Large language models can improve recommendation quality by reasoning explicitly over user history and candidate items - for example, extracting a user's preferences or explaining why one item fits better than another - rather than mapping history directly to a ranked list. This reasoning, however, is expensive to repeat on every ranking request and, once produced, is typically consumed once and discarded, leaving it neither reusable across future requests nor easy to inspect or correct as user tastes drift. Our insight is that reasoning does not need to be regenerated at every call if it can instead be compressed once into a compact, structured memory that a lightweight model retrieves from. We propose rEDMRec, which distills a teacher LLM's reasoning into four typed, editable experience channels - long-term preference, short-term context, item-perception, and counterfactual hard-negative comparisons - maintained by an LLM memory controller that performs Add/Delete/Modify/Keep operations and refines entries via K-agent debate. A lightweight student LLM then ranks candidates purely by retrieving from this memory, without invoking the teacher again, decoupling online inference cost from reasoning depth. Across ML-1M, Amazon Beauty, and Steam and ten student backbones, rEDMRec improves HR@1 over zero-shot, few-shot, and RAG on every backbone, and over GraphRAG on most backbones, with Impv up to 13.3% vs. the second-best baseline on ML-1M. Channel ablations show that short-term context is the only channel that helps consistently across capacity tiers, whereas long-term, item-perception, and counterfactual contributions are capacity-dependent (and can reverse on the strongest students); debate-based memory optimization lowers bank duplication by 7.4 percentage points while raising downstream HR@1 by up to +0.029 over six optimization epochs.
Model editing keeps large language models (LLMs) up to date without retraining, but temporal facts expose a limitation of the prevailing locate-and-edit paradigm: an update is not always a replacement. When a fact changes, the new answer should become current while the old answer may remain correct in historical time contexts. Building on this insight, we use causal tracing to show that LLMs already support this distinction via a two-stage internal computation: early MLP layers retrieve a time-agnostic subject representation, and later layers modulate it with temporal context to yield the time-correct answer. Motivated by this finding, we introduce PRISM Edit, which optimizes a single polysemous representation across temporal contexts and leverages the model's inherent modulation pathway to route it to temporally correct predictions without requiring any architectural modification. We evaluate on TimeConflict, a newly introduced temporal editing benchmark, and on temporally augmented CounterFact. PRISM Edit improves multiple core metrics over the best baseline, most notably +23.3 Temporal Consistency (TC) and +33.7 Current Relative-time Score (CRS) on LLaMA-3, while being more than 2x faster. Code and data are publicly available at https://github.com/CheerCHuang/PRISM-Edit.
Chen Huang, Qirui Zheng, Ruiqin Zheng et al.· 1 citation
Large language models (LLMs) can summarize heterogeneous user evidence in natural language, but current LLM recommenders often collapse enduring preferences, transient intent, and exposure-induced behavior into one profile. This makes recommendation vulnerable to feedback loops: repeated exposure is mistaken for preference, immediate clicks dominate delayed satisfaction, and fluent explanations need not reflect the ranking decision. We propose our method, a model-agnostic framework for long-horizon recommendation. Our method uses a frozen multimodal language model to convert item content and feedback into evidence-grounded semantic atoms, then maintains separate short-term, long-term, and exposure memories. Propensity-weighted updates reduce policy-induced exposure bias, while a conservative offline critic reranks candidates for delayed satisfaction under a behavior-support constraint. Explanations use only influential evidence atoms and are checked by counterfactual deletion. We provide an identification result and evaluate the framework in e-commerce-like, news-like, and short-video-like environments. Across ten seeds, our method improves discounted long-term value over the strongest alternative by 6.1%, 7.6%, and 6.7%, respectively. Twenty-seed paired ablations show significant value drops after removing propensity correction (0.739 +/- 0.191) or conservative support regularization (0.523 +/- 0.234). A frozen instruction language model also more than doubles semantic-atom NDCG over TF-IDF on a held-out paraphrase benchmark.
