Yesterday at Tech Event, which was great as always, I presented on sentiment analysis, taking as example movie reviews. I intend to write a little series of blog posts on this, but as I’m not sure when exactly I’ll get to this, here are the pdf version and a link to the notebook.
The focus was not on the classification algorithms per se (treating text as just another domain for classification), but on the difficulties emerging from this being language: Can it really work to look at just single words? Do I need bigrams? Trigrams? More? Can we tackle the complexity using word embeddings – word vectors? Or better, paragraph vectors?
I had a lot of fun exploring this topic, and I really hope to write some posts on this – stay tuned 🙂