speedup & bugfix in word2phrase

Faster vocab building and fixed bug when building dictionaries from huge datasets

bugfix in InitUnigramTable() - some words could not have been sampled as negative examples

Bug fix - added questions-phrases.txt

added script for training big word vector model using public corpora

fixed minor bugs

update to 0.1c version

and fixed the minimal cosine similarity to be -1 instead of 0

updated the code so that it would run even with extremely large N (number of closest words), such as 5000

fixed bug: the counter in K-means was updated too frequently

fixed typo back of words -> bag of words

bugfix - the vector computation was done before check for out-of-vocabulary words was performed, which resulted in accessing memory outside of the allocated block

changed makefile so that the .sh scripts would become executable after 'make'

distance -> cosine distance

changed index of OOV from 0 to -1 (so that first word vector is included in search)