知识卡片

普通读书笔记卡 · 6779.e · 生成模型:GPT-6

实体感知为关系模型提供线索

理解等级:中级

LUKE把实体信息纳入表示与注意,帮助建模词和实体间联系。预训练信号不等于已适配任意业务关系集;下游仍需选择任务头、实体跨度和有代表性标注,不能只加载模型就当完整关系库。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      One of the state-of-the-art models used nowadays to extract relations is Language Understanding with Knowledge-Based Embeddings (LUKE). LUKE uses entity-aware attention, meaning its training data included information on whether each token is an entity or not. It was also trained to be able to “guess” a masked entity in a Wikipedia-based dataset (rather than just guessing all masked words, like the BERT model was trained).

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第11章《信息抽取与知识图谱》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-11.html) - 结论依据:摘录中关于实体感知为关系模型提供线索的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:One of the state-of-the-art models used nowadays to extract relations is Language Understanding with Knowledge-Based Embeddings (LUKE). LUKE uses entity-aware attention, meaning its training data included information on whether each token is an entity or not. It was also trained to be able to “guess” a masked entity in a Wikipedia-based dataset (rather than just guessing all masked words, like the BERT model was trained).