Classifying code comments in Java software systems

Pascarella, Luca; Bruntink, Magiel; Bacchelli, Alberto

doi:10.1007/s10664-019-09694-w

Cited by 48 publications

(32 citation statements)

References 29 publications

(46 reference statements)

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…The most relevant approaches are References 7,8, which classify Java and C/C++ source code comments using a J48 decision tree algorithm and a NB multinominal classifier, respectively. Both use specific text preprocessing and feature engineering with rule‐based taxonomy classification, and achieve a weighted average precision and recall of 93% to 96% (on imbalanced datasets).…”

Section: Discussion Of Resultsmentioning

confidence: 99%

“…Pascarella et al employ supervised ML to build an automated classification approach 8 . Their study presents a publicly available dataset with manually classified comments on which they apply probabilistic classifiers, for example, Naive Bayes (NB), and decision tree algorithms, for example, J48 and RF, to detect and classify comments according to the hierarchical taxonomy of code comments introduced before.…”

Section: Related Workmentioning

confidence: 99%

“…8 Their study presents a publicly available dataset with manually classified comments on which they apply probabilistic classifiers, for example, Naive Bayes (NB), and decision tree algorithms, for example, J48 and RF, to detect and classify comments according to the hierarchical taxonomy of code comments introduced before. To the best of our knowledge, the work by Pascarella et al 8 is the only other work in this field which applies comment classification to industrial sample projects, finding that OSS use four times as many comments as closed source software. An interesting finding of their study is, that for unseen software projects precision and recall can be substantially improved by manually classifying between 40 to 80 code comments.…”

Section: Ml-based Approachesmentioning

confidence: 99%

“…To mitigate these problems, we are interested in whether we can replace heuristics by leveraging machine learning (ML) for comment classification for an industrial use case. ML methods were already applied for comment classification of C++, Java, and Python software systems with promising results in References 7‐9. Current practices for classification of source code comments in deep learning (DL) are rare.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Leveraging machine learning for software redocumentation—A comprehensive comparison of methods in practice

et al. 2020

View full text Add to dashboard Cite

Source code comments contain key information about the underlying software system. Many redocumentation approaches, however, cannot exploit this valuable source of information. This is mainly due to the fact that not all comments have the same goals and target audience and can therefore only be used selectively for redocumentation. Performing a required classification manually, for example, in the form of heuristics, is usually time‐consuming and error‐prone and strongly dependent on programming languages and guidelines of concrete software systems. By leveraging machine learning (ML), it should be possible to classify comments and thus transfer valuable information from the source code into documentation with less effort but the same quality. We applied classical ML techniques but also deep learning (DL) approaches to legacy systems by transferring source code comments into meaningful representations using, for example, word embeddings but also novel approaches using quick response codes or a special character‐to‐image encoding. The results were compared with industry‐strength heuristic classification. As a result, we found that ML outperforms the heuristics in number of errors and less effort, that is, we finally achieve an accuracy of more than 95% for an image‐based DL network and even over 96% for a traditional approach using a random forest classifier.

show abstract

Section: Discussion Of Resultsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Ml-based Approachesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Leveraging machine learning for software redocumentation—A comprehensive comparison of methods in practice

et al. 2020

View full text Add to dashboard Cite

show abstract

“…We use the approach proposed by Pascarella et al to classify the header and internal comments of public, private, and protected methods, respectively. As Tables 6 to 8 show, according to the work of Pascarella et al, each comment can be classified at coarse granularity, such as Purpose, Notice, Under developer, Style and IDE, Metadata, and Discarded. Meanwhile, each categorization can be divided into finer categories.…”

Section: Empirical Studymentioning

confidence: 99%

Does your code need comment?

et al. 2019

View full text Add to dashboard Cite

Code comments convey information about the programmers' intention in a more explicit but less rigorous manner than source code. This information can assist programmers in various tasks, such as code comprehension, reuse, and maintenance. To better understand the properties of the comments existing in the source code, we analyzed more than 450 000 comments across 136 popular open-source software systems coming different domains. We found that the methods involving header comments and internal comments were shown low percentages in software systems, ie, 4.4% and 10.27%, respectively. As an application of our findings, we propose an automatic approach to determine whether a method needs a header comment, known as commenting necessity identification. Specifically, we identify the important factors for determining the commenting necessity of a method and extract them as structural features, syntactic features, and textual features. Then, by applying machine learning techniques and noise-handling techniques, we achieve a precision of 88.5% on eight open-source software from GitHub. The encouraging experimental results demonstrate the feasibility and effectiveness of our approach.

show abstract