Jens Arvidsson & Simon Ström

Automatic classification of opinions

Abstract

The Internet is full of opinions. Reading an opinionated text and judging the author’s stance is something humans are generally expected to be able to do. In this report, the authors build and test a computer system able to read movie reviews and judge the author’s opinion of the film, thereby being able to give numerical grades to any film that has a decent number of reviews. The authors try different machine learning algorithms to judge which is best suited for this task, and compare the results to numerical movie grades found on the Internet Movie Database(IMDb). The algorithm Voted Perceptron comes out the winner, and the system built (named Judge Dr Film) manages to judge films quite close to their IMDb scores. The full source code can be found on github.