Stanford Nlp是一个比较牛叉的自然语言处理工具,其很多模型都是基于深度学习方法进行训练得到的,准确率比起原来的很多工具有了很大程度的提高。近年来很多开源项目也用到了其中的一些方法。
最近重拾这个工具做点语义分析的工作,但是发现中文资料比较少,入门比较困难,所以整理一下自己的使用方法,希望对有需要的童鞋能够有点帮助。
本文主要是讲如何在Java工程中调用Stanford NLP的API。
一.环境准备
Eclipse或者IDEA,JDK1.8,Apache Maven(注意,3.5及以后的版本都需要Java8环境才能运行,如果不想在Java8运行的话,请使用以前的版本)。
建立好一个新的Maven工程,在pom文件中加入如下代码:
<properties>
<corenlp.version>3.6.0</corenlp.version>
</properties>
<dependencies>
<dependency>
<groupId>edu.stanford.nlp</groupId>
<artifactId>stanford-corenlp</artifactId>
<version>${corenlp.version}</version>
</dependency>
<dependency>
<groupId>edu.stanford.nlp</groupId>
<artifactId>stanford-corenlp</artifactId>
<version>${corenlp.version}</version>
<classifier>models</classifier>
</dependency>
<dependency>
<groupId>edu.stanford.nlp</groupId>
<artifactId>stanford-corenlp</artifactId>
<version>${corenlp.version}</version>
<classifier>models-chinese</classifier>
</dependency>
</dependencies>
三个依赖包分别是CoreNlp的算法包、英文语料包、中文语料包,由于Maven默认镜像在国外,而Stanford NLP的模型文件很大,因此对网络要求比较高,网速慢的一不小心就time out下载失败了。 解决方法是找一个包含Stanford NLP依赖库的国内镜像,修改Maven的setting,xml中的mirror属性。
二.英文文本的处理
英文的处理官网也给出了示例代码,我这里只做一下整合,代码如下:package edu.zju.cst.krselee.examples.english;
import edu.stanford.nlp.dcoref.CorefChain;
import edu.stanford.nlp.dcoref.CorefCoreAnnotations;
import edu.stanford.nlp.ling.CoreAnnotations;
import edu.stanford.nlp.ling.CoreLabel;
import edu.stanford.nlp.pipeline.Annotation;
import edu.stanford.nlp.pipeline.StanfordCoreNLP;
import edu.stanford.nlp.semgraph.SemanticGraph;
import edu.stanford.nlp.semgraph.SemanticGraphCoreAnnotations;
import edu.stanford.nlp.trees.Tree;
import edu.stanford.nlp.trees.TreeCoreAnnotations;
import edu.stanford.nlp.util.CoreMap;
import java.util.List;
import java.util.Map;
import java.util.Properties;
/**
* Created by KrseLee on 2016/11/5.
*/
public class StanfordEnglishNlpExample {
public static void main(String[] args) {
StanfordEnglishNlpExample example = new StanfordEnglishNlpExample();
example.runAllAnnotators();
}
public void runAllAnnotators(){
// creates a StanfordCoreNLP object, with POS tagging, lemmatization, NER, parsing, and coreference resolution
Properties props = new Properties();
props.setProperty("annotators", "tokenize, ssplit, pos, lemma, ner, parse, dcoref");
StanfordCoreNLP pipeline = new StanfordCoreNLP(props);
// read some text in the text variable
String text = "this is a simple text"; // Add your text here!
// create an empty Annotation just with the given text
Annotation document = new Annotation(text);
// run all Annotators on this text
pipeline.annotate(document);
parserOutput(document);
}
public void parserOutput(Annotation document){
// these are all the sentences in this document
// a CoreMap is essentially a Map that uses class objects as keys and has values with custom types
List<CoreMap> sentences = document.get(CoreAnnotations.SentencesAnnotation.class);
for(CoreMap sentence: sentences) {
// traversing the words in the current sentence
// a CoreLabel is a CoreMap with additional token-specific methods
for (CoreLabel token: sentence.get(CoreAnnotations.TokensAnnotation.class)) {
// this is the text of the token
String word = token.get(CoreAnnotations.TextAnnotation.class);
// this is the POS tag of the token
String pos = token.get(CoreAnnotations.PartOfSpeechAnnotation.class);
// this is the NER label of the token
String ne = token.get(CoreAnnotations.NamedEntityTagAnnotation.class);
}
// this is the parse tree of the current sentence
Tree tree = sentence.get(TreeCoreAnnotations.TreeAnnotation.class);
System.out.println("语法树:");
System.out.println(tree.toString());
// this is the Stanford dependency graph of the current sentence
SemanticGraph dependencies = sentence.get(SemanticGraphCoreAnnotations.CollapsedCCProcessedDependenciesAnnotation.class);
System.out.println("依存句法:");
System.out.println(dependencies.toString());
}
// This is the coreference link graph
// Each chain stores a set of mentions that link to each other,
// along with a method for getting the most representative mention
// Both sentence and token offsets start at 1!
