The example input in the article is:

  [1,[2,3], [4, [5,6]]]

It happens to be representable as a binary tree. Indeterminacy regarding whether or not that is accidental or intentional is due to underspecification of the interview problem. The first time I read a description of Linkedin using logs as the fundamental data structure (for what turns out to be Kafka I later learned) I had an epiphany that there is no such thing as a generic simple data structure living in the wild because data is data because it has semantics, without semantics it is just noise.

To me, it looks like the root node is |1|.