NLP 다시 정리할 것

numeric or word representation
feature extraction
(feature) encoding
(word) embedding
word representation
numerical representation
vectorization

• categorical data를 one-hot encoding으로 분류 정리
  1 dimensional array : vector로 변경

• bag-of-word : 빈도 기반 단어 표현
  ex) BoW: [1,1,1,2]

• TF-IDF : 빈도 기반 단어 표현(문서 기반)
  Term Frequency X Inverse Document Frequency
  (Term == Word) 특정 단어가 등장한 문서의 개수

• Word2Vec : (dense) embedding
  • neural network model
  • 큰 dataset으로 (a 1.6 billion words data set) word embedding을 학습
  • "의미" == similar words?
  • "king" - "man" = "queen"

• Deep Neural Network

Model : CNN RNN LSTM GRU

Methodology :
sequence-to-sequence
encoder-decoder structure

• Transformer

• Bert
  - Transfer learning
  • Pre-training(Self-supervised Learning) and Fine-tuning(Supervised Learning)