Add parameters for alpha and beta values of the Dirichlet distributions

[GoogleCodeExporter]

The alpha values and the beta prior used for estimating the LDA topic model 
should be configurable. So far, the default values are always used (1.0 and 
0.01 respectively).

Solution:
Mallet takes two arguments:
- the sum of the alpha values over all topics
- one beta value for a single dimension of the Dirichlet prior.
These two values shall be passable as parameters to the Mallet component; if 
they are not given, the afore-mentioned defaults will be used further on.

Original issue reported on code.google.com by schno...@ukp.informatik.tu-darmstadt.de on 6 Jan 2015 at 1:35