Papers by Victoria Yaneva
Automated Prediction of Examinee Proficiency from Short-Answer Questions (2020.coling-main)
Copied to clipboard
| Challenge: | Existing approaches to predicting examinee proficiency from short-answer questions (SAQs) use of labeled data to train on is difficult, and requires expensive expert-rated data. |
| Approach: | They propose a method to predict examinee proficiency from short-answer questions . previous approaches train on manually labeled data to predict human-ratings assigned to SAQs . |
| Outcome: | The proposed model examines examinee proficiency directly and does not require manual training on labeled data. |
Predicting Item Survival for Multiple Choice Questions in a High-Stakes Medical Exam (2020.lrec-1)
Copied to clipboard
| Challenge: | Existing methods of pretesting are costly and require a large pool of test questions to be replenished, updated and expanded over time. |
| Approach: | They propose to automatically predict an item's probability to "survive" pretesting by embedding new items within a live exam and analyzing the responses. |
| Outcome: | The proposed method is based on human-produced MCQs for a medical exam and shows that survival is modelled through linguistic features and embedding types and features inspired by information retrieval. |
Classifying Referential and Non-referential It Using Gaze (D18-1)
Copied to clipboard
| Challenge: | a particular problem for anaphora resolution systems is the pronoun it, which can be used both referentially and non-referentially. |
| Approach: | They use eye-tracking data to learn how humans perform disambiguation and use it to improve automatic classification. |
| Outcome: | The proposed system outperforms a baseline and outperformed linguistic-based approaches. |
The USMLE® Step 2 Clinical Skills Patient Note Corpus (2022.naacl-main)
Copied to clipboard
| Challenge: | Large clinical note corpora are one of the most needed and one of least available resources in biomedical NLP due to patient confidentiality considerations and expert annotation cost. |
| Approach: | They present a corpus of 43,985 clinical patient notes (PNs) written by 35,156 examinees during the USMLE® Step 2 Clinical Skills examination. |
| Outcome: | The corpus of 43,985 clinical patient notes (PNs) written by 35,156 examinees during the high-stakes USMLE® Step 2 Clinical Skills examination is available via a data sharing agreement with NBME . |