A positional structure-oriented multimodal code summarization generation approach
For the task of automatic code summarization in software maintenance,an innovative model was proposed to address the limitations of existing methods in preserving semantic and structural information from source code.This model leveraged graph neural networks and Transformer technology to comprehensively capture both semantic and structural aspects of code.Additionally,byte pair encoding algorithm was employed to handle out-of-vocabulary words,and abstract syntax tree structure information was preserved using quadruples.This combination enabled the model to not only comprehensively capture the semantic features of source code but also accurately learn its syntactic structure.Experimental results on a real-world Java dataset demonstrate that this model outperforms baseline models in terms of BLEU,METEOR and ROUGE metrics,validating its effectiveness in generating more accurate code summarization.
万方数据
维普
