04_text_classification

refer

Text classification in general

label kinds: - Discrete labels: - Label is known: - Binary classification: spam filtering/sentiment analysis - Multi-class classification: categorization of goods - Multi-label classification: #hashtag prediction - Label is unknown: i. Text clasterization: user intent search - Continuous labels: predict a salary by CV, predict a price by a product description

Some common classification algorithms: - Maximum Entropy Method = sklearn.linear_model.LogisticRegression - Multinomial Naive Bayes = sklearn.naive_bayes.MultinomialNB

text representation: feature engineering

NBOW

Instead of sparse one-hot encoding we can use pretrained word embedding, which is so-called "Neural Bag-Of-Words".