Better Twitter Summaries?

Proceedings of NAACL-HLT 2013, pages 445–449, Atlanta, Georgia, 9–14 June 2013. c⃝2013 Association for Computational Linguistics Better Twitter Summaries? Joel Judd & Jugal Kalita Department of Computer Science University of Colorado Colorado Springs, Colorado Email: {jjudd2,jkalita}@uccs.edu Abstract This paper describes an approach to improve summaries for a collection of Twitter posts cre- ated using the Phrase Reinforcement (PR) Al- gorithm (Sharifi et al., 2010a). The PR algo- rithm often generates summaries with excess text and noisy speech. We parse these sum- maries using a dependency parser and use the dependencies to eliminate some of the excess text and build better-formed summaries. We compare the results to those obtained using the PR Algorithm. 1 Introduction Millions of people use the Web to express themselves and share ideas. Twitter is a very popular micro blogging site. According to a recent study approxi- mately 340 million Tweets are sent out every day1. People mostly upload daily routines, fun activities and other words of wisdom for readers. There is also plenty of serious information beyond the personal; according to a study approximately 4% of posts on Twitter have relevant news data2. Topics that may be covered by reputable new sources like CNN (Ca- ble News Network) were considered relevant. A topic is simply a keyword or key phrase that one may use to search for Twitter posts containing it. It is pos- sible to gather large amounts of posts from Twitter on many different topics in short amounts of time. Obviously, processing all this information by human hands is impossible. One way to extract information from Twitter posts on a certain topic is to automat- ically summarize them. (Sharifi et al., 2010a; Sharifi et al., 2010b; Sharifi et al., 2010c) present an al- gorithm called the Phrase Reinforcement Algorithm to produces summaries of a set of Twitter posts on 1http://blog.twitter.com/2012/03/ twitter-turns-six.htm 2http://www.pearanalytics.com/blog/wp-content/ uploads/2010/05/Twitter-Study-August-2009.pdf a certain topic. The PR algorithm produces good summaries for many topics, but for sets of posts on certain topics, the summaries become syntactically malformed or too wordy. This is because the PR Algorithm does not pay much attention to syntactic well-formedness as it constructs a summary sentence from phrases that occur frequently in the posts it summarizes. In this paper, we attempt to improve Twitter summaries produced by the PR algorithm. 2 The PR Algorithm Revisited Given a number of Twitter posts on a certain topic, the PR algorithm starts construction of what is called a word graph with a root node containing the topic phrase. It builds a graph showing how words occur before and after the phrase in the root node, considering all the posts on the topic. It builds a subgraph to the left of the topic phrase and another subgraph to its right in a similar manner. To con- struct the left graph, the algorithm starts with the root node and obtains the set of words that occur immediately before the current node’s phrase. For each of these unique words, the algorithm adds them to the …

Content truncated to 500 words