Facebook Graph Search: I got it, I even queried the author – but I don’t get it yet

Facebook’s new graph search is in closed beta at the moment.  I managed to hurl myself onto it  before the door shut.   I played around with it for a while, then I stopped, waiting for inspiration about how to make it as interesting for me as I was sure it was going to be.  I  haven’t figured it out yet.   So when I saw that the lead engineer from Graph Search, Kari Lee, was giving a Facebook Tech Talk this evening at Facebook’s London office, I hurled myself at the guest list, and squished myself on before the door shut.

My problem is not that I’m uninterested in graph search.  I think it’s fascinating.  Honest.  I even know more than a bit about it, and I still think it’s fascinating.   And I am convinced that for Facebook, the potential here is huge.  Beyond huge.  I am pre-sold on that.   It’s just that, as currently constituted, Facebook’s beta version doesn’t do anything for me.   I do wonder occasionally what I should ask of it.  And fail to engage myself.  Then I try not to be parochial and I think of myself as not being myself, but being at different life stages, doing and wanting different things.  So far, it hasn’t helped.

Facebook’s graph search icon

Facebook’s previous version of search was awful.   I blamed this on Bing, rightly or wrongly.  So, with graph search, I was looking forward to seeing something that shone a headlamp into the murk of the future, even if it what was illuminated wasn’t already a polished jewel.

The content of the talk this evening focussed almost entirely on the natural language processing done by the query constructor, which is impressively free form, and does a lot of “search ahead”, which, given the nature of the data it is trying to reach, is not trivial.  It’s all also updated in real time.   A lot of heavy lifting had clearly been done: they showed a picture of Kari’s team fillling up a massive “statement” staircase.  Mention was also made of how the indexes were stored.  I did wonder how much sense the talk made to much of the audience, as the majority had not been able to cram themselves onto the beta – possibly because they were ineligible as they spoke UK-english, which is not yet supported.  (I’m not sure how or why FB inferred that I spoke US English as one of my dialects.)

Kari went into considerable detail about their strategy and approach for query NLP and UI.  But, to my mind, the talk wasn’t about graph search.   I really thought it was about a front end to graph search.  I asked her about this, as best I could.  I thought she was treating the problem as if it was entirely about query resolution, and the problem ended there.  She answered that what she did was the parser, and that was what she talked about.   Which was certainly a straightforward and honest answer, but it left me feeling rather puzzled.  I think the good bit’s the next bit.  And by that I do not mean the cleverness of how it’s all distributed.   I mean what it does.

Facebook clearly are into graphs from an algorithmic perspective.   That was, in a way, the subject of their recent Kaggle competition which they are using to try to hire more data scientists.  So my bet is that someone, somewhere inside Facebook is thinking about it.  It just hasn’t been surfaced yet.   Graph search, as it was presented this evening,  is already being used internally in Facebook, in two products (ie. services) currently in development.    So we’ll see more of it.   Perhaps what it most needs is a context.  So I look forward to seeing more about how that develops.

[You may have noticed a recent blog theme of “stuff I see while wandering around London at night”.   (You’d be right…:-)]