Large language models (LLMs) are increasingly applied to recommendation, retrieval, and reasoning, yet deploying a single end-to-end model that can jointly support these behaviors over large, heterogeneous catalogs remains challenging. Such systems must generate unambiguous references to real items, handle multiple entity types, and operate under strict latency and reliability constraints requirements that are difficult to satisfy with text-only generation. While tool-augmented recommender systems address parts of this problem, they introduce orchestration complexity and limit end-to-end optimization. We view this setting as an instance of a broader research problem: how to adapt LLMs to reason jointly over multiple-domain entities, user behavior, and language in a fully self-contained manner. To this end, we introduce NEO, a framework that adapts a pre-trained decoder-only LLM into a tool-free, catalog-grounded generator. NEO represents items using semantic identifiers (SIDs) and trains a single model to interleave natural language and typed item identifiers within a shared sequence. Natural-language prompts control the task, target entity type, and output format (IDs, text, or mixed), while constrained decoding guarantees catalog-valid item generation without restricting free-form text. We refer to this instruction-conditioned controllability as language-steerability. Inspired by multimodal alignment, we treat SIDs as a distinct modality and study design choices for integrating discrete entity representations into LLMs via staged alignment and instruction tuning. We evaluate NEO at scale on a real-world catalog of over 10M items across multiple media types and discovery tasks, including recommendation, search, and user understanding. In offline experiments, NEO consistently outperforms strong task-specific baselines and exhibits positive cross-task transfer, demonstrating a practical path toward consolidating large-scale discovery capabilities into a single language-steerable generative model.
Marco De Nadai, Edoardo D'Amico, Maksym Lefarov et al.· Proceedings of the 32nd ACM...· 0 citations
For decades, search and recommendation systems have been optimized as distinct components within large-scale discovery platforms. The rise of generative AI is beginning to blur this boundary. At Spotify, we are exploring how large language models can evolve from tools that retrieve content into systems that reason over users, catalogs, and intent, while remaining steerable through natural language and user interaction. This talk presents lessons from deploying and studying generative retrieval and recommendation systems across Spotify's content ecosystem. I will describe how semantic identifiers enable language models to operate directly over large, heterogeneous catalogs, allowing search, recommendation, retrieval, explanation, and user understanding to be expressed within a common generative framework. I will discuss recent work on production-scale podcast discovery, language-steerable recommendation, and the NEO framework for unifying search, recommendation, and reasoning across multiple content types. These systems demonstrate how grounding language models in catalog entities and user behavior can improve discovery while preserving the flexibility of natural-language interaction. More broadly, they suggest a path toward discovery systems in which retrieval, recommendation, and reasoning are no longer separate stages, but capabilities of a shared generative model. Beyond model frameworks, I will discuss the emerging challenges of alignment and evaluation in discovery systems. Unlike traditional retrieval problems, generative recommendation often has many valid answers. I will present approaches for learning from large-scale behavioral signals, preference-aware optimization, and profile-aware LLM-as-a-judge evaluation, along with lessons from online experimentation at Spotify. These experiences suggest that future discovery systems will require new forms of personalization, controllability, and evaluation that extend beyond conventional ranking metrics. I will conclude with a research agenda for generative discovery systems, including language-steerable interfaces, unified retrieval-and-reasoning models, preference-aligned generation, and evaluation frameworks designed to measure user-specific relevance at scale. As search, recommendation, and conversational AI continue to converge, these directions point toward a new generation of discovery systems that can understand intent, reason over large catalogs, and help users navigate increasingly complex information spaces.
Paul N. Bennett· Proceedings of the 32nd ACM...· 0 citations
Agentic recommender systems ground each decision of a large language model (LLM) in a persistent memory of the user, and in existing agents that memory is text: a narrative written and maintained by further LLM calls. Text limits this memory in two ways. It is updated one rewrite at a time, so exploiting the full interaction history is prohibitively expensive; and collaborative evidence, graded similarity over an entire catalog, does not survive translation into sentences. We propose CoVeMem (Collaborative Vector Memory), which vectorizes the collaborative core of the agent's memory. Frozen LightGCN user and item states form the memory bank; at each decision, the candidate set itself retrieves the most relevant historical states, which enter the LLM's context as soft tokens alongside a light textual profile. Contrastive alignment to item-semantic anchors, followed by listwise co-training with masked candidates, teaches the model to read these states and to rank through them; a pointwise yes/no readout scores each candidate. Across four instruction-grounded recommendation benchmarks, CoVeMem matches or exceeds the strongest collaborative text-memory agent on 19 of 20 metric cells while requiring zero additional LLM calls for memory maintenance beyond the shared static profile, against per-interaction calls for text memory. The memory now takes gradients: the full interaction history, out of reach for text, becomes available as training data for what the agent remembers and for how it reads what it remembers.
Hanting Chen, Xing Tang, Lingyi Li et al.· 0 citations
RecGPT-V3 is presented, a stateful, hybrid-modal recommender that reasons over natural language for open-world knowledge and Semantic IDs (SIDs) for concrete item grounding and achieves consistent gains in large-scale online A/B tests.
A new method, called CW-Net, translates the reasoning process of an autonomous vehicle’s AI system into understandable concepts that explain its behavior.
