That's very interesting! What led you to this topic? I'm writing my master thesis about manipulative recommender systems and for it I need to make some simulation where I demonstrate the RecSys-human interaction. Haven't yet figured out how to simulate a human user but this could be one way to do it
November 12, 2025 at 7:19 PM
That's very interesting! What led you to this topic? I'm writing my master thesis about manipulative recommender systems and for it I need to make some simulation where I demonstrate the RecSys-human interaction. Haven't yet figured out how to simulate a human user but this could be one way to do it
Try For You - a personalized feed based on your likes.
It finds people who liked the same posts as you, and shows you what else they've liked recently.
That's a simple and surprisingly effective recsys!
It finds people who liked the same posts as you, and shows you what else they've liked recently.
That's a simple and surprisingly effective recsys!
November 6, 2025 at 1:24 AM
Try For You - a personalized feed based on your likes.
It finds people who liked the same posts as you, and shows you what else they've liked recently.
That's a simple and surprisingly effective recsys!
It finds people who liked the same posts as you, and shows you what else they've liked recently.
That's a simple and surprisingly effective recsys!
Bluesky is weird. The discover feed is 10 posts from people I follow and then a video of a guy jerking off. Put that in your recsys and smoke it.
November 5, 2025 at 2:49 PM
Bluesky is weird. The discover feed is 10 posts from people I follow and then a video of a guy jerking off. Put that in your recsys and smoke it.
while feeds like For You do a pretty good job of emulating ML-based recsys algorithms by just leveraging the social graph and interactions, it'd be cool to use cheap and scalable methods like these to surface relevant posts outside interaction graphs to connect "islands" of people better
i am entirely too employed to maintain this but it seems like it would be fairly trivial to make a service that embeds Every Post by some cheap method like "inverse document frequency weighted average of LLM token embeddings"
would be decent feedstock for custom recsys stuff
would be decent feedstock for custom recsys stuff
November 5, 2025 at 9:59 AM
while feeds like For You do a pretty good job of emulating ML-based recsys algorithms by just leveraging the social graph and interactions, it'd be cool to use cheap and scalable methods like these to surface relevant posts outside interaction graphs to connect "islands" of people better
i am entirely too employed to maintain this but it seems like it would be fairly trivial to make a service that embeds Every Post by some cheap method like "inverse document frequency weighted average of LLM token embeddings"
would be decent feedstock for custom recsys stuff
would be decent feedstock for custom recsys stuff
November 5, 2025 at 9:49 AM
i am entirely too employed to maintain this but it seems like it would be fairly trivial to make a service that embeds Every Post by some cheap method like "inverse document frequency weighted average of LLM token embeddings"
would be decent feedstock for custom recsys stuff
would be decent feedstock for custom recsys stuff
📜ACM Transactions on Recommender Systems introduces a special issue on deep reinforcement learning for recommender systems, collecting cutting-edge research that addresses pressing challenges and showcases innovations. #recsys #deeplearning #reinforcementlearning
dl.acm.org/toc/tors/202...
dl.acm.org/toc/tors/202...
TORS: Vol 4, No 1
dl.acm.org
October 30, 2025 at 1:59 PM
📜ACM Transactions on Recommender Systems introduces a special issue on deep reinforcement learning for recommender systems, collecting cutting-edge research that addresses pressing challenges and showcases innovations. #recsys #deeplearning #reinforcementlearning
dl.acm.org/toc/tors/202...
dl.acm.org/toc/tors/202...
Lol at Bluesky's recsys
October 27, 2025 at 7:40 PM
Lol at Bluesky's recsys
They said AGI is around the corner and yet the most sophisticated recsys still show Sky a bunch of women's titties. 😆
October 26, 2025 at 4:05 AM
They said AGI is around the corner and yet the most sophisticated recsys still show Sky a bunch of women's titties. 😆
My congratulations to Savvina Daniil of CWI Amsterdam for winning the Women in RecSys Journal Paper of 2025 award!
https://irrj.org/announcement/view/300
https://irrj.org/announcement/view/300
RecSys Journal Paper of the Year award for IRRJ paper of Savvina Daniil et al.
| Information Retrieval Research
irrj.org
October 24, 2025 at 8:55 AM
My congratulations to Savvina Daniil of CWI Amsterdam for winning the Women in RecSys Journal Paper of 2025 award!
https://irrj.org/announcement/view/300
https://irrj.org/announcement/view/300
Congratulations to Savvina Daniil for winning the Women in RecSys Journal Paper of the Year award for her #irrj paper "On the challenges of studying bias in Recommender Systems"!
