The paper introduces MARS, a multi‑resolution adaptive routing system for sequential recommendation that stores a user’s entire history in recurrent state tracks with different half‑lives. By sparsely routing queries to the appropriate temporal resolution, MARS preserves short‑term intent, medium‑term interests, and long‑term preferences, mitigating the temporal aliasing problem seen in monolithic cached memories. Experiments on three public datasets show that MARS outperforms strong baselines, with larger gains for longer histories and after user behavioral shifts, while maintaining comparable serving latency.
arXiv:2608. 12435v1 Announce Type: new Abstract: Transformers owe much of their strong long-context retrieval capability to a token-level memory that grows with context length.
SequenceO1 is an end‑to‑end framework that enables ultra‑long (up to 100K interactions) sequence modeling for recommendation systems. It compresses raw user histories into a fixed‑size sketch using Sketch Attention and then models short‑term and long‑term interests with Target‑to‑History Cross Attention. The system incorporates low‑rank caching, batching, pipeline lift, and a FlashSA kernel to keep training and inference efficient, achieving consistent offline and online performance gains when deployed at full traffic on Douyin.
arXiv:2606. 07546v1 Announce Type: cross Abstract: Capturing user interests across extensive watch histories is critical for short-form video recommendation, yet scaling sequence length is limited by two bottlenecks: the semantic sparsity of atomic Video IDs and the quadratic computational complexity of Transformers.
BehaveRec is a three-signal framework that combines a causal session transformer, a bidirectional User History Encoder over a 90‑day window, and a GraphSAGE module built from behavioral embedding similarity. On an industrial dataset of 137M sequences, it improves R@5 by +3.2% over SASRec, and its advantage grows with user history length, becoming significant for users with 101+ interactions. The study also establishes a practical deployment rule of roughly 20 interactions per user for when long‑term history and behavioral‑graph signals are worthwhile.
arXiv:2610.35323v1 Announce Type: cross Abstract: Large language models incur substantial training and inference costs in personalized search and recommendation, as growing interaction histories leng...
arXiv:2609.39007v1 Announce Type: cross Abstract: Sessionized interaction histories contain behavioral patterns that can improve sequential recommendation. However, existing models process all sessio...
arXiv:2606. 28533v1 Announce Type: cross Abstract: Sequence learning has emerged as the promising paradigm in recommendation systems, surpassing traditional Deep Learning Recommendation Models (DLRM) by capturing the temporal nuances of user behavior.
The paper introduces CGM-Rec, a continual graph memory framework designed for adaptive recommendation in the presence of intent drift. CGM-Rec treats the knowledge graph as writable memory, comprising a conservative Semantic Graph Memory for stable relational knowledge and a fast-reactive Episodic Lesson Memory for recent outcomes and corrective hints. Experiments show that, with frozen model parameters and one-pass reranking, CGM-Rec outperforms neural and LLM-based baselines across multiple recommendation settings, achieving significant gains such as a 29.58% improvement in HR@1 over the strongest LLM baseline on Bundle.
arXiv:2607. 25357v1 Announce Type: cross Abstract: Long-context recall in linear-time sequence models highlights a tradeoff in how they write to memory.
PRISM introduces personalized item representations in user history, allowing each item’s embedding to reflect its relationship to a specific user rather than using a shared embedding across all users. The method leverages upstream ranking computations and an autoencoder to compress user and item tower states into a single embedding per scored pair for downstream replay. Offline tests on a billion‑scale recommendation surface show a modest but measurable improvement in downstream normalized entropy, with gains scaling with sequence length according to a power‑law relationship.
arXiv:2610.23583v1 Announce Type: new Abstract: Long-term LLM assistants must repeatedly decide what prior context to expose and which model should answer as an interaction evolves. Existing memory s...
arXiv:2610.15642v1 Announce Type: new Abstract: Under a limited context budget, a long-term agent must recall not merely memories that are individually relevant, but evidence that is collectively suf...