<shrug>
Of course ecomm audience management at their scale is super hard, but these are challenges most etailers face.
I really agree on this. The Youtube homepage exemplifies it for me. It recommends only content the same as or very similar to what I've watched before, and only the most popular of that. I've watched a ton of conference talks, programming tutorials, etc. yet those never show up. It's mostly junk with clickbait titles from big channels.
I do like Youtube's recommendation algorithm for the side bar suggestions on a video page, though I get more sponsored-looking content now than in the past. It's been useful at least for finding music. And YT does have a page for improving homepage recommendations, which granted I haven't tried yet.
Whereas the "related" sidebar while actually watching a video is great.
You do? It does ok if I'm watching something technical or some reviews. It's laughably bad at music and wider interest content such as a documentary. If you hit a music track, 50% of the recommendations are the same damn track in degrading levels of quality, 30% the same band. The 10% that's left for other artists are normally a really bad connection to the mood. The last 10% lately is taken by awfully bad match, clearly sponsored, junk though never identified as sponsored.
Most of the time I'd like suggestions in the same genre or mood of music. If I've just watched an upbeat rock track why not suggest other upbeat rock tracks? Last.fm radio was superb at this, for years, until CBS destroyed it. So it can't be that hard.
I think the only times I use the sidebar is if the vid I landed on is terrible quality or a bad hit.
I tried it out just now for an album I like. I get about 25% unrelated junk or stuff I've already watched, but the rest are also full albums that seem like fairly solid recommendations based on the album at hand (Jan Jelinek's "Loop Finding Jazz Records").
Start with a random metal track in mind an I find I get a band mix suggestion, a nordic folk metal mix, 11 more tracks from the same band, 3 tracks I listened to very recently, 4 connected other artists and "5 CREEPIEST Old Found Footages - Recommended for you" as a perfect example of the crap being injected recently. They must notice I've never once clicked one.
I do clear cookies and history periodically, so perhaps not having a multiple year history counts against.
In that other link I posted YT suggests pruning your watched history to get better recommendations -- but that's a huge undertaking given the thousands of garbage videos that I've clicked out of in five seconds. Seems like it would be better if the algorithm only operated on data from videos a user manually "likes".
Never hear of it. Can you share a link to it or a name of it?
https://support.google.com/youtube/answer/6342839?hl=en
I didn't look closely at it the first time, but now that I have, it's not that useful.
Adding insult to injury, I only buy lumber from their competitor, who, although having a sucky website where I can't look up dimensions, will let me use their compound mitre saw for free, saving me a $600-$1000 purchase.
I'd argue that the new algorithmic feed makes things less democratic because naturally the most popular images and posters will trend to the top of my feed, leaving the friends who post once in a blue moon towards the bottom (even though I'd like to see their images just as much). The randomness of chronological was nice, now I just browse in my bubble of existing interests.
What the hell do they want out of me at that point? I already pay for an account, and if I want a second account then I already know where to get one, from the same place I got the first one: eveonline.com
Jesus, the whole thing was farcical, for all their tracking and profiling.
On FB, that's the remarketing side, and it's a lot more manual. Generally the goals and data will be selected by the advertiser themselves. It works, but it's a bit blunt-weapon-y - let's say it'll catch 10% of people who are absolutely going to convert, and 90% of people who are absolutely uninterested for campaign X. All percentages here are pulled from nowhere, they're just to give you a general idea.
Remarketing is popular because conversions. Also, if you're well-off or otherwise a desirable target, you cost more to reach through Facebook - so you're likely to see more remarketing campaigns, because automatic campaigns will adjust to target less expensive individuals until they've gone through all of them.
Facebook's automatic targeting is a lot more sophisticated, and much more in line with the suggestion engine idea. Let's say it will catch 0.2% of people who are absolutely going to convert, 9.8% of people who might possibly convert, 45% of people who vaguely appreciate the ad being there without converting (including people who might Like/Share/comment on the post), and 45% of people for whom the ad doesn't really connect. It's not usually quite as powerful for conversion, but on average it's pretty accurate at guessing if you'll be interested.
Also, StumbleUpon was a decent suggestion engine once upon a time - if they could actually design a proper extension that did something similar, I'm sure it would be a success.
My favourite one is travels. They keep insisting that I travel to the place I just come back.
In the other hand, I see this as a good thing. We really don't want they target you perfectly, because that means that they know everything.
Advertisers with perfect information is, in my opinion, a scary proposition, but probably unavoidable in the long term.
That's actually what made me install adblock on my new system even though I was resolved to try to withstand ads this time.
"Oh you just bough a mobile case, let me show you ten other types" or "You just bought this book, here's the paperback version of it"
However, even with their inferior targeting FB and Google are making billions. So it might be the case that there is no further incentive to improve in this area and they are focusing on making their core product better to get even more users.
If I just added some hiking boots, show me fluffy socks, gloves and coats. Maybe a compass and rucksack. Nope, more boots x4. Also bought: More boots x2, 1 pair of socks and a USB stick. Top Sellers in Athletic & Outdoor Shoes: Guess what? Your recently viewed items and featured recommendations... At which point it's getting tedious.
The only time the also-bought ad engines work is when buying music or books. I often buy those in 3s and 4s, and same artist/author or similar genre work pretty well there. For most everything else I buy they just don't work and are about as welcome as a party political tv slot.
I'm amazed none of them have tried to get a bit better at this.
The Holy Grail is Just In Time or Contextual purchases. The opportunity to buy something when your need is greatest. Calendar app which advertises to you based on your activities. Or the Amazon Dash.
The problem is pretty much all advertising is corruptible and becomes evil. I press your Dash button when you aren't looking.
Part of the reason ads annoy me is that I'm just not purchasing stuff every day. (Who does?)
If I got the ads I'm supposedly interested in once a week, that might be more useful.
Even if I summarize all the analytics solutions out there, like ads, Facebook stream, Youtube, Amazon, Netflix. What they usually do is they figure out what I did in recent history or in general and present me more of that. I don't know about other people, but that's absolutely not what I want. That's boring, or maybe even unnecessary.
Like, if I just bought a book about Linux system administration the only reason I would buy a second one would be because the first one sucked (and Amazon knows the ratings of the books). A reasonable suggestion after that would be something people do with Linux, like hosting a Wordpress blog. Honestly I might not even know what I would do with my newly acquired skill once I succeeded with that topic. Also it's quite tiresome to learn. Maybe after reading 30% of it I would like to insert a simple but interesting novel or comic book. Why not suggest something like that to me?
Or I just watched an action movie on Netflix. Then I don't need other action movie suggestions. After that maybe I want to watch a character focussed tv show.
What we need is not a "I know what you did and here is more of that" suggestion engine. What we need is a "oh that's getting boring, here is other awesome stuff that may surprise you" suggestion engine. Figure out the stuff I don't know I want and then suggest that.
And honestly, not a single of these billion-dollar-engines out there do that for me at the moment.