Cold-Starts in Generative Recommendation: A Reproducibility Study
arXiv:2603.29845v1 Announce Type: new Abstract: Cold-start recommendation remains a central challenge in dynamic, open-world platforms, requiring models to recommend for newly registered users (user cold-start) and to recommend newly introduced items to existing users (item cold-start) under sparse or missing interaction signals. Recent generative recommenders built on pre-trained language models (PLMs) are often expected to mitigate cold-start by using item semantic information (e.g., titles and descriptions) and test-time conditioning on limited user context. However, cold-start is rarely treated as a primary evaluation setting in existing studies, and reported gains are difficult to interpret because key design choices, such as model scale, identifier design, and training strategy, are
View PDF HTML (experimental)
Abstract:Cold-start recommendation remains a central challenge in dynamic, open-world platforms, requiring models to recommend for newly registered users (user cold-start) and to recommend newly introduced items to existing users (item cold-start) under sparse or missing interaction signals. Recent generative recommenders built on pre-trained language models (PLMs) are often expected to mitigate cold-start by using item semantic information (e.g., titles and descriptions) and test-time conditioning on limited user context. However, cold-start is rarely treated as a primary evaluation setting in existing studies, and reported gains are difficult to interpret because key design choices, such as model scale, identifier design, and training strategy, are frequently changed together. In this work, we present a systematic reproducibility study of generative recommendation under a unified suite of cold-start protocols.
Subjects:
Information Retrieval (cs.IR)
Cite as: arXiv:2603.29845 [cs.IR]
(or arXiv:2603.29845v1 [cs.IR] for this version)
https://doi.org/10.48550/arXiv.2603.29845
arXiv-issued DOI via DataCite (pending registration)
Submission history
From: Zhen Zhang [view email] [v1] Tue, 31 Mar 2026 15:06:31 UTC (161 KB)
Sign in to highlight and annotate this article
Conversation starters
Daily AI Digest
Get the top 5 AI stories delivered to your inbox every morning.
More about
modellanguage modeltraining
A Very Fine Untuning
How fine-tuning made my chatbot worse (and broke my RAG pipeline) I spent weeks trying to improve my personal chatbot, Virtual Alexandra , with fine-tuning. Instead I got increased hallucination rate and broken retrieval in my RAG system. Yes, this is a story about a failed attempt, not a successful one. My husband and I called fine tuning results “Drunk Alexandra” — incoherent answers that were initially funny, but quickly became annoying. After weeks of experiments, I reached a simple conclusion: for this particular project, a small chatbot that answers questions based on my writing and instructions, fine tuning was not a good option. It was not just unnecessary, it actively degraded the experience and didn’t justify the extra time, cost, or complexity compared to the prompt + RAG system
Google's TurboQuant saves memory, but won't save us from DRAM-pricing hell
<h4>Chocolate Factory’s compression tech clears the way to cheaper AI inference, not more affordable memory</h4> <p>When Google unveiled <a target="_blank" rel="nofollow" href="https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/">TurboQuant</a>, an AI data compression technology that promises to slash the amount of memory required to serve models, many hoped it would help with a memory shortage that has seen prices triple since last year. Not so much.…</p>
How We're Approaching a County-Level Education Data System Engagement
<p>When Los Angeles County needs to evaluate whether a multi-agency data system serving foster youth should be modernized or replaced, the work sits at the intersection of technology, policy, and people. That's exactly where we operate.</p> <h2> The Opportunity </h2> <p>The LA County Office of Child, Youth, and Family Well-Being is looking for a consulting team to analyze the Education Passport System (EPS), a shared data platform that connects 80+ school districts with the Department of Children and Family Services and the Probation Department. The system exists to ensure that when a foster youth moves between placements, their education records follow them.</p> <p>The question on the table: does the current system meet the needs of all stakeholders, or is it time to move to something new
Knowledge Map
Connected Articles — Knowledge Graph
This article is connected to other articles through shared AI topics and tags.
More in Models
A Very Fine Untuning
How fine-tuning made my chatbot worse (and broke my RAG pipeline) I spent weeks trying to improve my personal chatbot, Virtual Alexandra , with fine-tuning. Instead I got increased hallucination rate and broken retrieval in my RAG system. Yes, this is a story about a failed attempt, not a successful one. My husband and I called fine tuning results “Drunk Alexandra” — incoherent answers that were initially funny, but quickly became annoying. After weeks of experiments, I reached a simple conclusion: for this particular project, a small chatbot that answers questions based on my writing and instructions, fine tuning was not a good option. It was not just unnecessary, it actively degraded the experience and didn’t justify the extra time, cost, or complexity compared to the prompt + RAG system
Google's TurboQuant saves memory, but won't save us from DRAM-pricing hell
<h4>Chocolate Factory’s compression tech clears the way to cheaper AI inference, not more affordable memory</h4> <p>When Google unveiled <a target="_blank" rel="nofollow" href="https://research.google/blog/turboquant-redefining-ai-efficiency-with-extreme-compression/">TurboQuant</a>, an AI data compression technology that promises to slash the amount of memory required to serve models, many hoped it would help with a memory shortage that has seen prices triple since last year. Not so much.…</p>
Introducing The Screwtape Ladders
The time has come for me to find a new home for my writings. Like many an author before me, I've enjoyed improving my craft and getting feedback on my essays here. LessWrong is a good incubator for honing one's skills in that arena. There's a chance to get your point out in front of a really broad audience of really smart people. There's been some cool moments. My oldest visible post, Write A Thousand Roads to Rome , got cited in a discussion with Eliezer Yudkowsky once. I keep seeing people bring up Loudly Give Up, Don't Quietly Fade as a motivator for speaking out. Sometimes it's really cool people working on awesome projects, and I feel a flash of sadness at 'aww, it's not going to happen' and also a bit of cool 'whoa, they remember that post?' You've all also let me get away with a lot
Anthropic Executive Sees Cowork Agent as Bigger Than Claude Code - Bloomberg.com
<a href="https://news.google.com/rss/articles/CBMitgFBVV95cUxOM0VfSzdRYUNpT21XMlVuNXhsVEY4TUFxM3UzWUJDOEhFcUtJQnhTbjY2VjBXOUw1d1ZOUDRKeHVKMzkta3pFVWRWSGNoQkp3aWVndlRBQlpVUGxVN0ZnQW80OUZnYWN6RlhJWHRjT0V4RVhPcGhxMmE3b3oyVDlUV2RLY0g2NEx4M1dfMXhvTlhPTW50eFR1cEhxcHB3SXpURnRtbDZtZHp6bGQ2Z09IMjZBODBjdw?oc=5" target="_blank">Anthropic Executive Sees Cowork Agent as Bigger Than Claude Code</a> <font color="#6f6f6f">Bloomberg.com</font>
Discussion
Sign in to join the discussion
No comments yet — be the first to share your thoughts!