根据标签对一行中的每个句子进行评分并总结文本。 （爪哇）

我正在尝试用 Java 创建一个摘要器。我正在使用斯坦福对数线性词性标注器 http://nlp.stanford.edu/software/tagger.shtml标记单词，然后，对于某些标记，我对句子进行评分，最后在摘要中，我打印具有高分值的句子。 这是代码：

    MaxentTagger tagger = new MaxentTagger("taggers/bidirectional-distsim-wsj-0-18.tagger");

    BufferedReader reader = new BufferedReader( new FileReader ("C:\\Summarizer\\src\\summarizer\\testing\\testingtext.txt"));
    String line  = null;
    int score = 0;
    StringBuilder stringBuilder = new StringBuilder();
    File tempFile = new File("C:\\Summarizer\\src\\summarizer\\testing\\tempFile.txt");
    Writer writerForTempFile = new BufferedWriter(new FileWriter(tempFile));


    String ls = System.getProperty("line.separator");
    while( ( line = reader.readLine() ) != null )
    {
        stringBuilder.append( line );
        stringBuilder.append( ls );
        String tagged = tagger.tagString(line);
        Pattern pattern = Pattern.compile("[.?!]"); //Find new line
        Matcher matcher = pattern.matcher(tagged);
        while(matcher.find())
        {
            Pattern tagFinder = Pattern.compile("/JJ"); // find adjective tag
            Matcher tagMatcher = tagFinder.matcher(matcher.group());
            while(tagMatcher.find())
            {
                score++; // increase score of sentence for every occurence of adjective tag
            }
            if(score > 1)
                writerForTempFile.write(stringBuilder.toString());
            score = 0;
            stringBuilder.setLength(0);
        }

    }

    reader.close();
    writerForTempFile.close();

上面的代码不起作用。不过，如果我削减我的工作并为每一行（而不是句子）生成分数，它就会起作用。但摘要不是这样生成的，不是吗？ 这是代码：（所有声明与上面相同）

while( ( line = reader.readLine() ) != null )
        {
            stringBuilder.append( line );
            stringBuilder.append( ls );
            String tagged = tagger.tagString(line);
            Pattern tagFinder = Pattern.compile("/JJ"); // find adjective tag
            Matcher tagMatcher = tagFinder.matcher(tagged);
            while(tagMatcher.find())
            {
                score++;  //increase score of line for every occurence of adjective tag
            }
            if(score > 1)
                writerForTempFile.write(stringBuilder.toString());
            score = 0;
            stringBuilder.setLength(0);
        }

EDIT 1:

有关 MaxentTagger 功能的信息。显示其功能的示例代码：

import java.io.IOException;

import edu.stanford.nlp.tagger.maxent.MaxentTagger;

public class TagText {
    public static void main(String[] args) throws IOException,
            ClassNotFoundException {

        // Initialize the tagger
        MaxentTagger tagger = new MaxentTagger(
                "taggers/bidirectional-distsim-wsj-0-18.tagger");

        // The sample string
        String sample = "This is a sample text";

        // The tagged string
        String tagged = tagger.tagString(sample);

        // Output the result
        System.out.println(tagged);
    }
}

Output:

This/DT is/VBZ a/DT sample/NN sentence/NN

EDIT 2:

修改后的代码使用 BreakIterator 来查找断句。但问题仍然存在。

while( ( line = reader.readLine() ) != null )
        {
            stringBuilder.append( line );
            stringBuilder.append( ls );
            String tagged = tagger.tagString(line);
            BreakIterator bi = BreakIterator.getSentenceInstance();
            bi.setText(tagged);
            int end, start = bi.first();
            while ((end = bi.next()) != BreakIterator.DONE)
            {
                String sentence = tagged.substring(start, end);
                Pattern tagFinder = Pattern.compile("/JJ");
                Matcher tagMatcher = tagFinder.matcher(sentence);
                while(tagMatcher.find())
                {
                    score++;
                }
                scoreTracker.add(score);
                if(score > 1)
                    writerForTempFile.write(stringBuilder.toString());
                score = 0;
                stringBuilder.setLength(0);
                start = end;
            }

查找句子中断可能比仅仅查找 [.?!] 更复杂一些，请考虑使用中断迭代器 http://docs.oracle.com/javase/6/docs/api/java/text/BreakIterator.html.getSentenceInstance()

它的性能实际上与 LingPipe 的（更复杂）实现非常相似，并且比 OpenNLP 中的实现更好（至少从我自己的测试来看）。

示例代码

BreakIterator bi = BreakIterator.getSentenceInstance();
bi.setText(text);
int end, start = bi.first();
while ((end = bi.next()) != BreakIterator.DONE) {
    String sentence = text.substring(start, end);
    start = end;
}

Edit

我认为这就是您正在寻找的：

    Pattern tagFinder = Pattern.compile("/JJ");
    BufferedReader reader = getMyReader();
    String line = null;
    while ((line = reader.readLine()) != null) {
        BreakIterator bi = BreakIterator.getSentenceInstance();
        bi.setText(line);
        int end, start = bi.first();
        while ((end = bi.next()) != BreakIterator.DONE) {
            String sentence = line.substring(start, end);
            String tagged = tagger.tagString(sentence);
            int score = 0;
            Matcher tag = tagFinder.matcher(tagged);
            while (tag.find())
                score++;
            if (score > 1)
                writerForTempFile.println(sentence);
            start = end;
        }
    }