Transfer Learning for Closed Domain Question Answering in COVID-19
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
COVID-19 has been a popular issue around 2019 until today. Recently, there has been a lot of research being conducted to utilize a big amount of data discussing about COVID-19. In this work, we conduct a closed domain question answering (CDQA) task in COVID-19 using transfer learning technique. The transfer learning technique is adopted because a large benchmark for question answering about COVID-19 is still unavailable. Therefore, rich knowledge learned from a large benchmark of open domain QA are utilized using transfer learning to improve the performance of our CDQA system. We use retriever-reader framework for our CDQA system, and propose to use Sequential Dependence Model (SDM) as our retriever component to enhance the effectiveness of the system. Our result shows that the use of SDM retriever can improve the F-1 score of the state-of-the-art baseline CDQA system using BM25 and TF-IDF+cosine similarity retriever by 3,26% and 32,62%, respectively. The optimal parameter settings for our CDQA system are found to be as follows: using 20 top-ranked documents as the retriever’s output, five sentences as the passage length, and BERT-Large-Uncased model as the reader. In this optimal parameter setting, SDM retriever can improve the F-1 score of the state-of-the-art baseline CDQA system using BM25 by 5,06 % and TF-IDF+cosine similarity retriever by 24,94 %. Our last experiment then confirms the merit of using transfer learning, since our best-performing model (double fine-tune SQuAD and COVID-QA) is shown to gain eight times higher accuracy than the baseline method without using transfer learning. Further fine-tuning the transfer learning model using closed domain dataset (COVID-QA) can increase the accuracy of the transfer learning model that only fine-tuning with SQuAD by 27, 26%.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it