CS224N lecture02 笔记

发布于 2020-05-18  80 次阅读


Word Vectors 2 and Word Senses

Slides

  • Word Vectors(local vs global)
  • Glove
  • evaluate word vector(intrinsic and extrinsic)

Notes

  • Keyphrases:
    • Global vectors for word representations(Glove);
    • Intrinsic and extrinsic evaluations;
    • Effect of hyperparameters on analogy evalutions;
    • Correlation of human judgment with word vector distances;
    • Dealing with ambiguity in word using contexts;
    • window classification;
  • Evaluations for word vectors
    • Intrinsic: sub-structure; analogy task
    • Extrinsic: real-world task;
  • Glove

Suggested reading

  1. GloVe Global Vectors for Word Representation
    Glove词向量的源paper;全局词共现窗口;回归任务

  2. Improving Distributional Similarity with Lessons Learned from Word Embeddings
    基于预测的词向量模型增加了诸如负采样,词频平滑这样的处理,而传统的一些词向量方法并没有这些trick,这篇文章将这些trick加入传统方法与基于预测的方法生成的词向量进行了比较。

  3. Evaluation methods for unsupervised word embeddings
    intrinsic和extrinsic词向量评估方法在词频分布上的改进;Coherence评估方法;


阿克西斯上没有什么重要的东西