The award came with a free #recsys2025 registration and the opportunity to present the IRRJ paper at the main […]
The award came with a free #recsys2025 registration and the opportunity to present the IRRJ paper at the main […]
Original post on sigmoid.social
sigmoid.social
October 23, 2025 at 1:49 PM
Congratulations to Savvina Daniil for winning the Women in RecSys Journal Paper of the Year award for her #irrj paper "On the challenges of studying bias in Recommender Systems"!
The award came with a free #recsys2025 registration and the opportunity to present the IRRJ paper at the main […]
The award came with a free #recsys2025 registration and the opportunity to present the IRRJ paper at the main […]
Right i think it would work exactly like recsys where its some kind of hacky mess because anything else would be too hard to manage and test
October 20, 2025 at 8:40 PM
Right i think it would work exactly like recsys where its some kind of hacky mess because anything else would be too hard to manage and test
V funny to me that Spotify tried curated playlists and got backlash for appearance of pigeonholing artists by identity category (and was), but when SoundCloud does it it’s just the best RecSys you’ve ever heard
October 20, 2025 at 9:44 AM
V funny to me that Spotify tried curated playlists and got backlash for appearance of pigeonholing artists by identity category (and was), but when SoundCloud does it it’s just the best RecSys you’ve ever heard
What happens when you say:
“I want a horror -- comedy -- movie”? 🎥
That slip-of-the-tongue can confuse recommender systems.
Our INTERSPEECH 2025 paper shows some LLMs handle it better than others.
📄 mariateleki.github.io/pdf/HorrorCo...
#INTERSPEECH2025 #ConversationalAI #RecSys
“I want a horror -- comedy -- movie”? 🎥
That slip-of-the-tongue can confuse recommender systems.
Our INTERSPEECH 2025 paper shows some LLMs handle it better than others.
📄 mariateleki.github.io/pdf/HorrorCo...
#INTERSPEECH2025 #ConversationalAI #RecSys
October 18, 2025 at 6:15 PM
What happens when you say:
“I want a horror -- comedy -- movie”? 🎥
That slip-of-the-tongue can confuse recommender systems.
Our INTERSPEECH 2025 paper shows some LLMs handle it better than others.
📄 mariateleki.github.io/pdf/HorrorCo...
#INTERSPEECH2025 #ConversationalAI #RecSys
“I want a horror -- comedy -- movie”? 🎥
That slip-of-the-tongue can confuse recommender systems.
Our INTERSPEECH 2025 paper shows some LLMs handle it better than others.
📄 mariateleki.github.io/pdf/HorrorCo...
#INTERSPEECH2025 #ConversationalAI #RecSys
agree on the latter point, cf bsky.app/profile/tkuk...
on the former - it's costly if one assumes grok invocation per page load. but IMO scaling a grok-twitter-recsys is amenable to good system dimensioning (e.g. model cascade + periodic batching)
from that perspective, doesn't seem preposterous 🤷
on the former - it's costly if one assumes grok invocation per page load. but IMO scaling a grok-twitter-recsys is amenable to good system dimensioning (e.g. model cascade + periodic batching)
from that perspective, doesn't seem preposterous 🤷
agree re/feeds
what you'd normally do for a closed-platform is (1) scrape posts, (2) write-your-own ranking/clf. algo to get better content (tags are useful but easily co-opted, the platform's recsys is necessarily deficient due to partial obs).
_native_ support for scaling (2) I find exciting
what you'd normally do for a closed-platform is (1) scrape posts, (2) write-your-own ranking/clf. algo to get better content (tags are useful but easily co-opted, the platform's recsys is necessarily deficient due to partial obs).
