For the last half year I’ve been working on laying the groundwork for giving Lazy Scholar the ability to extract helpful information from a paper and put it in a sidebar. Most of this information is extracted using text matching rules, and some using machine learning. Moving forward, these extractions will be used to generate datasets to train more machine learning models (if you’d like to help with this effort see the instructions below). Extractions from this panel will improve iteratively over the coming months, and the panel’s design will likely change based on feedback. To this end, I have a new feedback form to make it easier to send ideas and bugs that is accessible by right clicking the icon next to the address bar -> Feedback. Please send me ideas and issues!
Here’s how you access the new panel, via the new yellow brain icon. Clicking any of the extractions will scroll the page to the extraction or sentence. Want it to open automatically on every scholarly page? Click the Options at the bottom of the panel to find that option.
Here are the current (attempted) extractions:
- Information Extractions:
- Population/Subjects: Number of participants, animals, etc in the study
- Intervention: If relevant, what the intervention(s) were in the study
- Outcomes: What the study looked at
- Measurement/Products: If the sentence contains a product to measure an outcome, the sentence will be extracted. If possible, only the outcome will be shown, and either way clicking it will scroll to the sentence
- COI/Funding: Extracts the conflict of interest/funding statement
- Paper outline: Extracts the basic outline of the paper and clicking scrolls to the section.
- References: Attempts to extract the references and shows quick links to the paper and to the paper on Google Scholar. Hovering over linked references within the paper should scroll to the reference (see demo below).
- Abbreviations: Attempts to extract abbreviations from the paper and puts them in the panel. Hovering over them shows the full definition, and highlighting the abbreviation within the paper will show the definition in the panel (see demo below).
- Supplemental Documents: If it finds a linked supplemental file on the page, it displays a link
- Paper Classification: ie Human, Animal, RCT, etc. Currently this is from MeSH tags and not extracted, but future iterations will attempt to classify automatically.
Help make it better
At this point, not all extractions will be right. I need more examples of extractions that are correct and incorrect to improve accuracy. If you would like to help, click the “Beta” bar in the panel as shown below and the link to open instructions. After you’ve read the instructions and clicked to accept, you’ll see thumbs up and down icons as shown below. Rating extractions (you can do as few or as many as you’d like) will help me make the panel much more useful. If you see really strange behavior, feel free to send me a link to the paper via the feedback link by right clicking the icon next to the address bar and hitting “Feedback”.
Update 5/2/18, version 2.1.0: extractions now work in PDFs (see video below). Also in this version, hypothes.is annotations load over a PDF within the extension instead of using their external proxy.
Currently, this update is for Chrome only (Firefox is coming soon).
Finally, an additional feature has been added to this version:
If you turn on the option to rename PDFs in the Options (right click the icon next to the address bar -> Options), Lazy Scholar will attempt to rename PDFs to a format with ‘FirstAuthor Year Journal.pdf’ instead of long filenames such as ‘3g7d828fjdf38rg8f.pdf’. This was a feature in version 0.x of Lazy Scholar but was removed until I could improve it.
Some bugs fixed and underlying improvements made. Stay tuned for more improvements in design and functionality in upcoming versions.