[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"aihot-art-86761":3},{"itemId":4,"vertical":5,"category":6,"source":7,"score":8,"title":9,"summary":10,"analysis":11,"url":12,"coverUrl":13,"direction":13,"marketSignal":13,"publishedAt":14},"86761","ai","论文研究","Meta Engineering Blog",68,"探索层次化兴趣表示用于Meta广告深度漏斗优化","Meta 探索层次化兴趣表示以优化广告深度漏斗。","Meta 探索层次化兴趣表示（HIR）用于广告深度漏斗优化，通过统一嵌入连接用户兴趣与广告主供给。\n· HIR 采用基于 Transformer 的图学习，结合偏差感知注意力和自监督跨视图蒸馏，学习多层级兴趣表示。\n· 利用 LLM 处理多模态广告主和产品内容，丰富稀疏交互，泛化到罕见实体。\n· 输出通用嵌入和“意义袋”兴趣 token，可赋能个性化、检索、排序等广告架构。\n· 在数十亿交互的真实 Meta 广告数据上端到端训练，旨在提升深度漏斗广告效果。\n看点：HIR 可能显著改善 Meta 广告系统的精准度，尤其对长尾和深度转化场景。","https:\u002F\u002Fengineering.fb.com\u002F2026\u002F07\u002F15\u002Fai-research\u002Fexploring-hierarchical-interest-representation-for-meta-ads-deep-funnel-optimization\u002F",null,"2026-07-16 01:00:52"]