WordNet詞網研究7——之JWS（Java Wordnet Similarity）語義相似度計算

JWS——Java WordNet Similarity是由University Of Sussex的David Hope等開發的基於java與WordNet的語義相似度計算開源項目。其中實現了許多經典的語義相似度算法。是一款值得研究的語義相似度計算開源工具。

 

JWS是WordNet::Similarity（一個Perl版的WordNet相似度比較包）的Java實現版本，想用Java實現用WordNet比較詞語相似度的朋友有福拉！！簡述使用步驟：

 

1、下載WordNet（Win、2.1版）：http://wordnet.princeton.edu/wordnet/download/；

 

2、下載WordNet-InfoContent（2.1版）：http://wn-similarity.sourceforge.net/ 或http://www.d.umn.edu/~tpederse/Data/；

 

3、下載JWS（現有版本：beta.11.01）：http://www.cogs.susx.ac.uk/users/drh21/；

 

4、安裝WordNet；

 

5、解壓WordNet-InfoContent-2.1，並將文件夾拷貝至WordNet目錄D:/Program Files/WordNet/2.1下；

 

6、將JWS中的兩個jar包：edu.mit.jwi_2.1.4.jar和edu.sussex.nlp.jws.beta.11.jar拷貝至Java的lib目錄下，並設置環境變量；

 

7、在Eclipse下運行JWS中的例子程序：TestExamples

 

     說明：由於下載的WordNet是2.1版本的，所以程序中有幾處需要修改

 

     String dir = "C:/Program Files/WordNet";    //這里指定WordNet的安裝路徑，按照你實際安裝的路徑加以修改

 

     JWS ws = new JWS(dir, "3.0");                   //把3.0改為2.1即可

程序實例：

 

import java.util.TreeMap;
import java.text.*;
import edu.sussex.nlp.jws.*;


// 'TestExamples': how to use Java WordNet::Similarity
// David Hope, 2008
public class TestExamples
{
     public static void main(String[] args)
    {

// 1. SET UP:
//   Let's make it easy for the user. So, rather than set pointers in 'Environment Variables' etc. let's allow the user to define exactly where they have put WordNet(s)
        String dir = "E:/Commonly Application/WordNet/";
//   That is, you may have version 3.0 sitting in the above directory e.g. C:/Program Files/WordNet/3.0/dict
//   The corresponding IC files folder should be in this same directory e.g. C:/Program Files/WordNet/3.0/WordNet-InfoContent-3.0

//   Option 1  (Perl default): specify the version of WordNet you want to use (assuming that you have a copy of it) and use the default IC file [ic-semcor.dat]
        JWS    ws = new JWS(dir, "2.1");
//   Option 2 : specify the version of WordNet you want to use and the particular IC file that you wish to apply
        //JWS ws = new JWS(dir, "3.0", "ic-bnc-resnik-add1.dat");


// 2. EXAMPLES OF USE:

// 2.1 [JIANG & CONRATH MEASURE]
        JiangAndConrath jcn = ws.getJiangAndConrath();
        //System.out.println("Jiang & Conrath\n");
// all senses
        TreeMap<String, Double>     scores1    =    jcn.jcn("apple", "banana", "n");            // all senses
        //TreeMap<String, Double>     scores1    =    jcn.jcn("apple", 1, "banana", "n");     // fixed;all
        //TreeMap<String, Double>     scores1    =    jcn.jcn("apple", "banana", 2, "n");     // all;fixed
        for(String s : scores1.keySet())
            System.out.println(s + "\t" + scores1.get(s));
// specific senses
        //System.out.println("\nspecific pair\t=\t" + jcn.jcn("apple", 1, "banana", 1, "n") + "\n");
// max.
        ///System.out.println("\nhighest score\t=\t" + jcn.max("java", "best", "n") + "\n\n\n");

//*/
// 2.2 [LIN MEASURE]
        Lin lin = ws.getLin();
        ///System.out.println("Lin\n");
// all senses
        TreeMap<String, Double>     scores2    =    lin.lin("like", "love", "n");            // all senses
        //TreeMap<String, Double>     scores2    =    lin.lin("kid", "child", "n");     // fixed;all
        //TreeMap<String, Double>     scores2    =    lin.lin("apple", "banana", 2, "n");     // all;fixed
        //for(String s : scores2.keySet())
            //System.out.println(s + "\t" + scores2.get(s));
// specific senses
        System.out.println("\nspecific pair\t=\t" + lin.lin("like", 1, "love", 1, "n") + "\n");
// max.
        System.out.println("\nhighest score\t=\t" + lin.max("From","date","n") + "\n\n\n");

// ... and so on for any other measure
    }
} // eof

 

 

簡單實現基於JWS的語義相似度計算程序，例如：

import edu.sussex.nlp.jws.JWS;
import edu.sussex.nlp.jws.Lin;


public class Similar {

    private String str1;
    private String str2;
    private String dir = "E:/Commonly Application/WordNet/";
    private JWS    ws = new JWS(dir, "2.1");
    
    public Similar(String str1,String str2){
        this.str1=str1;
        this.str2=str2;
    }
    
    public double getSimilarity(){
        String[] strs1 = splitString(str1);
        String[] strs2 = splitString(str2);
        double sum = 0.0;
        for(String s1 : strs1){
            for(String s2: strs2){
                double sc= maxScoreOfLin(s1,s2);
                sum+= sc;
                System.out.println("當前計算: "+s1+" VS "+s2+" 的相似度為:"+sc);
            }
        }
        double Similarity = sum /(strs1.length * strs2.length);
        sum=0;
        return Similarity;
    }
    
    private String[] splitString(String str){
        String[] ret = str.split(" ");
        return ret;
    }
    
    private double maxScoreOfLin(String str1,String str2){
        Lin lin = ws.getLin();
        double sc = lin.max(str1, str2, "n");
        if(sc==0){
            sc = lin.max(str1, str2, "v");
        }
        return sc;
    }
    
    public static void main(String args[]){
        String s1="departure";
        String s2="leaving from";
        Similar sm= new Similar(s1, s2);
        System.out.println(sm.getSimilarity());
    }
}

當時碰到想基於protege+Wordnet來處理語義分析這塊，所以接觸到JWS，但沒有太多的時間去深入研究，是一個非常的遺憾，希望有研究的朋友，發個Blog Url，大家參考參考！