Microsoft’s New CEO: Breaking News with the Knowledge Graph

The world awoke on Tuesday to a new Microsoft CEO. Satya Nadella, the former Head of Cloud Computing, had been promoted in a pretty uneventful affair replacing the incumbent CEO Steve Ballmer.

The media picked the news up exceptionally quickly and the story spread around the web like a wildfire with news outlets, bloggers and social media all talking about what the appointment meant for the business and what changes Nadella would be likely to make.

There was one place however that didn’t even notice that anything had even happened. Google’s Knowledge Graph.

Google’s news search served an updated story on the chief executive switch, of course, but the first visible result was provided by the Knowledge Graph, and despite it being a database containing encyclopedia entries on about 570m concepts, relationships, facts and figures, it was quickly made out of date by the Microsoft move. A fact I’m sure wasn’t lost on Nadella, a former Internet search employee.

Steve Ballmer

I was alerted to this anomaly by Samuel Gibbs of the Guardian who wrote about the lag in the system. So I jumped at the chance to examine the issue in real-time.

The Knowledge Graph is fueled by a number of knowledge bases that push facts to be used in information panels, however, as shown in search patents unearthed by Bill Slawski, these need to be verifiable. Essentially Google needs two sources of information to verify against before they will insert data into a panel. This seemed like an ideal area to investigate further.

As Wikipedia is seen as an important source of information for the Knowledge Graph I began scanning through dbpedia, a database of Wikipedia used by many semantic web applications. Diving into the RDF output for Nadella and Ballmer’s entries came up with nothing to cause alarm. Both sets of data had been updated to include their new employment status.

Updating the Knowledge Graph

Now that there was one definite source of data on the web, I took to Freebase to edit their profiles to see if the Knowledge Graph could be “kicked into gear”.

After a few minutes I had entered Nadella’s new CEO status at Microsoft and updated Ballmer’s new employment details. Now I had to wait.

Using a tool called Page Monitor I tracked the RDF output of Freebase to see of there was a correlation between the time of publication to the moment the Knowledge Graph updated with the new information.

Alas, just a few hours after editing, the RDF dump had updated followed quickly by an updated entry within the SERPs:

Microsoft CEO - Satya Nadella

So what does this tell us about the Knowledge Graph?

Verified sources
We have long understood that the Knowledge Graph needed multiple sources of information to populate a panel for an entity, and thanks to patents we had an indication that two separate sources of information would be enough to influence results. However, seeing this (albeit rough and ready) experiment in the wild gives a solid sign that this may well be the case.

Freebase as a source
Freebase is seen by many (myself included) as key to the growth of the Knowledge Graph and other semantic agents. This shows how Freebase data can also be used as a source of user-generated information that can be passed into the Knowledge Graph.

Time sensitivity is an issue
Last but not least, this debacle shows that for time sensitive information such as breaking news; the Knowledge Graph simply isn’t ready. The process of becoming (or editing) an entity isn’t well-known and as such will hamper the ability for Google to keep its panels updated to respond in (almost) real-time.

  • http://twitter.com/krystianszastok Krystian Szastok (@krystianszastok)

    It reminds me of the lag a few years ago in Google’s serps and the slow indexing.

    Do you think they’ll do a ‘caffeine’ update for Google’s KG soon?

    I think some of it is currently manually edited/approved, maybe that’s where the lag comes from too.

    • Andrew Isidoro

      Yeah definitely. There seems to have been an update back in 2013 that Dr Pete covered that opened up Knowledge Graph entries but they don’t seem to have cracked the speed vs quality conundrum quite yet.

  • http://www.jellyfish.co.uk/ Daryl Cygler

    Andrew,
    Great bit of insight. And this was around global news information that was significant on some many levels, big business, corporate news, political news, stock market news. So for the everyday snippets the lag will increase even more. KG; Good in principle just needs to be more real time, with quicker understanding and referencing, but i am sure this will develop. Hell, Twitter was not even around 10 years ago and look at the real time impact that has had as a platform..KG; Build it and they will come….
    All told, nice piece..my knowledge of the knowledge graph has just increased…quicker than Google will ever know…