Word Embeddings for Natural Language Processing in Python

Word Embeddings for Natural Language Processing in Python

Slides for my talk on word embeddings at PyCon Italy 2017 (PyCon Otto):
https://www.pycon.it/conference/talks/word-embeddings-for-natural-language-processing-in-python
Abstract:
Word embeddings are a family of Natural Language Processing (NLP) algorithms where words are mapped to vectors in low-dimensional space. The interest around word embeddings has been on the rise in the past few years, because these techniques have been driving important improvements in many NLP applications like text classification, sentiment analysis or machine translation.

In this talk we’ll describe the intuitions behind this family of algorithms, we’ll explore some of the Python tools that allow us to implement modern NLP applications and we’ll conclude with some practical considerations.

Aa38bb7a9c35bc414da6ec7dcd8d7339?s=128

Marco Bonzanini

April 08, 2017
Tweet

Transcript

  1. 8.
  2. 10.

    One-hot Encoding Rome Paris Italy France = [1, 0, 0,

    0, 0, 0, …, 0] = [0, 1, 0, 0, 0, 0, …, 0] = [0, 0, 1, 0, 0, 0, …, 0] = [0, 0, 0, 1, 0, 0, …, 0]
  3. 11.

    One-hot Encoding Rome Paris Italy France = [1, 0, 0,

    0, 0, 0, …, 0] = [0, 1, 0, 0, 0, 0, …, 0] = [0, 0, 1, 0, 0, 0, …, 0] = [0, 0, 0, 1, 0, 0, …, 0] Rome Paris word V
  4. 12.

    One-hot Encoding Rome Paris Italy France = [1, 0, 0,

    0, 0, 0, …, 0] = [0, 1, 0, 0, 0, 0, …, 0] = [0, 0, 1, 0, 0, 0, …, 0] = [0, 0, 0, 1, 0, 0, …, 0] V = vocabulary size (huge)
  5. 14.

    Bag-of-words doc_1 doc_2 … doc_N = [32, 14, 1, 0,

    …, 6] = [ 2, 12, 0, 28, …, 12] … = [13, 0, 6, 2, …, 0]
  6. 15.

    Bag-of-words doc_1 doc_2 … doc_N = [32, 14, 1, 0,

    …, 6] = [ 2, 12, 0, 28, …, 12] … = [13, 0, 6, 2, …, 0] Rome Paris word V
  7. 17.

    Word Embeddings Rome Paris Italy France = [0.91, 0.83, 0.17,

    …, 0.41] = [0.92, 0.82, 0.17, …, 0.98] = [0.32, 0.77, 0.67, …, 0.42] = [0.33, 0.78, 0.66, …, 0.97]
  8. 18.

    Word Embeddings Rome Paris Italy France = [0.91, 0.83, 0.17,

    …, 0.41] = [0.92, 0.82, 0.17, …, 0.98] = [0.32, 0.77, 0.67, …, 0.42] = [0.33, 0.78, 0.66, …, 0.97] n. dimensions << vocabulary size
  9. 19.

    Word Embeddings Rome Paris Italy France = [0.91, 0.83, 0.17,

    …, 0.41] = [0.92, 0.82, 0.17, …, 0.98] = [0.32, 0.77, 0.67, …, 0.42] = [0.33, 0.78, 0.66, …, 0.97]
  10. 20.

    Word Embeddings Rome Paris Italy France = [0.91, 0.83, 0.17,

    …, 0.41] = [0.92, 0.82, 0.17, …, 0.98] = [0.32, 0.77, 0.67, …, 0.42] = [0.33, 0.78, 0.66, …, 0.97]
  11. 21.

    Word Embeddings Rome Paris Italy France = [0.91, 0.83, 0.17,

    …, 0.41] = [0.92, 0.82, 0.17, …, 0.98] = [0.32, 0.77, 0.67, …, 0.42] = [0.33, 0.78, 0.66, …, 0.97]
  12. 32.

    I enjoyed eating some pizza at the restaurant I enjoyed

    eating some fiorentina at the restaurant
  13. 33.

    I enjoyed eating some pizza at the restaurant I enjoyed

    eating some fiorentina at the restaurant
  14. 34.

    I enjoyed eating some pizza at the restaurant I enjoyed

    eating some fiorentina at the restaurant Same context
  15. 35.

