Decision-based code annotation generation method fusing information retrieval and deep learning

Abstract

The invention provides a decision-based code annotation generation method fusing information retrieval and deep learning, which comprises the following steps: (1) based on a GitHub website, mining high-quality open source items, collecting functions and annotations in codes to form a corpus, and dividing the corpus into a training set and a verification set; (2) constructing a model based on information retrieval, searching a code segment csim most similar to the target code segment c in the training set by fusing semantic, lexical and grammatical similarities, and reusing annotations of the code segment csim; (3) constructing a model based on deep learning, considering an adversarial sample, and generating a code annotation by using a Transform-based model; (4) analyzing the relationship between the performance of the two models and the similarity score between the two code segments c and csim on a verification set, and determining a similarity threshold value; and (5) when an annotation is generated for a new target code, outputting a corresponding code annotation according to the decision module. The method has the beneficial effect that high-quality code annotations can be generated.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a code annotation generation method based on decision-making fusion information retrieval and deep learning. Background technique [0002] Code comment generation aims to help software developers understand the design intent and functionality of code snippets. Due to limited project development budget or insufficient attention to the writing of code comments, the quality of code comments is difficult to guarantee. Furthermore, code comments are often not automatically updated as code snippets evolve. Therefore, there is an urgent need to design effective methods to automatically generate high-quality code annotations after analyzing the semantics of target code segments. [0003] Previous studies usually use information retrieval for code comment generation tasks and achieve good performance. Since code cloning is prevalent in software development, information retrie...

AI Technical Summary

Problems solved by technology

Experimental research shows that these methods can achieve better performance and have higher generalization ability, but the generated annotations tend to contain high-frequency words, and it is difficult to cover low-frequency words

Images