/Minimalist_UIMA

BioNLP oriented UIMA library

Primary LanguageJavaBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

A minimalistic UIMA support library in the uimaFIT style of CCP. http://bionlp-uima.sourceforge.net/ This project includes some basic, simple, uima pipelines and analysis engines cribbed from both UIMA and uimaFIT tutorials, as well as past CCP UIMA work.

In a significant effort to simplify and make more accessible, this work breaks from past UIMA work at the CCP in that it does not include the "util" layer of code originally used to isolate use of UIMA. It fully accepts and embraces UIMA as a standard for Java NLP. Other aspects have also been left behind:

  • Importing Protege3 and Knowtator projects from the project files has been abandonded in favor of XML import that doesn't require the Protege API or complex code of the past.
  • The UIMA_Util functions to move between the util objects and UIMA objects has been abandonded.
  • Without "util", the evolution of that architecture with a set of base core interfaces and "wrapped" implementations, has also been abandonded.
  • Shims and View support is not implemented here. Collection readers include necessary XML to plain-text conversion.

Extending the functionality of UIMA generated annotation classes remains an area of exploration. This was accomplished in past efforts, largely through the use of static functions in the increasingly large UIMA_Util class. This work starts down a similar path using static functions, but divides them up by annotation class extension to aid devleopers in finding them. Utility functions for ClassMentions are found in the ClassMention extenstion class, ClassMentionX for example. Extending the classes was considered, but would require down casting. Directly modifying the classes was also considered, but steps away from easy type system modification and regeneration.

unfinished with apologies Chris Roeder, 2013

ccp_nlp submodules:

  • ts - UIMA type system xml files and generated java

  • ae

    • AnalysisEngineTest.java
    • Debug_AE.java
    • LingPipeSentenceDetector_AE.java
    • LingPipeSentenceDetectorAeTest.java
    • ClassMentionX.java, SlotMentionX.java Extensions of the generated ClassMention and SlotMention classes with static functions that might have been a member of ClassMention or ClassMentionX if it weren't for the fact you can't do type conversion as easily in Java or that UIMA is in control of what objects get created in the JCas.
    • ClassMentionRemovalFilter_AE.java
    • ConceptMapper2CCPTypeSystemConverter_AE.java
    • DbInsert_AE.java - for use with backend
    • Knowtator_AE.java (incomplete)
    • MapNameToIDSlot_AE.java
    • opendmap/OpenDMAP_AE.java
    • Protein_AE.java (simple toy example)

  • lingpipe_analysis_engines - a separate directory for lingpipe analysis engines to isolate the dependency on lingpipe and its license

  • doc - an interface to a local install of medline, as well as additions to integrate pmc oa and a local elsevier collection

    • Code to do a 2-stage parse of fulltext provided in XML. The first stage applies an XSL stylesheet to convert between the source's (complex) XML to a more simple XML style. The second stage parses that and creates various annotations for zoning and italics.
      • XsltConverter.java
      • CcpXmlParser.java
      • ElsevierArt5DtdClasspathResolver.java
      • PmcDtdClasspathResolver.java
      • PmcDtdUriResolver.java
    • Code to access the database and provide service for UIMA collection readers
      • DbConnect.java
      • DocumentProvider.java
      • DocumentProviderFactory.java
      • ElsevierArt5DocumentProvider.java
      • MedlineDocumentProvider.java
      • PmcDocumentProvider.java
    • ORM (JPA) code as well...

  • backend - an interface to tables in the same database as doc, above, but focussed on storing the results of a pipeline run

  • cr - a set of collection readers including DbCollectionReader and derivatives for using the doc directory's database

    • CcpXmlAnnotationFactory.java - a class for bridging from simple data from the doc module to uima annotations
    • DbCollectionReader.java - a base class for collection readers driven by a database to reference either abstracts in the database, such as the Medline CR, or to reference full-text paths such as the Pmc and Elsevier implementations.
    • ElsevierArt5DbCollectionReader.java
    • MedlineDbCollectionReader.java
    • PmcOaDbCollectionReader.java

  • util - a home for utility classes

    • SpacedProperties - a wrapper on java properties that allows for namespaces in the names

  • pipelines

    • BaseUimaFitPipeline.java - a base class using the UIMA base FileSystemCollectionReader

    • ProteinPipeline.java This is basically a type system modification over the RoomNumberAnnotator. The maven pom includes steps to generate and include for compilation the type system classes. It adds usage of the CASDumpWriter from uimafit. Protein.java ProteinAnnotator.java

    • ConceptMapperPipeline.java This set includes work from a number of people showing how to create a ConceptMapper Dictionary from an OBO file, then run CM in a pipeline to create RDF output. It uses xml files to describe the UIMA Analysis Engines. See: https://groups.google.com/forum/?fromgroups#!topic/uimafit-users/BiCdfJrwGBE

      Running ConceptMapper from a uimaFIT universe hasn't been cracked here yet. This code falls back on XML descriptor files and modifying the resulting in-memory descriptor instead of creating the in-memory descriptor directly from factory calls and no xml files. Adapting concept-mapper is awkward because, as far as I'm aware, at the time of this writing, to have an analysis engine use uiamFIT parameters requires that it inherit from a uimaFIT version of JCasAnnotator_ImplBase instead of the UIMA version. Future work would involve finding direct access to the initialization code there and calling it from the init method of a class that derives from ConceptMapper. This avoids modifying ConceptMapper source, though that would certainly work too.

    • KnowtatorPipeline (in devleopment) Past knowtator work at the CCP uses the protege API to access the pins/pont/pprj files. Work has begun on code to read exported XML files. While this avoids the Protege API code, it also avoids older CCP code that interfaces with it.

    • DMapPipeline.java

    • OpenNlpPipeline.java

  • ws - (work in progress)a web_service iniitally surfacing the concept_mapper pipeline

  • rdf_document_interface - an earlier version of doc using triple stores (abandoned)

  • sesame_interface - a connection factory for rdf_document_interface (unused, perhaps useful elswhere)