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

Improving recommendation systems and search in the age of LLMs

eugeneyan.com

Post reply on HN
Print view View original

91–98 of 98 posts

Re: Improving recommendation systems and search in the age of LLMs

#91

Post by RamblingCTO »

x1xx wrote: post #34

> Spotify saw a 9% increase in exploratory intent queries, a 30% rise in maximum query length per user, and a 10% increase in average query length—this suggests the query recommendation updates helped users express more complex intents To me it's not clear that it should be interpreted as an improvement: what I read in this summary is that users had to search more and to enter longer queries to get to what they neede…

100%. I've switched over to Apple Music because you really feel that they are pushing public playlists. Search terms maximize their playlists vs mine. I now had to go to my library to find my playlists because they wouldn't even show up.

Re: Improving recommendation systems and search in the age of LLMs

#92

Post by porridgeraisin »

x1xx wrote: post #34

> Spotify saw a 9% increase in exploratory intent queries, a 30% rise in maximum query length per user, and a 10% increase in average query length—this suggests the query recommendation updates helped users express more complex intents To me it's not clear that it should be interpreted as an improvement: what I read in this summary is that users had to search more and to enter longer queries to get to what they neede…

A conclusion true to the concerned executive's MBA.

Re: Improving recommendation systems and search in the age of LLMs

#93

Post by IanCal »

dmbche wrote: post #27
nine_k wrote: post #5

Earlier quoted context omitted.

> you can just search your files using your prefered file explorer This only work if you remember specific substrings. An LLM (or some other language model) can summarize and interpolate. It can be asked to find that file that mentions a transaction for buying candy, and it has a fair chance to find it, even if none of the words "transaction", "buying" or "candy" are present in the file, e.g. it says "shelled out $17…

But isn't that candy example non-sensical? In what situation do you need some information without any of the context(or without knowing any of the context)? i really believe that this is not an actual problem in need of solving, but instead creating a tool (personal ai assistant) and trying to find a usecase Edit0: note to self, rambling - assuming there exist valuable information that one needa to access in their fi…

> But isn't that candy example non-sensical? In what situation do you need some information without any of the context(or without knowing any of the context)?

I know the context and the content but not the specific substrings in an email I received several years ago.

Here's one of the first things that gemini in gmail actually helped with. I wanted to check when I bought a car seat for my kids, which one it was and how much it cost.

So I knew the rough time it was when I bought it, I know it's a receipt I'm looking for, it's for a child seat, and roughly when. I know the context here.

What I struggled with was finding the exact text that would be in that. There are hundreds or more emails with invoice/receipt/order in. I didn't recall exactly who I bought it from, and there are large numbers of more advertising emails with kids seats in.

I couldn't easily find it, because the actual email I wanted did not say child seat in it. It had a brand and other information, but nothing in the text had a substring I was searching for. I might have found it with "booster seat" but I didn't think of that exact phrase at the time.

Instead I asked gemini to find it. That can then trawl through a bunch of emails and find things that mean but do not say child seat.

Re: Improving recommendation systems and search in the age of LLMs

#94

Post by dmbche »

IanCal wrote: post #93
dmbche wrote: post #27

Earlier quoted context omitted.

But isn't that candy example non-sensical? In what situation do you need some information without any of the context(or without knowing any of the context)? i really believe that this is not an actual problem in need of solving, but instead creating a tool (personal ai assistant) and trying to find a usecase Edit0: note to self, rambling - assuming there exist valuable information that one needa to access in their fi…

> But isn't that candy example non-sensical? In what situation do you need some information without any of the context(or without knowing any of the context)? I know the context and the content but not the specific substrings in an email I received several years ago. Here's one of the first things that gemini in gmail actually helped with. I wanted to check when I bought a car seat for my kids, which one it was and h…

Makes total sense, I'm not entirely sure why but I had assumed we were talking about an AI assistant being device specific, which I understood as "based on my offline data and files" - I'm saying I'm not sure why because the parent comment is specifically mentionning the cloud. Anyhoot.

Enjoy your day.

Re: Improving recommendation systems and search in the age of LLMs

#95

Post by mhuffman »

anon8764352 wrote: post #89

@7d7n Eugene / others experienced in recommendation systems: for someone who is new to recommendation systems and uses variants of collaborative filtering for recommendations, what non-LLM approach would you suggest to start looking into? The cheaper the compute (ideally without using GPUs in the first place) the better, while also maximizing the performance of the system :)

IMHO it depends on the types of things you are recommending. If you have a good way of accurately and specifically textually classifying items it is hard to beat the performance of good old-fashioned embeddings and vector search/ANN. There are plenty of embeddings that do not need GPU like the newer LLM-based ones all crave. Word2Vec, GloVe, and FastText are all high-performance and you wouldn't need GPUs. There are plenty of vector-search libraries that are high-performance and predate the vector-db popularity of late, so also would not depend on GPUs to be high-performance. Most are memory-hungry however, so something to keep in mind. That performance, especially with the embeddings, will come at the cost of loss of some context. No free lunch.

Re: Improving recommendation systems and search in the age of LLMs

#96

Post by singron »

x1xx wrote: post #34

> Spotify saw a 9% increase in exploratory intent queries, a 30% rise in maximum query length per user, and a 10% increase in average query length—this suggests the query recommendation updates helped users express more complex intents To me it's not clear that it should be interpreted as an improvement: what I read in this summary is that users had to search more and to enter longer queries to get to what they neede…

This is a hard problem. We had similar issues evaluating success with real users. In the literature, there is "abandonment" (i.e. I couldn't find what I wanted and gave up) and "positive abandonment" (I got what I wanted from the SERP and didn't click on anything). A flurry of requests might be a series of positive abandonment, a natural fruitful process of refining the request, or rage querying where the user repeatedly fails to correct a model that is incapable of understanding the query. It's especially devious if they rage query for a while before switching to an easier task and succeeding (e.g. clicking a result) since you might count that whole interaction as positive when it was really quite negative.

Re: Improving recommendation systems and search in the age of LLMs

#97

Post by wildrhythms »

whatever1 wrote: post #2

Why we don’t have an LLM based search tool for our pc / smartphones? Specially for the smartphones all of your data is on the cloud anyway, instead of just scraping it for advertising and the FBI they could also do something useful for the user?

Pixels already have the Screenshots app that indexes screenshots and makes them searchable. My assumption is the context window size is still too small for all of your data to go into it.

Re: Improving recommendation systems and search in the age of LLMs

#98

Post by appleorchard46 »

Melatonic wrote: post #62
appleorchard46 wrote: post #52

Earlier quoted context omitted.

I liked when you could make a playlist radio and do that manually. That's been removed now of course.

On desktop I believe you can still take any of your playlists and tell it to generate a "similar" playlist. Works really well.

Oops, I'm four days late but - no, it appears to have been removed there too. If I'm missing it please let me know.
Post reply on HN