David’s Reviews > Natural Language Processing in Action > Status Update

Not absorbing as much of this past chapter. Feels like the authors are speaking to their peers at this point rather than to students. Could also be that there are some non-listed prerequisites I need to learn first, but am still absorbing a high enough percentage to keep going with this book.

Deeper dive into inverse document frequency. Showed how to build a TF-IDF matrix (using scikit-learn). Listed several - and walked through one - alternative algorithm for TF-IDF normalization

Revisiting vector spaces in more detail, plotting two vectors and calculating cosine similarity between multiple vectors.

Also introduced Zipf's law: given a large enough document the number of occurrences of distinct words - when in descending order - will often differ from the next distinct word(s) in surprisingly neat distances.

Sentiment analysis: rule-based vs machine learning (naive Bayes).

Next, bag-of-words with term-frequency times inverse document frequency (TF-IDF) factored in.

A formal introduction to stop words, and drilling into n-grams a little further. Text normalization techniques, like case folding and stemming.

A tour through the simplest and increasingly more complex and flexible ways to tokenize text (documents)

Introduced Panda DataFrames and Series, illustrated the value of condensing one hot vectors into more compact structures (based on dictionaries). Showcasing the value of bag-of-words vectors and word frequency vectors to simplify a large document to its essence.

Also introduced dot product (aka matrix product)