Open source Java implementation of the TextRank algorithm by Mihalcea, et al. Note that this code only implements key phrase extraction based on keyword co-occurance described in section 3 of the Mihalcea-Tarau paper. This code does not yet implement the sentence extraction described in section 4 of that paper. See also: http://lit.csci.unt.edu/index.php/Graph-based_NLP GitHub code repo: http://github.com/ceteri/textrank/ GoogleGroups discussion: http://groups.google.com/group/textrank-dev Paco NATHAN pacoid@cs.stanford.edu @pacoid http://www.google.com/profiles/ceteri NB: There is a known issue with use of JWNL (Java libraries for WordNet) such that if the graph size exceeds a particular threshold, then low-level Java I/O reads to the WordNet database on disk will cause Java thread to block -- even though JVM tools show no blocked threads. A potential remedy is to dump WordNet, or at least the parts of it used here, into some DBD structure with an in-memory cache. --------- simple test: ant run test with a specific data file FOO.txt ant -Ddata.file=FOO.txt run build the JAR for export to another project: ant jar --------- Sources for third-party JAR files: commons-logging-1.1.1.jar http://commons.apache.org/downloads/download_logging.cgi commons-math-1.2.jar http://commons.apache.org/downloads/download_math.cgi log4j-1.2.15.jar http://logging.apache.org/log4j/1.2/download.html porterstemmer.jar http://snowball.tartarus.org/download.php opennlp-tools-1.3.0.jar http://opennlp.sourceforge.net/ maxent-2.4.0.jar https://sourceforge.net/projects/maxent/ sptoolkit.jar http://text0.mib.man.ac.uk:8080/scottpiao/sent_detector trove-2.0.2.jar http://trove4j.sourceforge.net/ jwnl-1.4rc1.jar http://sourceforge.net/projects/jwordnet jdom-1-1.jar http://jdom.org/downloads/index.html