我在同时注释多个句子时面临并发问题。我不清楚我是做错了什么，或者CoreNLP中是否有错误。
我的目标是使用并行运行的多个线程使用管道“tokenize，ssplit，pos，lemma，ner，parse，dcoref”来注释句子。每个线程分配自己的StanfordCoreNLP实例，然后将其用于注释。
问题是在某些时候会引发异常：
java.util.ConcurrentModificationException
at java.util.ArrayList$Itr.checkForComodification(ArrayList.java:901)
at java.util.ArrayList$Itr.next(ArrayList.java:851)
at java.util.Collections$UnmodifiableCollection$1.next(Collections.java:1042)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:463)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.analyzeNode(GrammaticalStructure.java:488)
at edu.stanford.nlp.trees.GrammaticalStructure.<init>(GrammaticalStructure.java:201)
at edu.stanford.nlp.trees.EnglishGrammaticalStructure.<init>(EnglishGrammaticalStructure.java:89)
at edu.stanford.nlp.semgraph.SemanticGraphFactory.makeFromTree(SemanticGraphFactory.java:139)
at edu.stanford.nlp.pipeline.DeterministicCorefAnnotator.annotate(DeterministicCorefAnnotator.java:89)
at edu.stanford.nlp.pipeline.AnnotationPipeline.annotate(AnnotationPipeline.java:68)
at edu.stanford.nlp.pipeline.StanfordCoreNLP.annotate(StanfordCoreNLP.java:412)
我正在附加一个应用程序的示例代码，该代码在我的Core i3 370M笔记本电脑（Win 7 64位，Java 1.8.0.45 64位）上大约20秒内重现问题。此应用程序读取识别文本蕴涵（RTE）语料库的XML文件，然后使用标准Java并发类同时解析所有句子。需要将本地RTE XML文件的路径作为命令行参数提供。在我的测试中，我在这里使用了公开的XML文件： http://www.nist.gov/tac/data/RTE/RTE3-DEV-FINAL.tar.gz
package semante.parser.stanford.server;
import java.io.FileInputStream;
import java.io.InputStreamReader;
import java.io.OutputStream;
import java.io.PrintStream;
import java.nio.charset.StandardCharsets;
import java.util.ArrayList;
import java.util.List;
import java.util.Properties;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.Executors;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import javax.xml.bind.JAXBContext;
import javax.xml.bind.Unmarshaller;
import javax.xml.bind.annotation.XmlAccessType;
import javax.xml.bind.annotation.XmlAccessorType;
import javax.xml.bind.annotation.XmlAttribute;
import javax.xml.bind.annotation.XmlElement;
import javax.xml.bind.annotation.XmlRootElement;
import edu.stanford.nlp.pipeline.Annotation;
import edu.stanford.nlp.pipeline.StanfordCoreNLP;
public class StanfordMultiThreadingTest {
@XmlRootElement(name = "entailment-corpus")
@XmlAccessorType (XmlAccessType.FIELD)
public static class Corpus {
@XmlElement(name = "pair")
private List<Pair> pairList = new ArrayList<Pair>();
public void addPair(Pair p) {pairList.add(p);}
public List<Pair> getPairList() {return pairList;}
}
@XmlRootElement(name="pair")
public static class Pair {
@XmlAttribute(name = "id")
String id;
@XmlAttribute(name = "entailment")
String entailment;
@XmlElement(name = "t")
String t;
@XmlElement(name = "h")
String h;
private Pair() {}
public Pair(int id, boolean entailment, String t, String h) {
this();
this.id = Integer.toString(id);
this.entailment = entailment ? "YES" : "NO";
this.t = t;
this.h = h;
}
public String getId() {return id;}
public String getEntailment() {return entailment;}
public String getT() {return t;}
public String getH() {return h;}
}
class NullStream extends OutputStream {
@Override
public void write(int b) {}
};
private Corpus corpus;
private Unmarshaller unmarshaller;
private ExecutorService executor;
public StanfordMultiThreadingTest() throws Exception {
javax.xml.bind.JAXBContext jaxbCtx = JAXBContext.newInstance(Pair.class,Corpus.class);
unmarshaller = jaxbCtx.createUnmarshaller();
executor = Executors.newFixedThreadPool(Runtime.getRuntime().availableProcessors());
}
public void readXML(String fileName) throws Exception {
System.out.println("Reading XML - Started");
corpus = (Corpus) unmarshaller.unmarshal(new InputStreamReader(new FileInputStream(fileName), StandardCharsets.UTF_8));
System.out.println("Reading XML - Ended");
}
public void parseSentences() throws Exception {
System.out.println("Parsing - Started");
// turn pairs into a list of sentences
List<String> sentences = new ArrayList<String>();
for (Pair pair : corpus.getPairList()) {
sentences.add(pair.getT());
sentences.add(pair.getH());
}
// prepare the properties
final Properties props = new Properties();
props.put("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref");
// first run is long since models are loaded
new StanfordCoreNLP(props);
// to avoid the CoreNLP initialization prints (e.g. "Adding annotation pos")
final PrintStream nullPrintStream = new PrintStream(new NullStream());
PrintStream err = System.err;
System.setErr(nullPrintStream);
int totalCount = sentences.size();
AtomicInteger counter = new AtomicInteger(0);
// use java concurrency to parallelize the parsing
for (String sentence : sentences) {
executor.execute(new Runnable() {
@Override
public void run() {
try {
StanfordCoreNLP pipeline = new StanfordCoreNLP(props);
Annotation annotation = new Annotation(sentence);
pipeline.annotate(annotation);
if (counter.incrementAndGet() % 20 == 0) {
System.out.println("Done: " + String.format("%.2f", counter.get()*100/(double)totalCount));
};
} catch (Exception e) {
System.setErr(err);
e.printStackTrace();
System.setErr(nullPrintStream);
executor.shutdownNow();
}
}
});
}
executor.shutdown();
System.out.println("Waiting for parsing to end.");
executor.awaitTermination(10, TimeUnit.MINUTES);
System.out.println("Parsing - Ended");
}
public static void main(String[] args) throws Exception {
StanfordMultiThreadingTest smtt = new StanfordMultiThreadingTest();
smtt.readXML(args[0]);
smtt.parseSentences();
}
}
在我试图找到一些背景信息时，我遇到了来自斯坦福的Christopher Manning和Gabor Angeli给出的答案，这些答案表明当代版本的Stanford CoreNLP应该是线程安全的。但是，CoreNLP 3.4.1版上的最新bug report描述了并发问题。如标题中所述，我使用的是版本3.5.2。
我不清楚我面临的问题是由于错误还是因为我使用包的方式有问题。如果有更多知识渊博的人能够对此有所了解，我将不胜感激。我希望示例代码对于重现问题很有用。谢谢！
[1]：
答案 0 :(得分：9)
您是否尝试过使用foo选项？您可以为单个threads管道指定多个线程，然后它将并行处理句子。
例如，如果您要处理8个核心的句子，请将StanfordCoreNLP选项设置为threads：
8尽管如此，我认为您的解决方案也应该有效，我们会检查是否存在并发错误，但在此期间使用此选项可能会解决您的问题。
答案 1 :(得分：0)
我有同样的问题，并使用最新的github修订版（今天）的构建解决了这个问题。所以我认为这是自3.5.2以来已经解决的CoreNLP问题。