Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Real World Recommendation System

blog.fennel.ai

Post reply on HN
Print view View original

31–40 of 156 posts

Re: Real World Recommendation System

#31

Post by imilk »

rexreed wrote: post #10

Isn't this obvious list-building promotion for a company (Fennel) that sells recommendation systems? "Fennel AI: Building and deploying real world recommendation systems in production Launched 18 hours ago" Caveat reader.

I'll never understand why people think this is a valid criticism of an article, rather than pointing out an issue they have with the actual content of the article. There's nothing inherently wrong with a company sharing info about the space they operate in. In fact, it should be encouraged as long as what they share is useful.

Re: Real World Recommendation System

#32

Post by imilk »

chuckcode wrote: post #21
fmakunbound wrote: post #5

With all of this technology applied, I am still disappointed by Netflix's recommendations – to the point of just giving up and doing something else.

Do you think part of this is that Netflix has assumed zero effort from user model? My experience has been that Netflix does an ok job of recommendations, but fails at overall discovery experience. There is no way for me to drive or view content from different angles easily. I end up googling for expert opinions or hitting up rotten tomatoes to get better reviews. Netflix knows a ton about me and their content, but se…

Google TV has the best content discovery I've come across so far. Recommendations across most streaming services based on overall similar movies, different slices of the genre, and movies with similar directors/cast members. Plus as soon as you select another movie, you can see all the same "similar" recommendations for that movie.

Re: Real World Recommendation System

#33

Post by samstave »

vikingcaffiene wrote: post #3

Gentle reminder to anyone reading this that your problems are probably not FAANG problems. If you architect your system trying to solve problems you don't have, you are gonna have a bad time.

Wow, this is something that has been a floater-in-mind for decades ;

I'll top it off with an interview at Twitter with the Eng MGR ~2009-ish?

--

Him: So tell me how you would do things differnetly here at twitter based n your experience?

ME: "Well, I have no idea what your internal processes are, or architecture, or problems, so my previous experience wouldn't be relevant."

I'd go for the best option that suits goals.

[This was my literal response to the question, which I thought was a trap but responded honestly -- as a previous mgr of teams, the "well, we did it at my last company as such"]

Dont reply this way. Here was his statement:

This is a literal quote from a hiring manager for DevOps/Engineering at Twitter:

"Thank god!, We have hired so many people from FB, where that was there only job out of school, and no other experience, and the biggest thing they told me was "well - the way we did this at FB was... X"

--

His biggest concern was engineering-culture-creep...

Re: Real World Recommendation System

#34

Post by samstave »

siskiyou wrote: post #24

All I know is that Facebook's recommendation systems always show me things that I hate to see. I suppose they may "work" at scale, but at an individual level it's epic failure.

FB needs an Ad-Rev-Share-Model with ALL of its users...

Imagine if FB were to pay a fraction% of how yur data was used and paid you for it...

It may be a small amount, but in super 4th world countries, it could affect change in their lives...

Now imagine that this becomes big... and it works well.

Now imagine that the populous is aware of the hand of god above them just pressing keys to affect land masses (yes I am referring to the game from the 80s)

but this cauterizes them into union building...

So when the people realize their metrics are the product to feed consumerism for capitalistic profits, and decide to organize, what happens?

Is FB going to need a military force to protect their DCs?

---

With "Zuck Bucks" (I still am not sure if true)

This makes this ultimate "company store"

Tokens?

So how get?

How EARN? (What service on FB GENERATES '$ZB'?)

How spend?

WHAT GET? (NFTs?, Goods? Services?)?

The entire fucking model of EVERYTHING FB DOES is to MAP SENTIMENT!

Sentiment is the tie btwn INTENT and SENTIMENTAL VALUE

The idea is to map interest with emotional drivers which make someone buy (spend resources their time and effort went into building up a store-of)...

---

So map out your emotinal response over N topics and forums.. Eval your documented Online comments, NLP the fuck out of that, see what your demos are and build this profile to you....

