LearningToRank extracts six features based on domain knowledge, including the lexical similarity score, the collaborative filtering score, the bug-fixing recency score, the bug-fixing frequency score, and so on. Then, a learning-to-rank approach is utilized to train the features’ weights.
Contributions of The Paper
Key contribution-
using API description to bridge the lexical gap between bug reports and source code
exploiting previously fixed bug reports as training example for determining model parameters using learn-to-rank technique
new benchmark dataset (created by checking out before-fix-version of the source code package for each bug report)
six Features: surface lexical similarity (normal textual cosine similarity), API-enriched lexical similarity, similar bug information scores (collaborative filtering score), class name similarity score, bug fixing recency score, bug fixing frequency score
ranking mode is based on a weighted combination of features that capture domain dependent relationships between a bug report and a source code file. The model parameters are trained using the learning to rank [1] approach as implemented in the SVM rank package
Ref:
[1] https://dl.acm.org/doi/10.1145/775047.775067
Publisher
FSE
Link to The Paper
https://dl.acm.org/doi/10.1145/2635868.2635874
Name of The Authors
Xin Ye , Razvan Bunescu , Chang Liu
Year of Publication
2014
Summary
LearningToRank extracts six features based on domain knowledge, including the lexical similarity score, the collaborative filtering score, the bug-fixing recency score, the bug-fixing frequency score, and so on. Then, a learning-to-rank approach is utilized to train the features’ weights.
Contributions of The Paper
Key contribution-
Comments
No response