text - How can I convert probability into score? -

i working on document recommendation program , kinda stuck here. each document, have score assigned according user's actions. then, when new document comes in, need predict how user , rerank whole documents again according scores. solution use threshold divide scores "recommend" , "not recommend". naivebayes or other classification models can either give me label or return possibility of label (i using nltk package text analytics). on right way? question when possibility, how can convert score use ranking? or should use logistic regression in scikit instead? thanks!

i suggest trying out svm-rank algorithm. takes input set of "recommended" , "not recommended" vectors , learns how rank them recommended ones come first. there simple python tool in dlib can use it. see here example: http://dlib.net/svm_rank.py.html