java - 斯坦福 NLP 注释文本非常慢

Question

我正在使用 Stanford CoreNLP 在 Windows 机器上运行 Java 的 NLP 项目。我想从这个注释一篇大文本文章。我写的代码如下；

Properties props = new Properties();
props.put("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref, regexner");
StanfordCoreNLP pipeline =   new StanfordCoreNLP(props);
Annotation document = new Annotation("Text to be annotated. This text is very long!");
pipeline.annotate(document); // this line takes a long time

文本的注释占用了相当长的时间。对于大约 60 个单词，这一特定行大约需要 16 秒，这太长了。

有没有办法加快这个处理过程，或者有什么我遗漏的。请告诉我我能做什么。提前致谢:-)

编辑

代码示例

    public TextReader() {
props.put("annotators", "tokenize, ssplit, pos, lemma, ner, parse, regexner");
pipeline = new StanfordCoreNLP(props);
extractor = CoreMapExpressionExtractor.
                            createExtractorFromFiles(TokenSequencePattern.getNewEnv(), "Stanford NLP\\stanford-corenlp-full-2015-01-29\\stanford-corenlp-full-2015-01-30\\tokensregex\\color.rules.txt");
text = "Barak Obama was born on August 4, 1961,at Kapiolani Maternity & Gynecological Hospital "
+ " in Honolulu, Hawaii, and would become the first President to have been born in Hawaii. His mother, Stanley Ann Dunham,"
+ " was born in Wichita, Kansas, and was of mostly English ancestry. His father, Barack Obama, Sr., was a Luo from Nyang’oma"
+ " Kogelo, Kenya. He studied at the University of Westminster. His favourite colour is red.";
Logger.getLogger(TextReader.class.getName()).log(Level.INFO, "Annotator starting...", text); // LOG 1
Annotation document = new Annotation(text);
pipeline.annotate(document);
Logger.getLogger(TextReader.class.getName()).log(Level.INFO, "Annotator finished...", props); // LOG 2
sentences = document.get(SentencesAnnotation.class);
for (CoreMap sentence : sentences) {
   //the tokens of the sentence are taken and iterated over
   // the NER, POS and lemma of the tokens are stores iteratively
}
}

我意识到 LOG 1 和 LOG 2 之间的时间大约为 16 秒。我需要的是处理更长的文本，这需要很长时间。请告诉我我做错了什么？

谢谢=D

Answer 1

文本是一个长句子吗？相对于句子的长度，解析器的运行时间为 O(n^3)，对于超过 40 个单词的句子来说会变得相当慢。如果删除“parse、dcoref、regexner”注释器，它会加速吗？而且，如果您重新添加“解析”，它会再次变慢吗？

如果您关心的是依赖性解析而不是选区解析，新的“depparse”注释器将更快地生成这些；不过，我们的 coref 还不能用于依赖解析(即将推出!)。