Microsoft Academic

» see academic.microsoft.com as it lived, on the wayback machine

~ the obituary ~

i read papers for work. computer vision on small hardware means living in arxiv and in whatever search layer sits over it, and for me that layer has always been google scholar. microsoft academic is the one i opened when scholar could not find a citation count i trusted, which was not often. this is a grave for a tool i used the way you use a second dictionary.

microsoft had been at this longer than most people remember. windows live academic search came out in 2006 and died two years later. microsoft academic search followed in 2009 and development stopped on it around 2012. the third version, plain microsoft academic, launched on 22 february 2016 and was a different animal: instead of indexing pages it ran the bing crawl through machine learning to pull out entities, so a paper arrived already linked to its authors and to the institutions behind them.

it got large. over 260 million publications by the end, 88 million of them journal articles, with the graph underneath exposed through a free api that anyone could query. bibliometrics people used it for h-index and g-index math. it was the only serious free competitor to the paid citation databases and to scholar, which is free but shut.

on 4 may 2021 microsoft announced that the site and its apis would be retired on 31 december that year. no drama, no explanation beyond retirement. the graph stopped updating on the last day of 2021 and the search box went away.

~ the case file ~

~ what made it different ~

google scholar tells you what it knows one query at a time. there is no bulk export and no open api, so you cannot ask it anything about the corpus as a whole. every large-scale study of who cites whom has to work around that.

microsoft academic published its entire dataset under an open license and shipped a new dump every two weeks. researchers could download the graph, run their own analysis and check the result against someone else's. a company sitting on a proprietary index chose to hand it out, which was odd in 2016 and looks stranger now.

~ what replaced it ~

because the data was open, the death was survivable. a nonprofit called ourresearch took the final dump as a starting corpus and launched openalex on 1 january 2022, days after the graph went dark. the data carried over. the software did not, since the machinery that built the graph stayed inside microsoft and had to be written again from nothing.

openalex now sits underneath a lot of research infrastructure that most academics never see by name, and every citation in it traces back to a microsoft product that no longer exists. that is a better ending than most shutdowns get. it only happened because somebody at microsoft research picked a license in 2016 that made the corpus impossible to take back.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »