CS 224 N Final Project Summarizing Movie Reviews

Text summarization is a classic problem in natural language processing. Given a body of text, is there an automated way to generate a few sentences that sum up its content? Using movie reviews downloaded from RottenTomatoes.com, along with summary sentences provided by the site, we attempt to find statistical machine learning methods to find acceptable summary sentences in previously unseen movie reviews. The task is inherently difficult due to the relatively unstructured nature of online movie reviews, the large variability in writing styles, and the presence of many possible “good” sentences among which only one will be tagged as the correct “RottenTomatoes” choice. Our best system first classifies each review as either positive or negative in opinion and then uses a unigram language model along with a ranking support vector machine to guess the correct sentence with 26% precision. Human precision on this task was tested to be 40%. In addition, many of the “incorrect” sentences that the system returns are subjectively plausible summaries.