Machine Learning Lecture 3: working with text + nearest neighbor classification

We continue our work with sentiment analysis from Lecture 2. I go over common ways of preprocessing text in Machine Learning: n-grams, stemming, stop words, wordnet, and part of speech tagging. In part 2 I introduce a common approach to k-nearest neighbor classification with text (It is very similar to something called the vector space model with tf-idf encoding and cosine distance)

Code and other helpful links:
http://karpathy.ca/mlsite/lecture3.php

Category:

Science & Technology

Tags:

• machine
• learning
• data
• mining
• text
• classfication
• stemming
• ngram
• stem
• classifier
• naive
• bayes
• nearest
• neighbor
• sentiment
• analysis
• code
• python
• parsing
• movie
• reviews

[oneparagraphanswers]

Andrej does an _excellent_ job of motivating us as to why log(x) would be better. I have not come across a better explanation of why log works as a "squasher" function.

Wish he'd make more videos!

[koolsatan]

waiting for the next lecture... :)

[ashamrad]

very nice and simple explanation. I hope for more videos from you, thanks very much for this video

[chetan528]

ty. waiting for your SVM video.

[avneetchugh]

very nice explanation of tf-idf, thanks

[persist911]

This is one of the best video I have seen so far and clear and concise also . Can't wait for your SVM videos. Hope you post them soon. Cheers

[kwilliaa]

Good pace and easy to understand. Thanks!