Back to results
US 10,761,839 B1Granted
Natural language search engine with a predictive writing tool for coding
AssigneeGlobant España, S.A.
InventorsMartin Migoya, Haldo Sponton
Life of this patent
Estimating the term…
Abstract
A system including a source code classifier, a natural language classifier, and a search engine operative on a search engine index associating outputs of the source code classifier and the natural language classifier to match a natural language search query with at least one computer code section from a computer program.
Claims18
A system comprising:
- a commented source code parser to extract source code comments from commented source code;
- a source code classifier to receive source code corresponding to the commented source code;
- a natural language classifier to receive the source code comments from the commented source code parser;
- a search engine operative on a search engine index associating outputs of the source code classifier and the natural language classifier to match a natural language search query with at least one code section in the source code;
- a relational model disposed between the source code classifier and the natural language classifier;
- and
- logic to apply a hyper-parameter adjustment algorithm on outputs of the search engine to adjust one or more hyper-parameters of one or both of the source code classifier and the natural language classifier, wherein a hyper-parameter of a classifier comprises a parameter that determines one of a number of layers in the classifier and a learning rate of the classifier.
4 more elements
A method comprising:
- generating a first embedding for source code;
- generating a second embedding for natural language word sequences derived from source code comments of the source code;
- relating the first embedding and the second embedding to generate a search engine index;
- operating a search engine on the search engine index to match a natural language search query with at least one code section of the source code;
- using a relational model for relating the first embedding and the second embedding to generate the search engine index;
- and
- applying a hyper-parameter adjustment algorithm on outputs of the search engine to adjust at least one hyper-parameter of both of a source code classifier utilizing the first embedding and a natural language classifier utilizing the second embedding, wherein a hyper-parameter of a classifier comprises a parameter that determines one of a number of layers in the classifier and a learning rate of the classifier.
4 more elements
Dependencies are inferred from the claim text.