Pinterest and Tumblr vs. Hunch: Two Approaches to Solving the “Discovery” Problem
One feature that I think helps drive Pinterest’s (and Tumblr’s) popularity is the ability to re-pin (in the case of Pinterest) or re-blog (in the case of Tumblr) something someone else has flagged or published. Hunch, which was sold to eBay a few months ago, is primarily a B2B app, but it was initially a consumer-facing app that tried to predict what content from the web you’d like based on your past activities/selections (something they call the “Taste Graph”).
One way to think about this is as two approaches to the same problem — how to determine what content you want to hear about. Pinterest and Tumblr are socially-driven and in that sense are low-tech: you do a little work up front to choose certain people or topics to follow, and from there what topics and content you see is driven by organic social mechanics — coming across an interesting person that someone you follow also follows (on Pinterest), seeing who else re-blogged the same article you re-blogged (on Tumblr), etc.
In contrast, Hunch’s method as I understand it is more “high-tech” — essentially developing a machine learning algorithm, similar to other recommendation engines such as Netflix’s. Bayes’ Theorem and all that.
Interesting to think about which is “better”, particularly from the perspective of “work required to develop per unit of user satisfaction”. Of course the answer probably depends on what context you’re thinking about. Socially-driven sites will score very high on the “user satisfaction” score (I know I like something if I choose to follow the person that created it, whereas we can all think of times when we were really confused by a recommendation that an algorithm came up with). It also would seem to score higher on the “less work required” scale as compared to machine learning approaches.
But perhaps the machine learning approach is better on the user satisfaction score for (a) less social activities such as shopping on an online marketplace or watching a movie at home on my couch, and (b) enterprise (instead of consumer) more broadly, particularly for “big data” uses (buzzword alert!) — BloomReach comes to mind.
Machine learning / natural language processing is a huge area with plenty more research and product-ization to come, so it will be interesting to see what applications come to market and whether there are successful new consumer-facing applications that rely heavily on machine learning (are there already such applications that I’m not thinking about or don’t know about?).
As an aside, I don’t use Etsy much, but it could be an interesting case to study, since it’s a marketplace like eBay but it “feels” more like Pinterest in terms of its design and as such buyers may be tempted to buy based on what others with similar design tastes are buying.
