[theknight]

@sadikkapadia - Any idea why Recommendation Engine as a Service has not picked up? I realize that building a use-case specific recommendation engine is unique.

However, I am wondering is there a recommendation engine as a service, which is similar to algolia available/possible.

I see only 2 players - yusp and recombee.

I'd appreciate any thought you have on this.

[bbulkow]

I worked for a recommendation engine as a service company called aggregate knowledge in 2007 when they got their second round of funding, 25m from kleiner. It was a remarkably lousy business, and they didn't do well.

There was really nothing wrong with the concept. A little JavaScript on the page, a bunch of back end magic, the ability to use a larger data pool because you are collecting from multiple sites.... But people wouldn't pay, and the engagement of recommendations was never that good.

The secret we learned was, after a bunch of math and research, was that 'best in this category and 3 closest adjacent categories works so well in retail that a naive algo did very well.

Better than the fancy math, which once associated the Koran with the sports illustrated swim suit edition ( and vice versa ).

There is a massive data sufficiency problem, and no company with the real data would go into this low end business. Small companies can't get enough data to be relevent.

The world is 10 years later, ml is better understood, so it might all be different now.... Pm me if you are interested in further detail

[sebst]

> Any idea why Recommendation Engine as a Service has not picked up?

These kind of services aren't so much exposed to the public and likely don't start at <100 bucks a month, which could be why those services are not that visible. However, there are some e-commerce services going in that direction, such like AgilOne...

[sadikkapadia1]

It is hard to sell technology to companies when they have their own teams (often using free libraries). Embedded teams are always experts and will often discredit better technology. Often the only method of testing is A/B. These can easily be manipulated. For instance at Netflix (ignoring more blatant practices), P-hacking (run thousands of simulations and report ones that worked), and HARKing (come up with a hypothesis after the results are known) are rampant. That is part of the reason the recommender has been degrading over the years.

[prades]

> That is part of the reason the recommender has been degrading over the years.

Netflix is clearly promoting its original shows. Do you think your system is still in use now that they've moved to thumbs rating and percent match score?

[sadikkapadia1]

That is what I have been told. It is however clearly messed up.

[prades]

I feel for you. It must be a huge disappointment to see what they've done to your work. Netflix recommendation system used to be very good a few years ago.

[prades]

https://news.ycombinator.com/item?id=15607383

Amazing story! I don't know if it's true or if you're delusional, but I'm inclined to believe you as it would explain why Netflix recommendation system went downhill instead of improving.