Functional code clone detection with syntax and semantics fusion learning

Fang, Chunrong; Liu, Zixi; Shi, Yangyang; Huang, Jeff; Shi, Qiyun

doi:10.1145/3395363.3397362

Cited by 98 publications

(47 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…2) Code Clone Detection: There are supervised and unsupervised approaches to detect clones. While deep learning methods are applied to detect code clones, they require labelled data to train a supervised learning model [14,44,55]. As such, one needs human annotators to mark the pairs of snippets as clones, limiting the ability to detect clones by large amount of the data one can collect.…”

Section: U S E Ca S E Smentioning

confidence: 99%

“…Note that we do not compare with techniques such as Oreo [55], CCD [14], ASTNN [44] because they use supervised learning techniques to build clone classifiers. We believe that the code embeddings or the weights from the pretrained InferCode can be used for training supervised clone classifiers too, and with further improvement on self-supervised learning techniques such as improving the encoder, the auto-identified labels, and the loss function, the performance of unsupervised code clone detection may also get close to supervised ones.…”

Section: B Code Clone Detection 1) Datasets Metrics and Baselinesmentioning

confidence: 99%

“…Such techniques may miss useful information across different kinds of code units, and the trained representations may not perform well for various downstream tasks either. Some other techniques based on graph embeddings [14,15] share a similar drawback and in addition need the overheads of graph construction which may introduce inaccurate information in the graphs. Such limitations have been illustrated in a recent study: Kang et al [16] show that the pre-trained code2vec [8] representation does not perform well for other tasks when it was trained specifically for the method-name prediction task.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

InferCode: Self-Supervised Learning of Code Representations by Predicting Subtrees

Bui

Jiang

2021

2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE)

View full text Add to dashboard Cite

Learning code representations has found many uses in software engineering, such as code classification, code search, comment generation, and bug prediction, etc. Although representations of code in tokens, syntax trees, dependency graphs, paths in trees, or the combinations of their variants have been proposed, existing learning techniques have a major limitation that these models are often trained on datasets labeled for specific downstream tasks, and as such the code representations may not be suitable for other tasks. Even though some techniques generate representations from unlabeled code, they are far from being satisfactory when applied to the downstream tasks. To overcome the limitation, this paper proposes InferCode, which adapts the self-supervised learning idea from natural language processing to the abstract syntax trees (ASTs) of code. The novelty lies in the training of code representations by predicting subtrees automatically identified from the contexts of ASTs. With InferCode, subtrees in ASTs are treated as the labels for training the code representations without any human labelling effort or the overhead of expensive graph construction, and the trained representations are no longer tied to any specific downstream tasks or code units.We have trained an instance of InferCode model using Tree-Based Convolutional Neural Network (TBCNN) as the encoder of a large set of Java code. This pre-trained model can then be applied to downstream unsupervised tasks such as code clustering, code clone detection, cross-language code search, or be reused under a transfer learning scheme to continue training the model weights for supervised tasks such as code classification and method name prediction. Compared to prior techniques applied to the same downstream tasks, such as code2vec, code2seq, ASTNN, using our pre-trained InferCode model higher performance is achieved with a significant margin for most of the tasks, including those involving different programming languages. The implementation of InferCode and the trained embeddings are available at the link: https://github.com/bdqnghi/infercode.

show abstract

Section: U S E Ca S E Smentioning

confidence: 99%

Section: B Code Clone Detection 1) Datasets Metrics and Baselinesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

InferCode: Self-Supervised Learning of Code Representations by Predicting Subtrees

Bui

Jiang

2021

2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE)

View full text Add to dashboard Cite

show abstract

“…An abstract syntax tree (AST) is a tree representation of the abstract syntactic structure of source code written in a programming language [11]. The abstract syntax tree clearly describes the structure of the source code.…”

Section: B Abstract Syntax Treementioning

confidence: 99%

“…In many existing studies, source code is parsed into abstract syntax trees to produce code representations that capture the semantic relationships between different code elements [12], [13]. Code representation based on abstract syntax trees is now being used for code clone detection [11], defect prediction [14], auto program repair [15], and other problems. In metric-based code smell detection methods, abstract syntax trees may also be used to compute a set of source code metrics [7], [16].…”

Section: B Abstract Syntax Treementioning

confidence: 99%

Multi-Granularity Code Smell Detection using Deep Learning Method based on Abstract Syntax Tree

Xu¹

2021

International Conferences on Software Engineering and Knowledge Engineering

View full text Add to dashboard Cite

Code smell refers to poor design that is perceived to have a negative impact on readability and maintainability during software evolution, and it implies the possibility of refactoring. Therefore, the effective detection of code smell is of great importance. Many approaches including metric-based, heuristic-based, and machine learning approaches have been proposed to detect code smells. However, all these methods use manually selected features, which is highly subjective and difficult to select the most appropriate features. Recently, deep learning methods without extensive feature engineering have been proposed. Nevertheless, these token-based approaches may not achieve good results because they ignore many semantic and structural information of source code. To this end, we propose a novel deep learning approach based on abstract syntax trees(ASTs) to detect multi-granularity code smells, which captures the semantic and structural features of code fragments from the ASTs. The experimental results on four types of smells show that this approach achieves better results than the stateof-the-art approaches for detecting code smells with different granularities.

show abstract

CroLSSim: Cross‐language software similarity detector using hybrid approach of LSA‐based AST‐MDrep features and CNN‐LSTM model

Ullah

Naeem

et al. 2022

Int J of Intelligent Sys

View full text Add to dashboard Cite

Software similarity in different programming codes is a rapidly evolving field because of its numerous applications in software development, software cloning, software plagiarism, and software forensics. Currently, software researchers and developers search cross‐language open‐source repositories for similar applications for a variety of reasons, such as reusing programming code, analyzing different implementations, and looking for a better application. However, it is a challenging task because each programming language has a unique syntax and semantic structure. In this paper, a novel tool called Cross‐Language Software Similarity (CroLSSim) is designed to detect similar software applications written in different programming codes. First, the Abstract Syntax Tree (AST) features are collected from different programming codes. These are high‐quality features that can show the abstract view of each program. Then, Methods Description (MDrep) in combination with AST is used to examine the relationship among different method calls. Second, the Term Frequency Inverse Document Frequency approach is used to retrieve the local and global weights from AST‐MDrep features. Third, the Latent Semantic Analysis‐based features extraction and selection method is proposed to extract the semantic anchors in reduced dimensional space. Fourth, the Convolution Neural Network (CNN)‐based features extraction method is proposed to mine the deep features. Finally, a hybrid deep learning model of CNN‐Long‐Short‐Term Memory is designed to detect semantically similar software applications from these latent variables. The data set contains approximately 9.5K Java, 8.8K C#, and 7.4K C++ software applications obtained from GitHub. The proposed approach outperforms as compared with the state‐of‐the‐art methods.

show abstract

Functional code clone detection with syntax and semantics fusion learning

Cited by 98 publications

References 33 publications

InferCode: Self-Supervised Learning of Code Representations by Predicting Subtrees

InferCode: Self-Supervised Learning of Code Representations by Predicting Subtrees

Multi-Granularity Code Smell Detection using Deep Learning Method based on Abstract Syntax Tree

CroLSSim: Cross‐language software similarity detector using hybrid approach of LSA‐based AST‐MDrep features and CNN‐LSTM model

Contact Info

Product

Resources

About