    I enjoyed eating some pizza at the restaurant I enjoyed

    eating some fiorentina at the restaurant Same context Pizza = Fiorentina ?
  16. 37.
  17. 38.
  18. 43.
  19. 47.

    I enjoyed eating some pizza at the restaurant Maximise the

    likelihood 
 of the context given the focus word
  20. 48.

    I enjoyed eating some pizza at the restaurant Maximise the

    likelihood 
 of the context given the focus word P(i | pizza) P(enjoyed | pizza) … P(restaurant | pizza)
  21. 58.

    I enjoyed eating some pizza at the restaurant Move to

    next focus word and repeat Example
  22. 65.
  23. 66.

    P( vout | vin ) P( vec(eating) | vec(pizza) )

    P( eating | pizza ) Input word Output word
  24. 67.

    P( vout | vin ) P( vec(eating) | vec(pizza) )

    P( eating | pizza ) Input word Output word ???
  25. 73.
  26. 79.
  27. 80.
  28. 83.
  29. 86.

    Case Study 1: Skills and CVs Data set of ~300k

    resumes Each experience is a “sentence” Each experience has 3-15 skills Approx 15k unique skills
  30. 87.

    Case Study 1: Skills and CVs from gensim.models import Word2Vec

    fname = 'candidates.jsonl' corpus = ResumesCorpus(fname) model = Word2Vec(corpus)
  31. 88.
  32. 89.
  33. 90.

    Case Study 1: Skills and CVs Useful for: Data exploration

    Query expansion/suggestion Recommendations
  34. 92.

    Case Study 2: Beer! Data set of ~2.9M beer reviews

    89 different beer styles 635k unique tokens 185M total tokens
  35. 93.

    Case Study 2: Beer! from gensim.models import Doc2Vec fname =

    'ratebeer_data.csv' corpus = RateBeerCorpus(fname) model = Doc2Vec(corpus)
  36. 94.

    Case Study 2: Beer! from gensim.models import Doc2Vec fname =

    'ratebeer_data.csv' corpus = RateBeerCorpus(fname) model = Doc2Vec(corpus) 3.5h on my laptop … remember to pickle
  37. 95.

    Case Study 2: Beer! model.docvecs.most_similar('Stout') [('Sweet Stout', 0.9877), ('Porter', 0.9620),

    ('Foreign Stout', 0.9595), ('Dry Stout', 0.9561), ('Imperial/Strong Porter', 0.9028), ...]
  38. 97.

    Case Study 2: Beer! model.most_similar([model.docvecs['Wheat Ale']]) 
 [('lemon', 0.6103), ('lemony',

    0.5909), ('wheaty', 0.5873), ('germ', 0.5684), ('lemongrass', 0.5653), ('wheat', 0.5649), ('lime', 0.55636), ('verbena', 0.5491), ('coriander', 0.5341), ('zesty', 0.5182)]
  39. 98.

    PCA

  40. 101.
  41. 102.
  42. 104.

    Case Study 2: Beer! Useful for: Understanding the language of

    beer enthusiasts Planning your next pint Classification
  43. 107.

    But we’ve been
 doing this for X years • Approaches

    based on co-occurrences are not new • Think SVD / LSA / LDA • … but they are usually outperformed by word2vec • … and don’t scale as well as word2vec
  44. 109.

    Efficiency • There is no co-occurrence matrix
 (vectors are learned

    directly) • Softmax has complexity O(V)
 Hierarchical Softmax only O(log(V))
  45. 111.

    Garbage in, garbage out • Pre-trained vectors are useful •

    … until they’re not • The business domain is important • The pre-processing steps are important • > 100K words? Maybe train your own model • > 1M words? Yep, train your own model
  46. 112.
  47. 113.

    Summary • Word Embeddings are magic! • Big victory of

    unsupervised learning • Gensim makes your life easy
  48. 115.

    Credits & Readings Credits • Lev Konstantinovskiy (@gensim_py) • Chris

    E. Moody (@chrisemoody) see videos on lda2vec Readings • Deep Learning for NLP (R. Socher) http://cs224d.stanford.edu/ • “word2vec parameter learning explained” by Xin Rong More readings • “GloVe: global vectors for word representation” by Pennington et al. • “Dependency based word embeddings” and “Neural word embeddings as implicit matrix factorization” by O. Levy and Y. Goldberg