Text this: Comparative study of probability models for compound similarity searching