Somehow, I follow more than 200 people on twitter. I may need a whom-not-to-follow feature more than a whom-to-follow feature. My current location on this sinusoidal interest graph not withstanding, this is actually a very interesting problem. How do you connect up people to that which they are receptive to, that which they are looking for, that which they will appreciate when they see it – but may not know to look for it! Catering to “I want something, I dont know what, but I will know it when I see it” is not easy. But it makes for an interesting problem. Today Recommendation Engines are mostly based on available algorithms to mine existing data, determine a pattern and suggest something else that falls within the same pattern. But I am ready for more. I want Mr. Jeeves in place of the robot. I want some whimsy. I want someone who pays attention to what I am suggestible-to at a given moment.
Netflix, Amazon, Pandora, Facebook, LinkedIn, Twitter – all have ways to recommend things to you. Pandora has a few interesting things working in its favor (in addition to the core of their recommendation engine).
It is interesting to look at how Twitter compares to Pandora along these axis.
The Recommendation Moment and the Consumption Moment:
People will consume Pandora’s recommendations at the very moment they are made. Compare this to Netflix. Anyone who has returned a movie unwatched knows what I mean. It seems great when you put it on the queue. Later, not so much. So it sits there. And it is returned. Unlike this, with Pandora (and Netflix streaming) the instant of making the recommendation is the same as the instant of consuming the recommendation. On this dimension twitter represents something that you the user will be consuming for a long, long time after the instant of recommendation.
The Cost of Accepting
With Pandora or Netflix streaming, the cost of accepting a recommendation is rather low. You check it out, if you don’t like it, you stop and move on. The design of Pandora, it is all about reducing the cost of accepting a bad recommendation. With twitter, it may seem that the cost of accepting a recommendation is low – but in reality it is not. Un-follow is a heavy and emotionally expensive click.
Homogeneous or Distinct
I don’t know if it is conditioning by radio channels or if radio channels evolved this way due to the way we are designed to consume music: but typically when you are in the mood for a certain kind of music, you want more and more and more music of that same kind. This is less true of books or even movies. Infact right after watching an intense historical drama, I might be most suggestible to a situational comedy show. Now with twitter, my experience has become richer every-time I have found an entire new interest represented. For instance interaction designers, or snarky commentators on India, comfortable in Hinglish. Once I found these groups, I could see who the people I found are following and add to people who represent this interest. Who to follow makes this part easier. However, it is the other thing – finding oasis of people who represent a different interest – that makes for a quantum leap in twitter-interestingess.
Length of Engagement:
For Pandora, the length of engagement with the recommendation is the average length of a song. For twitter, it is in some sense an infinitely long length of time.
Recommendation Basis: The basis for recommendations in Pandora is an analysis of the object itself. Contrast this with what Amazon does. Amazon looks to see purchase and browsing patterns, to approximate a recommendation ( they do not use a deep categorization of the books, for instance). (Netflix, ofcourse having run the million dollar contest, does a variety of things that are interesting). If we agree that twitter is ultimately trying to hook us up with more interesting tweets in our timeline, this current mechanism is more of a secondhand recommendation (like amazons) than Pandora’s.
In addition to those meta observations, here are some suggestions based on my personal experience so far:
Who-not-to-follow: Find out who I have explicitly un-followed. Do not ask me to follow them. Find out people that you have shown me that I have ignored, do not show me those people. If they have appeared 3 times on my sidebar and been ignored (less than 3 actually), that is an explicit (but no click) thumbs down.
The sounds of silence: If they have not said a word on twitter for three months, they should not be on the WTF list.
Recognize an Omission: For some people on twitter, who to follow could be showing a lot of names that they were choosing explicitly not to follow. If 5 of the people I follow, also follow someone, that is actually a clear indicator that I choose not to follow that person.
Sometimes, Less is More: It is also possible that not all the people who are on twitter will really have a richer experience if they start following even more people. It really depends many things, including how many people they are already following. Users with a very active timeline may have a higher resistance to adding more people. I, for instance, would prefer to get one recommendation every week or even every other week, rather than seeing something new every time I refresh the screen.
Tell me why: I want to know why you think this is a good person for me to follow. Tell me. Right up front. Right now all I have is the persons name. It works for the very famous (who turn out to be really ineffective tweeters at times, but that is another story) – but for everyone else – I want to know more. I want to know why you think this is a good fit. I want to know why they are on my wtf.
Show me the content: I want to see the tweets. That is what helps me make the decision. If you don’t show it to me, I have to click and look.There is a part of me that wants one or two “suggested” people to simply appear in my timeline with an explanatory hashtag? I can see that this suggestion is fraught with peril, but really, a name is nothing to me many times.
And in closing: People often comment that while Facebook is the place to connect to people you know, twitter is the place to connect to the people you -wish- you knew. So you want to recommend people for me to follow on twitter – you have to figure out what I -wish- for. Tall order? Yes. But I am afraid, it is the feature I would request. I do not wish to be “Rolodex-complete” on twitter. Make it an opt in feature if you need to, but be adventurous. Not a lot of people, just one or two a month. But recommendations that would make me smile, even if I decided not to act on them.
- Some people on my timeline voicing similar opinions: @psivadasan, @daddy_san , @i_r_squared
- My stretch goal for this post: do not use “social graph” anywhere.
- Tangentially related reading: http://www.slate.com/id/2263462/pagenum/all/