… In the remainder of this paper, we first elaborate on the OneRec architecture (Section 2),
detailing … Lastly, we conclude this work, discuss the existing limitations of OneRec, and propose …

… users demonstrates that OneRec-V2 achieves significant improvements compared to
OneRec-V1, … In the remainder of this paper, we first elaborate on the OneRec-V2 architecture and …

… Specifically, we develop two model variants based on the scale of the training corpus:
OneRec and OneRec-Pro. The standard variant is trained exclusively on our publicly released …

… In this paper, we introduced ONeRec, a proactive news recommendation framework designed
to mitigate filter bubbles by strategically guiding users toward new topics. We proposed …

… (2) Auto-regressive-based framework, employing the beam search to decode the item from
whole space, eg, OneRec. In this paper, we devise a unified encoder-decoder framework to …

… Furthermore, to map the items (referred to as item tags in this paper) predicted by LLMIT to
specific items within the in-domain item corpus, we propose a tag-aware semantic relevance …

… Quantized inference has become an essential technique for … OneRec introduces a unified
generative framework that in… 2025b), and subsequent extensions such as OneRec-V2 fur…

… Following OneRec’s successful two-stage post-training pipeline, we adopt SFT on user-item
interaction sequences, followed by group relative policy gradient (GRPO) (DeepSeek-AI et al…

… Using a 1.29% traffic experimental group, we compare OneRec-Think with our online model
for one week and report the result in Table 3, where the primary metric is APP Stay Time (…

… iv) Unifying with OneRec. With the KV Cache friendly architecture, building a world … We
believe such an integration of KSA and OneRec could bridge the gap between language-model-…