Preserving Item Semantics for Free: Rethinking Token Initialization in LLM-Based Generative Recommendation
2026-08-11T08:52:10Z•2d05dde53553084939b982d30448a1e941da3fd84420b4f1844d52245edb640d
LLMsagentic systemsarXivdataset leakagefairnessgenerative recommendationinformation retrievalrecommendation systemsretrieval-augmented generationsearch controlsemantic searchvisual retrieval
What happened
A batch of newly announced arXiv computer-science information-retrieval papers covering generative and agentic recommendation, semantic IDs, fairness, visual-memory search infrastructure, retrieval-augmented debate, deep-search control, modality bias, and dataset leakage auditing. The material is research-oriented and does not describe a confirmed cybersecurity vulnerability, exploit, malicious activity, or security incident.
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_ir
- Record identifier
- 2d05dde53553084939b982d30448a1e941da3fd84420b4f1844d52245edb640d
- Enrichment time
- 2026-08-11T08:52:10Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.