I have searched for titles of articles from BBC, Google news, Yahoo News (trustable source) for negative data. for the positive data I have searched on Google for one liner, jokes, one liner database and for each site I have created a crawler because most of them there are unstructured sites. I have verified the sources not to contains the same sentences
and after that I have chosen randomly 100 negative and 100 positive data to see how good is my created data base.