Part 1: 25 Points. Sentiment analysis for movie reviews.
Implement a naive Bayes classifier for up/down movie reviews. For this part you should use a simple all-words approach to classification. For data, use the polarity dataset v2.o found here.
You should perform a 10-fold cross-validation to report your results.
The application of naive Bayes to this task follows along the same lines as the application to WSD. For a detailed description of NB classifiers as applied to text classification see Manning et al's new IR textbook.
Part 2: 25 Points. Improved sentiment analysis.
Find a way to improve the performance of your system. This part is entirely up to you. The point is to get a score better than the basic NB approach. You can stick with NB and improve the feature set, or dump NB and use a better machine learning model (or both). You might read some of the articles pointed to on the data site for inspiration.
For both parts, email me a short writeup with your results, along with your code.