1.7 KiB
1.7 KiB
type:: conferencepaper date:: 15-03-2024 - 16:11 full-title:: Good things come in three: Generating SO Post Titles with Pre-TrainedModels, Self Improvement and Post Ranking external-links:: ESEM2024-FILLER - Online LaTeX Editor Overleaf todoist:: 7795832448 year:: 2024 date-start:: status:: ACCEPTED date-submitted:: 06-05-2024 deadline:: 06-05-2024 venue:: ESEM priority:: P1 parent:: leader:: people/phuong progress:: {{renderer :todomaster}}
- We need to highlight that the technical contribution is not on a new way of combining the two strategies (i.e., self-improvement and post-ranking), but instead on their application on a very specific SE problem
- Clarify the usage of TextRank and its suitability and effectiveness in the context of post-ranking
- TextRank is typically employed for selecting keywords or key sentences
- The settings of RQ3 is lacking a configuration that excludes self-improvement alone to understand its individual impact on the performance of the model
- The self-improvement expands the size of the training data, thus the observed improvement could be attributed to the increase in the quantity of the dataset. Thus to ascertain whether the data augmentation inherent in the self-improving process mitigates such issue, we could compare the performance of the models trained on datasets generated through the self-improving process with those trained on datasets derived from duplicating the original dataset.
- Tasks
- DONE Sezione 2 [[30-04-2024]]
- DONE Sezione 3 [[30-04-2024]]
- DONE Sezione 4 [[01-05-2024]]
- DONE Sezione 5
- DONE Sezione 6
-