Modeling lifetime user behavior sequences is essential for capturing long-range preferences in industrial recommendation, yet sequences containing up to 100K interactions make target-aware ranking prohibitively expensive. Search-based methods that retrieve from raw sequences incur candidate-dependent costs that grow wi...
Wen-Hao Li, Qing-Lin Jia, Cheng-Hou Jin et al.· 0 citations
The KSA V2 is developed, which not only achieves a more flexible KV cache memory compression but also reduces the computation cost in Full attention layer and finds that the summary tokens can serve two roles: the KV memory and sparse indexer.
OneRec Team Chenglong Chu, Guo-Hong Mu, Guo-Wang Zhang et al.· 0 citations
AgentX-Model, the next generation of AgentX's model research framework, is presented, which connects proposal development and model experimentation within sandboxes defined by business inputs and prediction tasks within sandboxes defined by business inputs and prediction tasks.
Shuang Yang, Zi-Jie Zhuang, Chang-Xin Lao et al.· 0 citations
Value-guided decoding is proposed, named RGD, a controllable decoding framework for industrial value-oriented generative recommendation that deriving a closed-form reward guided decoding distribution that principledly combines generation probability with reward signals.
Results show that generative retrieval can combine shared modeling with objective-specific control and complementary candidate generation, and under the same 512-item retrieval budget, Multi-Decoder OneRec improves over the single-decoder OneRec baseline.
This work proposes UniR, a decoder-only Transformer that unifies Generative and Multi-Objective ranking within a single heterogeneous sequence comprising user context, SID trajectory, and item features, validating the practicality of unified model in large-scale recommendation systems.
These results show that RecoReward trains the MLLM to produce item features that benefit downstream recommendation while retaining content-only serving, and shows that RecoReward-9B outperforms its Qwen3.5-9B baseline and all other evaluated models across seven recall metrics.
Guohong Mu, Yue-Yang Liu, Jiangxia Cao et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.