Natural language processing with Naive Bayes using NLTK
Last night I gave a presentation at Carbon 5 hack night meetup. The intention was to give an intro to natural language processing (NLP) and Naive Bayes. There’s a lot of different types NLP, but I thought that the best way to introduce it in one night was to focus on Naive Bayes classification and sequential probability estimation. The examples I used were classifying tweets into positive/negative/neutral/unsure categories, and sequentially estimating the probability of emotions in EMO20Q, which I’ve been working on at hack night.
It was fun presenting about programming because in academic talks it usually stops at the equations, while last night I got to go into the details of how it is implemented in NLTK, which I find very clever. It went well but I still had some constructive feedback. By the end of the talk I realized that I was using too many variations on the names of the terms of Naive Bayes, like p(model|data), p(label|feature), or p(category|word). On one hand, I thought it would be good to inoculate the audience to the different jargon, but I realized that it would have been better to stick to one convention and save that jargon inoculation to the end.
Thanks again to Carbon Five for hosting. I wish all presentations could be accompanied by beer and pizza!
Here is the handout version of the slides:
presentationForCarbon5_naiveBayes_handout
Originally published on abewrites.
clustrmaps.html