Nutri-bullets: Summarizing Health Studies by Composing Segments

Authors: Darsh J Shah, Lili Yu, Tao Lei, Regina Barzilay13780-13788

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct human and empirical evaluation to comprehensively study the applicability and quality of these approaches. While seq2seq models enjoy high fluency scores, the extract-compose method performs much stronger on metrics such as content, relevance, faithfulness and informativeness.
Researcher Affiliation Collaboration 1Computer Science and Artificial Intelligence Lab, MIT 2ASAPP, Inc. darsh@csail.mit.edu, liliyu@asapp.com, tao@asapp.com, regina@csail.mit.edu
Pseudocode No The paper describes its methods using text and mathematical equations but does not present any structured pseudocode or algorithm blocks.
Open Source Code No Our code and data is submitted and will be made publicly available on acceptance.
Open Datasets Yes Our Healthline6 and Breast Cancer7 datasets consist of scientific abstracts as inputs and human written summaries as outputs. 6https://www.healthline.com/nutrition 7https://foodforbreastcancer.com/
Dataset Splits Yes We randomly split both Health Line and Breast Cancer datasets into training, development and testing sets(see Table 2). Table 2: Statistics for scientific abstracts, Health Line and Breast Cancer datasets. Data Train Dev Test Input Scientific Abstracts 6110 750 866 ... Health Line summaries 1522 179 193
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or cloud computing instance specifications used for running the experiments.
Software Dependencies No The paper mentions software components like 'Neural CRF tagger', 'BERT text classifiers', and 'Blank Language Model (BLM)', and cites their respective papers, but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes We set we, wd and ws to 1 and wm to 0.75. rp is 0.02 and δ is 0.99. We use a large, 6 layer BLM (Shen et al. 2020) for fusion.