A Method for Learning Function-Hierarchical Embedding Representations in Source Code in Hyperbolic Spaces

Abstract

The invention discloses a method for learning the function level embedding representation in the source code in the hyperbolic space, the method has the following steps: S1, data collection and construction stage: S2, Ridge curvature analysis stage: S3, Poincaré model learning Function embedding representation stage: the present invention applies the latest data mining technology, dimension reduction technology, deep learning processing technology, and visualization technology to hyperbolic space function embedding learning, which can effectively accommodate massive data, just like source code data; efficient and Accurately achieve data dimensionality reduction, and then visualize and analyze hierarchical information.

Description

technical field [0001] The invention relates to the technical field of network research, in particular to a function-level embedding representation method for learning source codes in a hyperbolic space. Background technique [0002] The successful application of distributed representations in natural language processing and the naturalness assumption have inspired researchers to apply distributed representations to source code mining. Most existing work on learning distributed representations from source code typically treats programs as sequences of symbols or bags of words. Some of them first parse the program to obtain the corresponding Abstract Syntax Tree (AST), which is suitable for tree-based neural networks or expand the AST to a graph for further processing. According to these underlying neural network models, existing works can generally be classified into the following categories: feed-forward neural networks, recurrent neural networks, convolutional neural netw...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the deficiencies of existing research methods, the present invention provides a method for learning the hierarchical embedding representation of functions in source codes in hyperbolic space, which solves the problems of lack of hierarchical information and insufficient space in existing research methods

Images