activation mechanism of the processor-based machine learning system, the attention-based activation mechanism being configured to determine a plurality of subgraphs respectively associated with the plurality of historical knowledge graphs based on nodes and edges included in the plurality of historical knowledge graphs, wherein attention measures are generated for respective ones of the edges by processing, for each of the edges, weighted edge information for that edge and associated weighted first and second node information for respective first and second nodes connected by that edge, utilizing a rectified linear unit of the attention-based activation mechanism, and further wherein the subgraphs are determined based on the attention measures;
• applying the subgraphs to a first neural network of the processor-based machine learning system, the first neural network being configured to determine node embeddings for respective nodes of the subgraphs;
• determining, in the processor-based machine learning system, statistical embeddings for the respective nodes of the subgraphs, wherein each of the statistical embeddings is generated by expanding a dimension of statistical information associated with a corresponding one of the nodes through application of a designated function to the statistical information, the designated function generating respective different positional entries of the statistical embedding from the statistical information;
• combining, in the processor-based machine learning system, the node embedding with the statistical embedding for the respective nodes of the subgraphs, to generate respective corresponding combined embeddings;
• applying the combined embeddings to a second neural network of the processor-based machine learning system, the second neural network comprising a plurality of graph convolutional networks coupled to respective ones of a plurality of gated recurrent units, with the graph convolutional networks and the gated recurrent units being collectively configured to determine a sequence feature associated with the historical sequence of the knowledge graphs based on the plurality of subgraphs, by processing at least a portion of the combined embeddings generated for the respective nodes of the subgraphs;
• and
• training a prediction model of the processor-based machine learning system based on the sequence feature such that the prediction model can describe a correlation among the plurality of historical knowledge graphs, wherein the prediction model comprises a multi-layer perceptron.

A computer program product that is tangibly stored on a non-transitory computer-readable medium and comprises machine-executable instructions, wherein the machine-executable instructions, when executed by a machine, cause the machine to perform actions comprising:

• acquiring, in a processor-based machine learning system, a historical sequence of knowledge graphs, wherein the historical sequence comprises a plurality of historical knowledge graphs respectively associated with a plurality of historical time points, and a historical knowledge graph in the plurality of historical knowledge graphs includes a plurality of nodes and at least one edge between the plurality of nodes;
• applying the historical knowledge graphs to an attention-based activation mechanism of the processor-based machine learning system, the attention-based activation mechanism being configured to determine a plurality of subgraphs respectively associated with the plurality of historical knowledge graphs based on nodes and edges included in the plurality of historical knowledge graphs, wherein attention measures are generated for respective ones of the edges by processing, for each of the edges, weighted edge information for that edge and associated weighted first and second node information for respective first and second nodes connected by that edge, utilizing a rectified linear unit of the attention-based activation mechanism, and further wherein the subgraphs are determined based on the attention measures;
• applying the subgraphs to a first neural network of the processor-based machine learning system, the first neural network being configured to determine node embeddings for respective nodes of the subgraphs;
• determining, in the processor-based machine learning system, statistical embeddings for the respective nodes of the subgraphs, wherein each of the statistical embeddings is generated by expanding a dimension of statistical information associated with a corresponding one of the nodes through application of a designated function to the statistical information, the designated function generating respective different positional entries of the statistical embedding from the statistical information;
• combining, in the processor-based machine learning system, the node embedding with the statistical embedding for the respective nodes of the subgraphs, to generate respective corresponding combined embeddings;
• applying the combined embeddings to a second neural network of the processor-based machine learning system, the second neural network comprising a plurality of graph convolutional networks coupled to respective ones of a plurality of gated recurrent units, with the graph convolutional networks and the gated recurrent units being collectively configured to determine a sequence feature associated with the historical sequence of the knowledge graphs based on the plurality of subgraphs, by processing at least a portion of the combined embeddings generated for the respective nodes of the subgraphs;
• and
• training a prediction model of the processor-based machine learning system based on the sequence feature such that the prediction model can describe a correlation among the plurality of historical knowledge graphs, wherein the prediction model comprises a multi-layer perceptron.