Java - Chinese Ontology Library API

Prerequisite

• JavaSE-1.7
• UTF-8 File Encoding

EHowNet

Load EHowNet Library and EHowNet Ontology

• Add ontologyAcquisition.jar to classpath

• Get an instance of the Ontology file ehownet_ontology.txt

   EHowNetTree tree = EHowNetTree.getInstance("./docs/ehownet_ontology.txt");

Search

• For example, we search for 「開心」

   List<EHowNetNode> results = tree.searchWord("開心");
   EHowNetNode node = results.get(0);

• If there's no result, an empty List will be returned

Data within a Node

• node.getNodeType(): return NodeType.WORD or NodeType.TAXONOMY
  • Node with type NodeType.WORD has no Hyponym, since it is at the bottom of the Ontology
• For word node:
  • node.getSid(): return an integer denoting the id of the word, for example 61549
  • node.getNodeName(): return a string denoting the name of the word, for example 開心
  • node.getPos(): return a string denoting the part-of-speech tag of the word, for example Nv4,VH21
  • node.getEhownet(): return a string denoting the ehownet's definition of the word, for example {joyful|喜悅}
• For taxonomy node:
  • node.getNodeName(): return a string denoting the name of the taxonomy, for example 物體
  • node.getEhownet(): return a string denoting the ehownet's definition of the word, for example object|物體

Hypernym

• node.getHypernym(): return an EHowNetNode instance, which is the parent of the node. If the node is at the top of the Ontology, the returned value will be null

Hyponym

• node.getHyponymList(): return a List<EHowNetNode> instance, containing all the children of the node. If the node is at the bottom of the Ontology, an empty List will be returned

CKIP Document Converter

Convert a Text File into CKIP-Tagged Document

• Add ontologyAcquisition.jar and jsoup-1.9.2.jar to classpath

• Set the input/output files and convert

   Converter.toCKIP("ckip_input.txt", "ckip_output.txt");

• We can also convert the documents online: http://sunlight.iis.sinica.edu.tw/uwextract/demo.htm

Ontology Acquisition

Load the Acquisition Tools

• Add ontologyAcquisition.jar and jxl.jar to classpath

• Initialize and start with root concept, CKIP-documents and EHowNet

   OntologyAcquisition oa = new OntologyAcquisition("課綱", "./docs/ckip", "./docs/ehownet_ontology.txt");
   oa.start();

Search for a Specific Concept

• For example, we search for 「會議」

   OntologyNode node = oa.searchConcept("會議");

• If the concept does not exist, null will be returned

Data within a Node

• node.getConcept(): return a string denoting the name of the concept, for example 會議 and 記錄
• node.getAttr(): return a List<String> instance, containing all the related concept(but not Hypernym or Hyponym) of the node. If the node has no attributes, an empty List will be returned

Hypernym

• node.getHypernym(): return an OntologyNode instance, which is the parent of the node. If the node is at the top of the Ontology, the returned value will be null

Hyponym

• node.getCategories(): return a List<OntologyNode> instance, containing all the children of the node. If the node is at the bottom of the Ontology, an empty List will be returned

Term/Document Frequency

• oa.getTermFreq("教育"): return an integer, which is the term frequency of 教育
• oa.getDocFreq("教育"): return an integer, which is the document frequency of 教育

Save the Ontology

• oa.dump(): save the Ontology into a new sheet in result.xls

UI Version

• new UIFrame()

Ontology Doc2Vec

Load the Tools and Build the Model

• Add ontologyAcquisition.jar to classpath

• Build the model with domain concept, CKIP-documents, EHowNet and dimension of the output vector

  Doc2Vec d2v = new Doc2Vec("課綱", "./docs/ckip", "./docs/ehownet_ontology.txt", 5);
  VectorModel model = d2v.build();

Features and Valid Dimension

• model.getFeatures(): return a List<String> instance, denoting the features extraced by the model. An empty list will be returned if the process fails
• model.getDimension(): return an integer equal to the valid dimension

Vectors

• model.getDocVectors(): return a Map< String, List<Double> > instance containing all the document vectors. key is the absolute path of a document while value is the vector
• model.getDocVector("docs/ckip/97815.txt"): return a List<Double> instance denoting the vector of the document with path docs/ckip/97815.txt. Both path and absolute path are acceptable for the parameter

Compile and Run the Sample Project

• OntologyDemo is an Eclipse sample project of EHowNet, CKIP-Converter, Ontology Acquisition and Doc2Vec
• For Eclipse:
  • Open the project in workspace
  • Properties-JavaBuildPath-Libraries: add all the JAR files in libs
  • Windows-Perferences-General-Workspace: set the text file encoding to UTF-8
• For Shell:

  • Makefile is available

    • OntologyDemo$ make to compile, OntologyDemo$ make run to run

  • Commands to Compile and Run

    OntologyDemo$ javac -d bin -sourcepath src -encoding utf8 -cp libs/jsoup-1.9.2.jar;libs/jxl.jar;libs/ontologyAcquisition.jar src/Main.java
    
    OntologyDemo$ java -Dfile.encoding=UTF-8 -cp bin;libs/jsoup-1.9.2.jar;libs/jxl.jar;libs/ontologyAcquisition.jar Main
    

Reference

• JExcel
• JSoup
• CKIP Service
• EHowNet