Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
Java HTML Lex CSS Shell R
Fetching latest commit…
Cannot retrieve the latest commit at this time.
|Failed to load latest commit information.|
Hoop: a text mining and language exploration workbench in Java. Primarily designed for event extraction and text-based event forecasting, the workbench is a generic text and language machine learning sandbox that can be adapted to a wide variety of tasks. Hoop is a collection of modules (hoops) each of which can take linguistic input from one, process it and pass it onto another. Combining those modules or hoops allows you to create complex analysis systems. Author: Martin van Velsen <email@example.com>,<firstname.lastname@example.org> Background When analyzing any kind of textual data you will soon find yourself in a situation where you're writing the same code over and over again. Each new project requires a version of string tokenizing, each new effort has some form of custom text filtering and cleaning. You feel as if you have to jump through the same hoops over and over again. Hence the creation of a modular toolkit called hoops in which those building blocks are pre-packaged for you and can be easily customized. Goals Within large complex language analysis systems we often focus on analyzing our results statistically without examining the correctness of the individual steps. Even worse, we tend to not look at those cases that get rejected by parsers or have been mis-classified by machine learning classifiers. Hoop attempts to provide a means whereby each hoop in a transformation process or analysis step can be examined and interrogated as it is doing its job. All in all Hoop attempts to provide: - Inspectability Most systems allow you to do an after-action review of a completed pipeline (a CPE in UIMA terms). This makes it very difficult to inspect what data was produced in each step (CASes) and what data was discarded. Hoop integrates an inspection system which can be activated at any time during or after the running of a Hoop sequence. By clicking on the magnifying glass in a selected Hoop panel you can see the data that was created in that step and you can also inspect what data was discarded. - Explainability In a complex system that runs on a cluster in which essentially all the steps happen in a parallel fashion, it can be difficult if not impossible to understand what exactly is happening to the system and the data is processes. Hoop aims to provide both visualization tools to understand how data is managed, manipulated and pushed through the pipeline, as well as make the results comprehensible through enhanced text visualizations (e.g. a document wall showing text highlighting based on likelihood estimates) - Repeatability Initially the Hoop code should make it possible to repeat an experiment hundreds or thousands of times, perhaps in such a way that each time different permutations are tried of a pipeline. However the ultimate goal is to create system that can be run indefinitely akin to online learning but with a strong feedback loop that can integrate previously discarded data if the system detects faults in previously used assumptions. History Initially written as a set of support code for graduate classes in Language Technologies and smaller narrative projects, the code is slowly growing to encompass a larger text-based data mining framework. This source set includes code written for other various graduate courses and is part of a larger research effort in the field of interactive narrative (IN). Language Technologies (CMU, LTI http://www.lti.cs.cmu.edu/) Used Packages: - JDom (included), used as the XML processing and creation workhorse http://www.jdom.org/ - Cobra (included), for webpage rendering http://lobobrowser.org/cobra.jsp - JGraph (included), to render the hoop graph and query trees, etc http://www.jgraph.com/jgraph.html - Hadoop (included), for the indexing part in case you're running through a cluster http://hadoop.apache.org/ - Apache Xerces (included), used to populate certain swing controls from xml, for example the INHoopXMLTreeView. This package is included in the lib directory but can be separately downloaded and linked to externally if so desired. http://xerces.apache.org/mirrors.cgi - MySQL jdbc driver (not included), provided as a means to load data directly from a database. Please download this package separately and add the jar to your classpath. http://dev.mysql.com/downloads/connector/j/ - BerkeleyDB driver (included), mainly integrated to function as a rapid indexing and retieval backend that lives alongside an XML document tree http://www.oracle.com/technetwork/products/berkeleydb/overview/index.html - Stanford NLP (included), mostly provided in hoop form for the purpose of tokenizing, parsing and low-level processing of text data http://nlp.stanford.edu/software/ - UIMA (included, integrated), used for pipeline and cluster management as well as high-level data type specifications http://uima.apache.org/downloads.cgi - frej (integrated), fuzzy pattern recognition based on regular expressions http://frej.sourceforge.net/index.html - JDesktop (integrated) - cjwizard (integrated), used for the generation of wizards such as the application builder http://code.google.com/p/cjwizard/ - Lucene (integrated), used to optionally create a searchable version of the Hoop native document data set http://lucene.apache.org/ Legend: - Included, means that the package is simply made available and is usually the form of a jar or set of jars - Integrated, means that the package is included but also used in various classes and most likely altered or customized - Not Included, means that the code links to needed jars but additional drivers might be needed to make the code operational at runtime Notice! At some point the IDE might be migrated to the Eclipse workbench, although it looks like we don't need to go that elaborate and the current system is almost done anyway.