Something about systems knowing our preferences deeply unsettles us, at a visceral level. Other people knowing our preferences, hell no. Not unless it is carefully crafted to fit a social narrative. Preferences are a part of our personality. And we care about how people see us.
Here comes LinkedIn and Facebook which broadcasts everything we do to our network. They liked this article, they shared this post, they connected with this person, they followed this news. Everything we do, broadcasted. And everyone who comes across this forming their own opinions, coloured by their own biases, about us.
The first hypothesis is people don’t like giving out these signals because it also signals their preferences to their network, which they don’t really want to do? I don’t really want my boss to know I like watching cat videos on loop, what if he thinks I am not working as much as I should?
2. The stupid, greedy recommendation systems
Here’s a scenario. The first time I had momos I come home and tell my mother about it. Raving about how good it is. She says momos are simple, I could make it for you at home. That doesn’t preclude the fact that momos are amazing. She took it to heart (how can he appreciate momos so much and not my biryani?? that is significantly harder to prepare, so much effort - or something along those, I dunno. I am guessing)
Next day she makes momos at home. I like it. Say it is amazing. Then she makes momos the day after again. It took me just two days to get over momos. I went from loving momos to almost gagging at their sight. It was too much!
Just because I tell you I like something doesn’t mean I want to keep doing that immediately and repeatedly. Amazon - I bought an oven. Like 2 weeks back. Why are you still recommending ovens to me? You know I have one. I bought it from you!*
YouTube! Please! Once I watched that video. Sure I liked it, but I want to other stuff too man. Or the flip side, I have watched every big content piece of this particular creator, the lifecycle is done. Why are you still showing me pieces “related” to them? I just like the comedian, I have seen all their skits. I don’t want to watch interviews and scandals and what they had for dinner on Golden Globes night, come on.
The second hypothesis is people are afraid of telling their preference to the system because it will go berserk-mom-mode all over them. A weakly held hypothesis, one I think mostly applicable to tech savvy people. Probably a minuscule portion of the overall population. Just the geeks?
But it may not be as minuscule I am thinking. You can see the effects your watch history is having on your feed. Question is, is there cross learning happening in users across platforms?
I believe so. We rely on the explicit signals for our ground truth and try to map the implicit signals to them.
Our recsys literate consumers don’t give us signals because they have been burnt in the past, our dataset doesn’t have their representation, and our models perform poorly for them….
Netflix seems to be onto something with their single thumbs-up, single thumbs-down and double-thumbs-up approach. Almost like saying “Hey just tell us if you like this, it is extremely useful for us to know. We promise not to force content like this down your throat. Unless you LOVE it. Tell us that and we will know. We are here to help :) . ”
Part of evening discussions over nachos with @Ritesh
* I have been told Amazon doesn’t do this anymore, it is smarter with respect to the item life cycle now.