THEN THEN THEN THEN

Offer an "earnable" (i.e. Grindable by farms and bots alike) -- "Zuck Buck" which is a TOKEN (etymology that fucking word for yourself)

of value...

Meaning, zero INTRINSIC value, Zero accountability (managed by a central Zuck Bank) And the vaule both determined AND available to you via not INTRINSIC CONTROL, nor VALUE.

---

FB Bots Galore.

Re: Real World Recommendation System

#35

Post by notafraudster »

imilk wrote: post #31
rexreed wrote: post #10

Isn't this obvious list-building promotion for a company (Fennel) that sells recommendation systems? "Fennel AI: Building and deploying real world recommendation systems in production Launched 18 hours ago" Caveat reader.

I'll never understand why people think this is a valid criticism of an article, rather than pointing out an issue they have with the actual content of the article. There's nothing inherently wrong with a company sharing info about the space they operate in. In fact, it should be encouraged as long as what they share is useful.

It's a short-hand for the treatment of the subject being pretty shallow and non-descript, which seems to apply to this article exactly. I read this and didn't learn anything.

Re: Real World Recommendation System

#36

Post by greatpostman »

I’ve built one of these at FAANG. Generally the different parts of the system are completely separate teams that interact through apis and ingest systems. Usually there’s a mix of online and offline calculations, where features are stored in a nosqldb and some simple model runs in a tomcat server at inference time, or the offline result is just retrieved. Almost everything is precomputed.

We had an api layer where another team runs inference on their model as new user data comes in, then streams it to our api which inboards the data.

On top of this, you have extensive A/B testing systems

Re: Real World Recommendation System

#37

Post by nixpulvis »

These steps read to me like: first we filter, then we filter, then we filter; all of this being done based on some various orders of the data.

The devil's in the details, which are surely domain specific and hopefully not too morally questionable.

Re: Real World Recommendation System

#39

Post by KaiserPro »

ocrow wrote: post #30

These recommendation systems take control away from individuals over what content they see and replace that choice with black box algorithms that don't explain why you are seeing the content that you are or what other content was excluded. All of the companies who have deployed these content selection algorithms could have also given you manual choice over the content that you see, but chose instead to let the algori…

Did you know how a news paper used to choose what articles it wanted to run?

Do you know how a TV channel decides to schedule stories?

Humans, its all humans. Looking at the metrics, and steering stuff that feeds that metric.

Content filters are dumb and easy to understand. seriously, open up a fresh account at FB, instagram, twitter or tiktok.

First it'll try and get a list of people you already know. Don't give it that.

Then it'll give you a bunch of super popular but click baity influencers to follow. why? because they are the things that drive attention.

if you follow those defaults, you'll get a view of whats shallow and popular: spam, tits, dicks and money.

If you find a subject leader, for example a independent tool maker, cook, pattern maker, builder, then most of your feed will be full of those subjects, save for about 10% random shit thats there to expand your subject range (mostly tits, dicks, spam or money)

What you'll see is stuff related to what you like and stare at.

And thats the problem, they are dumb mirrors. Thats why you don't let kids play with them. Thats why you don't let people with eating disorders go on them, thats why mental health needs to be more accessible, because some times holding up a mirror to your dark desires is corrosive.

Could filter designers do more? fuck yeah, be we also have to be aware that filters are a great whipping boy for other more powerful things.

Re: Real World Recommendation System

#40

Post by lmc »

hallqv wrote: post #22

Anyone have recommendations (no pun) for more in depth resources on the subject (large scale recommendation systems)?

Much of the field seems to be fixated on throwing massive compute resources at models with results that can neither be evaluated nor reproduced.

"the Recommender Systems research community is facing a crisis where a significant number of papers present results that contribute little to collective knowledge […] often because the research lacks the […] evaluation to be properly judged and, hence, to provide meaningful contributions"

https://doi.org/10.1145%2F2532508.2532513

More here... https://en.wikipedia.org/wiki/Recommender_system#Reproducibi...

Post reply on HN