Training on large data and testing on small data using liblinear

Training on large data and testing on small data using liblinear

first of all, I'm very new to machine learning and liblinear in specific,
I've implemented code using python to build the features dictionary and
train the models (-c MaximumCost).. all this part was successfully done,
with a good accuracy. on the testing part, my models are trained on a
large set (more than 50k) while the testing set are very small (around
100), so the result of the accuracy is 50%; recall and precision became
zero ?? after training the classifier using small number of data or equal
to the testing data, accuracy became around 75%, also recall and precision
was reasonable ! ... this suggesting that some parameters are wrong or as
what found i need to do scaling ?? which I really couldn't figure out how
am also not sure if I'm right!?
Thanks in advance!