Neural bit embeddings for graphs
View Patent ↗An approach is provided in which a system transforms a set of embedding approximation values corresponding to a set of knowledge graph nodes into a set of binary valued embedding vectors. The system evaluates the set of binary valued embedding vectors against a query and a selects one of the binary valued embedding vectors based on the evaluation. The system then identifies one of the knowledge graph nodes that corresponds to the selected binary valued embedding vector and in turn, provides a result to the query based on the identified knowledge graph node.
1. An information handling system comprising:
one or more processors;
a memory coupled to at least one of the processors;
a set of computer program instructions stored in the memory and executed by at least one of the processors in order to perform actions of:
transforming a plurality of embedding approximation values into a plurality of binary valued embedding vectors, wherein the plurality of embedding approximation values correspond to a plurality of nodes in a knowledge graph;
determining that a first one of the plurality of nodes matches a query, wherein the first node corresponds to a first one of the plurality of binary valued embedding vectors;
creating a plurality of second binary valued embedding vectors from the first binary valued embedding vector, wherein each one of the plurality of second binary valued embedding vectors comprises a single bit that is different from the first binary valued embedding vector, and wherein each one of the plurality of second binary valued embedding vectors is different from each other;
identifying a second one of the plurality of nodes that corresponds to one of the plurality of second binary valued embedding vectors; and
providing a result to the query based on the second node.
2. The information handling system of claim 1 wherein the one or more processors perform additional actions comprising:
initializing a plurality of embedding elements having a number of dimensions, wherein the number of dimensions is based on an amount of feature sets corresponding to each of the plurality of nodes;
generating a matrix based on an amount of the plurality of nodes and the number of dimensions; and
optimizing the matrix to minimize an objective function, wherein the optimization produces the plurality of embedding approximation values.
3. The information handling system of claim 1 wherein the one or more processors perform additional actions comprising:
assigning a binary value of 0 to each of the plurality of embedding approximation values that is less than 0.5; and
assigning a binary value of 1 to each of the plurality of embedding approximation values that is greater than or equal to 0.5.
4. The information handling system of claim 1 wherein the plurality of nodes are sparsely distributed in the knowledge graph.
5. The information handling system of claim 1 wherein the one or more processors perform additional actions comprising:
mapping an element in the query into the knowledge graph; and
identifying a closest neighbor to the mapped element, wherein the closest neighbor is the first node.
6. A computer program product stored in a computer readable storage medium, comprising computer program code that, when executed by an information handling system, causes the information handling system to perform actions comprising:
transforming a plurality of embedding approximation values into a plurality of binary valued embedding vectors, wherein the plurality of embedding approximation values correspond to a plurality of nodes in a knowledge graph;
determining that a first one of the plurality of nodes matches a query, wherein the first node corresponds to a first one of the plurality of binary valued embedding vectors;
creating a plurality of second binary valued embedding vectors from the first binary valued embedding vector, wherein each one of the plurality of second binary valued embedding vectors comprises a single bit that is different from the first binary valued embedding vector, and wherein each one of the plurality of second binary valued embedding vectors is different from each other;
identifying a second one of the plurality of nodes that corresponds to one of the plurality of second binary valued embedding vectors; and
providing a result to the query based on the second node.
7. The computer program product of claim 6 wherein the information handling system performs further actions comprising:
initializing a plurality of embedding elements having a number of dimensions, wherein the number of dimensions is based on an amount of feature sets corresponding to each of the plurality of nodes;
generating a matrix based on an amount of the plurality of nodes and the number of dimensions; and
optimizing the matrix to minimize an objective function, wherein the optimization produces the plurality of embedding approximation values.
8. The computer program product of claim 6 wherein the information handling system performs further actions comprising:
assigning a binary value of 0 to each of the plurality of embedding approximation values that is less than 0.5; and
assigning a binary value of 1 to each of the plurality of embedding approximation values that is greater than or equal to 0.5.
9. The computer program product of claim 6 wherein the plurality of nodes are sparsely distributed in the knowledge graph.