CS224N Final Project: Movie Title Recognition in E-Mails

For this project, a system was designed to rst identify whether or not an email mentioned a movie, and if it does, to extract the title, time and date of the movie in question. The classi er used to determine whether an email is a movie or non-movie is an extension of the Naive Bayes classi er. The classi er was fairly successful in terms of precision, although it tended to yield a lot of false positives. A named entity recognition system using MEMM was utilized to tag movie titles, locations, addresses, dates, and times. The NER system saw much less success than the classi er, although some labels like 'time' did fairly well. The dataset used to train both the classi er and the NER turned out to be fairly small so with more training data the system could see some improvement.