We looked inside some of the tweets by @RaulREsteban and here's what we found interesting.
Inside 100 Tweets
Preprint of our work on the COVID-19 Open Research Dataset is now out: Reference ontology and database annotation of the COVID-19 Open Research Dataset (CORD-19) https://t.co/EK23QLeRZK
Just published our article comparing social media to other sources to identify the side effects of statins. Social media provides information on side effects important to patients and more quickly than other sources. https://t.co/Q6410iRIv4
From our publication @DrugSafetyJour: "Patients on social media are proportionally far more likely to complain about musculoskeletal symptoms than other adverse events. Most adverse events showed a high level of agreement between Twitter and regulatory data" @UPennDBEI @nlm_news https://t.co/SodoC5PJaO
@ani_nenkova I looked at this some with @Xiaolei33 https://t.co/Sye3TaEhXH like the quoted tweet, our advice is to split train/dev/test chronologically, training on earlier data, tuning hyperparams on more recent data, and testing on latest. domain adaptation also improves robustness to time.
Analysis of temporal changes in reviews, news and tweets https://t.co/NayNO7GNdB
BioAsq session program online (CLEF 2020) Great talks and free registration, including our Mesinesp track on semantic indexing of medical literature, clinical trials and health related public projects in Spanish. Thanks a lot George, Tasos and Anastasia https://t.co/L5KdsidRD6
Excited to share our viewpoint, “The myth of generalizability in clinical research & ML in health care”, now out in @LancetDigitalH w/ Morgan Simons, @basslinetherapy @FinaleDoshi & Leo Celi https://t.co/bFtn0gGonc 1/x
Cool thread. I pretty much agree with everything here regarding our at times unreasonable overemphasis on cross site generalizability. It may be difficult to convince study sections though 🙂 https://t.co/6CCB9RmDLL
#nlproc folks: Has anyone used MTurk to get annotations for tweets to generate training data. Is this okay. What are best practices for anyone trying to do this. What are some ethical issues that we need to consider before doing this. How can one handle them appropriately.
Well, the bot announced it, so I should too :) https://t.co/s3oqAoxUoj #COVID19 Scientific Evidence Explorer Still in dev, feedback welcome. @SimonSuster @YuliaOtmakhova @eltimster @eltoroquerie @jibmaird & Shevon Mendis, Zenan Zhai, Biaoyan Fang, Jey Han Lau @ARC_AIMedTech https://t.co/BeUejE64k1 https://t.co/VuWKzgRXfp
"This tool uses a ML model trained on 1.7M PubMed Central documents to recommend suitable journals based on the textual content of your bioRxiv preprint." ✨ https://t.co/UTY18LHOYv A stellar proof-of-concept leveraging the full text of preprints/OA articles. (ht @carlystrasser) https://t.co/sX6FRXj0s9
MIMIC-IV is public! ~70,000 ICU stays, deidentified, ready for research. It's only 7 GB! ... but with >300 million charted observations, there's a lot to dig through. Quick thread on the highlights. https://t.co/N6Rzrrkjxg
Recent advances of automated methods for searching and extracting genomic variant information from biomedical literature https://t.co/1NMghi2OfD
Neural networks for open and closed Literature-based Discovery. https://t.co/jNtsI7V1ie