A New Method of Measuring Document Similarity for Movie Recommendation

Abstract

One of good ways for a person to find a movie the person would like is, supposedly, to find a movie that is similar to other movies that the person found interesting before. For those who like to find movies in this way, there are a few websites where users can search for similar movies to a certain movie. Recommendation systems of such websites rely on rating information or experts' analysis which take a lot of time and efforts of humans to get. This paper suggests a method for scoring similarity among movies using movie scripts. The method we suggest doesn't need any rating information or metadata to analyze. Instead of using those kinds of information, our method is about measuring similarities among documents. There are several conventional methods for the purpose, but it is based on matching keywords. To overcome the limit of matching keywords directly, our method compares two documents in an indirect way.