Source code level word embeddings in aiding semantic test-to-code traceability (2019)
| Proceedings of the 10th International Workshop on Software and Systems Traceability
Source code level word embeddings in aiding semantic test-to-code traceability
https://scrapbox.io/files/60d83ecb80849a0021adac4f.png
https://scrapbox.io/files/60d84613b90319001c391cf9.png
三行まとめ
構文木からidentifierを抜き出してdoc2vecにかけて、similarity が高い順に TCT を推薦したら良い精度が出た
hr.icon
Motivation
hr.icon
どんなもの?
We provide a method, that automatically links test cases and production classes relying on only conceptual information.
conceptural information って何?
We propose a generic and efficient technique that maps test cases to production classes
We adapted the doc2vec technique as a traceability link retrieval method, which to the best of our knowledge is a novelty in software engineering
Code to vec とか使わないの???
We show that textual similarities provided by the doc2vec technique approximates the naming convention technique rather well
これはうまく行かないと思う。naming convention がうまく言ってるのはガチでカチッとしたprotocolに則ってるからであって、機械学習使った approximation を行うとうまくいかないと思う。
We demonstrate that the doc2vec approach can substitute and even outperform the LSI technique in a traceability task
hr.icon
先行研究と比べてどこがすごい?
hr.icon
技術や手法のキモはどこ?
hr.icon
どうやって有効だと検証した?
hr.icon
議論はある?
hr.icon
次に読むべき論文は?
NC 以外の traceability recovery 手法