Source folder -
pth.py : holds the path to training and test files. The git only stores tokenized dictionaries. Not the entire xml files given in the website.
pan_util.py: contains utilities to mine the xml train files. for example - given a conversation id get all the conversations
pan_alg.py: creates a self designed feature set with 4 parameters. these parameters were crossvalidated and hard coded in it. These features are then used to train a naive bayes classifier.
Results folder -
The best result so far is 93 % true positive 25 % false positive Scope for improvement - making the features taking into account the frequency when the bayes classifier is being constructed.
For anydoubt contact elango at kth dot se