This post is contributed by Paul Mucur, CTO at Altmetric.
On Wednesday, 24th September, eleven people armed with laptops and some altmetrics-related ideas descended on Macmillan Campus for the first 1:AM London Hack Day.
If you’re unfamiliar with hack days, the premise is to put developers and domain experts together and see what can be built within a day or two (for other examples, see the NHS Hack Day or BERG’s Little Printer Hack Day). In our case, we had the loose theme of altmetrics (though any scholarly idea was welcomed) and developers came from Altmetric, Papers, Elsevier, writeLaTeX and Nowomics to work together.
After caffeination and introductions, those with ideas (some shared beforehand on our wiki) pitched them to the others and three teams were formed:
- Buzzmetrics: following research that there is a link between the length of a paper’s title and its resulting citations, an exploration of how titles affect altmetrics (if at all);
- User journeys: looking at altmetrics in social media and reference managers to predict usage (e.g. is there a spike in reference manager usage after a paper is tweeted?)
- Altmetric ORCID profiles: given the ORCID ID of a researcher, generate a profile with altmetrics for each of their papers.
Several hours (and quite a few sandwiches) later, we wrapped things up and each team presented their work.
John Lees-Miller of writeLaTeX presented the “Buzzmetrics” team’s findings (consisting of Richard Smith from Nowomics and Oliver Martell and Matt Peperell of Altmetric). They downloaded a set of 10,000 articles from Altmetric’s API and wrote a series of feature extractors to analyse how their titles affected their score (if at all) such as:
- Title length
- Presence of question marks
- Presence of colons
Their code can be found at https://github.com/amconference/buzzmetrics and result of their analysis at http://rpubs.com/jdleesmiller/30864. While John was careful to state none of the regressions resulted in anything with a high coefficient of determination, we enjoyed learning that having question marks in your title counts against you but a judicious colon and some pithy jargon might help.
The second team consisted of Matt MacLeod, Maciej Gajewski and Scott Matthewman of Altmetric and Ben Blackburne of Papers. Their project, combining Altmetric data with ORCID profiles, was presented by Matt MacLeod.
Their code can be found at https://github.com/amconference/altmetric-orcid-profiles and a test version is running at http://altmetric-orcid-profiles.herokuapp.com/
Given an ORCID ID (such as Jason Priem’s), their web application will present a profile of that person’s papers with any relevant altmetrics data, giving a breakdown of their mentions.
While all of that was very impressive, it was the spinning Altmetric donut-themed loading indicator that drew the most gasps from the audience and the team admitted a disproportionate amount of time was spent perfecting its hypnotic animation.
The final team of Narayanan Subramanian, Terry Adeagbo and Mike Taylor of Elsevier presented their findings regarding the relationship between altmetrics and reference manager usage.
They started by taking a hot trending topic from Google Trends (in this case, the recent Scottish Referendum) and then used that to find papers from CrossRef about that topic. Having collected some DOIs, they then queried the Altmetric API for altmetrics around those papers and compared those to readership numbers from Mendeley.
A correlation seemed to emerge between Twitter activity and Mendeley readers: that a spike in tweets leads to an uptick in readership but there wasn’t enough time to explore a more thorough time series of the data and to bring in Mendeley’s demographics.
Terry has since shared these visualisations from the group:
- A breakdown of article types, subjects and publishers
- A breakdown of geography, academics and discipline
- A graph of issued date vs social media first seen date
All in all, it was a successful day and it was great to see people from a variety of backgrounds working together on the same theme. Roll on next time where hopefully we can develop these ideas even further!