_native_ support for scaling (2) I find exciting
October 18, 2025 at 4:45 PM
agree on the latter point, cf bsky.app/profile/tkuk...
on the former - it's costly if one assumes grok invocation per page load. but IMO scaling a grok-twitter-recsys is amenable to good system dimensioning (e.g. model cascade + periodic batching)
from that perspective, doesn't seem preposterous 🤷
on the former - it's costly if one assumes grok invocation per page load. but IMO scaling a grok-twitter-recsys is amenable to good system dimensioning (e.g. model cascade + periodic batching)
from that perspective, doesn't seem preposterous 🤷
@swyx https://x.com/swyx/status/1979291014990074030 #x-swyx
Quote: https://x.com/aiDotEngineer/status/1945553551239123251
now seems a particularly good time to remind you why I asked @eugeneyan to curate an entire track of LLMs in Recsys talks at @aidotengine...
Quote: https://x.com/aiDotEngineer/status/1945553551239123251
now seems a particularly good time to remind you why I asked @eugeneyan to curate an entire track of LLMs in Recsys talks at @aidotengine...
October 17, 2025 at 9:15 PM
@swyx https://x.com/swyx/status/1979291014990074030 #x-swyx
Quote: https://x.com/aiDotEngineer/status/1945553551239123251
now seems a particularly good time to remind you why I asked @eugeneyan to curate an entire track of LLMs in Recsys talks at @aidotengine...
Quote: https://x.com/aiDotEngineer/status/1945553551239123251
now seems a particularly good time to remind you why I asked @eugeneyan to curate an entire track of LLMs in Recsys talks at @aidotengine...
Leonie Winter: Dataset Pruning in RecSys and ML: Best Practice or Mal-Practice? https://arxiv.org/abs/2510.14704 https://arxiv.org/pdf/2510.14704 https://arxiv.org/html/2510.14704
October 17, 2025 at 6:32 AM
Leonie Winter: Dataset Pruning in RecSys and ML: Best Practice or Mal-Practice? https://arxiv.org/abs/2510.14704 https://arxiv.org/pdf/2510.14704 https://arxiv.org/html/2510.14704
Leonie Winter
Dataset Pruning in RecSys and ML: Best Practice or Mal-Practice?
https://arxiv.org/abs/2510.14704
Dataset Pruning in RecSys and ML: Best Practice or Mal-Practice?
https://arxiv.org/abs/2510.14704
October 17, 2025 at 4:07 AM
Leonie Winter
Dataset Pruning in RecSys and ML: Best Practice or Mal-Practice?
https://arxiv.org/abs/2510.14704
Dataset Pruning in RecSys and ML: Best Practice or Mal-Practice?
https://arxiv.org/abs/2510.14704
📢 We're also now on LinkedIn!
Follow the Glasgow Information Retrieval Group for updates on IR research, @irglasgow.bsky.social activities, events, and collaborations:
🔗 linkedin.com/company/glasgow-information-retrieval-group
#InformationRetrieval #IR #AI #recsys #RecommenderSystems #Glasgow
Follow the Glasgow Information Retrieval Group for updates on IR research, @irglasgow.bsky.social activities, events, and collaborations:
🔗 linkedin.com/company/glasgow-information-retrieval-group
#InformationRetrieval #IR #AI #recsys #RecommenderSystems #Glasgow
Glasgow Information Retrieval Group | LinkedIn
Glasgow Information Retrieval Group | 298 followers on LinkedIn. Founded in 1986, the Glasgow IR Group has been at the forefront of Research & Development in search and recommendation.
linkedin.com
October 16, 2025 at 3:57 PM
📢 We're also now on LinkedIn!
Follow the Glasgow Information Retrieval Group for updates on IR research, @irglasgow.bsky.social activities, events, and collaborations:
🔗 linkedin.com/company/glasgow-information-retrieval-group
#InformationRetrieval #IR #AI #recsys #RecommenderSystems #Glasgow
Follow the Glasgow Information Retrieval Group for updates on IR research, @irglasgow.bsky.social activities, events, and collaborations:
🔗 linkedin.com/company/glasgow-information-retrieval-group
#InformationRetrieval #IR #AI #recsys #RecommenderSystems #Glasgow
A big reason social media took over from feed readers as a content discovery mechanism was that people will choose the lower effort of recsys vs curating their own feeds. I suspect access to a recsys feed that ranks on quality vs just various measures of popularity will succeed if sufficiently lazy.
