arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2401.12959cs.SE
收藏

理解有用的代码审查评论中的表情符号 :)

Understanding Emojis :) in Useful Code Review Comments

Sharif Ahmed, Nasir U. Eisty

AI总结:

研究通过分析代码审查评论中表情符号的情感与语义,结合表情符号特征与预训练嵌入构建模型,显著提升了评论有用性评估的准确率。

摘要:

表情符号和颜文字作为非言语线索,在现代代码审查(Modern Code Review)等各种平台中日益普遍。这些线索通常对开发者具有情感或指导意义。我们的研究通过审查这些评论中表情符号传达的情感和语义,深入探讨了代码审查评论(CR comments)的实用性。为了评估CR评论的有用性,我们用“表情符号特定”特征和预训练嵌入来增强传统的“文本”特征和预训练嵌入。为了加强我们的探究,在现有关于GitHub表情符号使用研究的指导下,我们用表情符号注释扩展现有数据集,并相应地重新评估CR评论。我们的模型结合了文本和基于表情符号的情感特征及对表情符号的语义理解,显著优于基线指标。CR评论中常被忽视的表情符号元素成为有用性的关键指标,表明这些符号具有重要分量。

英文摘要:

Emojis and emoticons serve as non-verbal cues and are increasingly prevalent across various platforms, including Modern Code Review. These cues often carry emotive or instructive weight for developers. Our study dives into the utility of Code Review comments (CR comments) by scrutinizing the sentiments and semantics conveyed by emojis within these comments. To assess the usefulness of CR comments, we augment traditional 'textual' features and pre-trained embeddings with 'emoji-specific' features and pre-trained embeddings. To fortify our inquiry, we expand an existing dataset with emoji annotations, guided by existing research on GitHub emoji usage, and re-evaluate the CR comments accordingly. Our models, which incorporate textual and emoji-based sentiment features and semantic understandings of emojis, substantially outperform baseline metrics. The often-overlooked emoji elements in CR comments emerge as key indicators of usefulness, suggesting that these symbols carry significant weight.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