October 14, 2025 at 4:43 PM
A big reason social media took over from feed readers as a content discovery mechanism was that people will choose the lower effort of recsys vs curating their own feeds. I suspect access to a recsys feed that ranks on quality vs just various measures of popularity will succeed if sufficiently lazy.
Yeah I agree. I still believe in algorithmic choice, but the history of this platform has made clear that *at least one* of the choices must use personalized recsys stuff. Trying to cut corners on that part was very nearly fatal
October 14, 2025 at 11:25 AM
Yeah I agree. I still believe in algorithmic choice, but the history of this platform has made clear that *at least one* of the choices must use personalized recsys stuff. Trying to cut corners on that part was very nearly fatal
🎯 Jak dobře chápou doporučovací systémy své uživatele?
Doporučovací systémy nám denně radí např. co si pustit. Snaží se nabídnout to, co je pro nás nejrelevantnější. Ale stačí to?
O tom, nakolik si systémy a uživatelé rozumí, mluvil Ladislav Peška z KSI MFF UK na konferenci RecSys 2025.
Doporučovací systémy nám denně radí např. co si pustit. Snaží se nabídnout to, co je pro nás nejrelevantnější. Ale stačí to?
O tom, nakolik si systémy a uživatelé rozumí, mluvil Ladislav Peška z KSI MFF UK na konferenci RecSys 2025.
October 13, 2025 at 7:26 AM
🎯 Jak dobře chápou doporučovací systémy své uživatele?
Doporučovací systémy nám denně radí např. co si pustit. Snaží se nabídnout to, co je pro nás nejrelevantnější. Ale stačí to?
O tom, nakolik si systémy a uživatelé rozumí, mluvil Ladislav Peška z KSI MFF UK na konferenci RecSys 2025.
Doporučovací systémy nám denně radí např. co si pustit. Snaží se nabídnout to, co je pro nás nejrelevantnější. Ale stačí to?
O tom, nakolik si systémy a uživatelé rozumí, mluvil Ladislav Peška z KSI MFF UK na konferenci RecSys 2025.
a recsys that maximizes time on app is kind of a hostile thing to build when you think about it. it also seems like the kind of thing that would emerge from extended use + algorithmic diversity + directed search
October 11, 2025 at 11:21 PM
a recsys that maximizes time on app is kind of a hostile thing to build when you think about it. it also seems like the kind of thing that would emerge from extended use + algorithmic diversity + directed search
(I don’t remember if this piece does it, but some in RecSys have used the terms “justification” and “explanation” to distinguish between post-hoc justifications of why a recommendation is good vs. internally-faithful explanations of why it was actually recommended. IMO that’s very useful.)
October 8, 2025 at 8:47 PM
(I don’t remember if this piece does it, but some in RecSys have used the terms “justification” and “explanation” to distinguish between post-hoc justifications of why a recommendation is good vs. internally-faithful explanations of why it was actually recommended. IMO that’s very useful.)
I guess I do maybe have a paper lol. IMO Tintarev’s work does a good job of laying out the relationship of purpose and explanation in recommender systems, e.g. doi.org/10.1007/978-....
(IMSaltyO, some XAI has been rediscovering concepts the RecSys explanation people were figuring out a while ago.)
(IMSaltyO, some XAI has been rediscovering concepts the RecSys explanation people were figuring out a while ago.)
Explaining Recommendations: Design and Evaluation
This chapter gives an overview of the area of explanations in recommender systems. We approach the literature from the angle of evaluation: that is, we are interested in what makes an explanation “good”. The chapter starts by describing how explanations...
doi.org
October 8, 2025 at 8:45 PM
I guess I do maybe have a paper lol. IMO Tintarev’s work does a good job of laying out the relationship of purpose and explanation in recommender systems, e.g. doi.org/10.1007/978-....
(IMSaltyO, some XAI has been rediscovering concepts the RecSys explanation people were figuring out a while ago.)
(IMSaltyO, some XAI has been rediscovering concepts the RecSys explanation people were figuring out a while ago.)