Map<Integer, CorefChain> graph =
document.get(CorefCoreAnnotations.CorefChainAnnotation.class);
}
}
值得注意的是,Stanford NLP采用的是pipeline的方式,给用户一个参数的设置接口,之后的过程全都被封装好了,使用起来非常方便。所有的返回结果都保存在一个<pre>Annotation对象中,根据需要去获取。<a target=_blank href="http://nlp.stanford.edu/pubs/StanfordCoreNlp2014.pdf">The Stanford CoreNLP Natural Language Processing Toolkit</a> 一文中对PileLine方式做了详细的介绍,这里就不多说了,
需要提到一点就是参数中,后面的参数往往依赖于前面的参数(直观的讲,就是标注pos依赖于分词tokenize,语法分析paser依赖于标注,等等)。
三.中文文本的处理
相对于英文来说,中文文本的处理稍微麻烦一点,主要的地方在于一个配置文件。中文语料模型包中有一个默认的配置文件StanfordCoreNLP-chinese.properties ,文件内容如下:# Pipeline options - lemma is no-op for Chinese but currently needed because coref demands it (bad old requirements system)annotators = segment, ssplit, pos, lemma, ner, parse, mention, coref# segmentcustomAnnotatorClass.segment = edu.stanford.nlp.pipeline.ChineseSegmenterAnnotatorsegment.model = edu/stanford/nlp/models/segmenter/chinese/ctb.gzsegment.sighanCorporaDict = edu/stanford/nlp/models/segmenter/chinesesegment.serDictionary = edu/stanford/nlp/models/segmenter/chinese/dict-chris6.ser.gzsegment.sighanPostProcessing = true# sentence splitssplit.boundaryTokenRegex = [.]|[!?]+|[。]|[!?]+# pospos.model = edu/stanford/nlp/models/pos-tagger/chinese-distsim/chinese-distsim.tagger# nerner.model = edu/stanford/nlp/models/ner/chinese.misc.distsim.crf.ser.gzner.applyNumericClassifiers = falsener.useSUTime = false# parseparse.model = edu/stanford/nlp/models/lexparser/chineseFactored.ser.gz# corefcoref.sieves = ChineseHeadMatch, ExactStringMatch, PreciseConstructs, StrictHeadMatch1, StrictHeadMatch2, StrictHeadMatch3, StrictHeadMatch4, PronounMatchcoref.input.type = rawcoref.postprocessing = truecoref.calculateFeatureImportance = falsecoref.useConstituencyTree = truecoref.useSemantics = falsecoref.md.type = RULEcoref.mode = hybridcoref.path.word2vec =coref.language = zhcoref.print.md.log = falsecoref.defaultPronounAgreement = truecoref.zh.dict = edu/stanford/nlp/models/dcoref/zh-attributes.txt.gz主要是指定相应pipeline的操作步骤以及对应的语料文件的位置。实际使用中我们可能用不到所有的步骤,或者要使用不同的语料库,因此可以自定义配置文件,再引入代码中。主要的Java程序代码如下:
package edu.zju.cst.krselee.examples.chinese;
import edu.stanford.nlp.dcoref.CorefChain;
import edu.stanford.nlp.dcoref.CorefCoreAnnotations;
import edu.stanford.nlp.ling.CoreAnnotations;
import edu.stanford.nlp.ling.CoreLabel;
import edu.stanford.nlp.pipeline.Annotation;
import edu.stanford.nlp.pipeline.StanfordCoreNLP;
import edu.stanford.nlp.semgraph.SemanticGraph;
import edu.stanford.nlp.semgraph.SemanticGraphCoreAnnotations;
import edu.stanford.nlp.trees.Tree;
import edu.stanford.nlp.trees.TreeCoreAnnotations;
import edu.stanford.nlp.util.CoreMap;
import edu.stanford.nlp.util.PropertiesUtils;
import edu.zju.cst.krselee.examples.english.StanfordEnglishNlpExample;
import java.util.List;
import java.util.Map;
import java.util.Properties;
/**
* Created by KrseLee on 2016/11/4.
*/
public class StanfordChineseNlpExample {
public static void main(String[] args) {
StanfordChineseNlpExample example = new StanfordChineseNlpExample();
example.runChineseAnnotators();
}
public void runChineseAnnotators(){
String text = "克林顿说,华盛顿将逐步落实对韩国的经济援助。"
+ "金大中对克林顿的讲话报以掌声:克林顿总统在会谈中重申,他坚定地支持韩国摆脱经济危机。";
Annotation document = new Annotation(text);
StanfordCoreNLP corenlp = new StanfordCoreNLP("StanfordCoreNLP-chinese.properties");
corenlp.annotate(document);
parserOutput(document);
}
public void parserOutput(Annotation document){
// these are all the sentences in this document
// a CoreMap is essentially a Map that uses class objects as keys and has values with custom types
List<CoreMap> sentences = document.get(CoreAnnotations.SentencesAnnotation.class);
for(CoreMap sentence: sentences) {
// traversing the words in the current sentence
// a CoreLabel is a CoreMap with additional token-specific methods
for (CoreLabel token: sentence.get(CoreAnnotations.TokensAnnotation.class)) {
// this is the text of the token
String word = token.get(CoreAnnotations.TextAnnotation.class);
// this is the POS tag of the token
String pos = token.get(CoreAnnotations.PartOfSpeechAnnotation.class);
// this is the NER label of the token
String ne = token.get(CoreAnnotations.NamedEntityTagAnnotation.class);
System.out.println(word+"\t"+pos+"\t"+ne);
}
// this is the parse tree of the current sentence
Tree tree = sentence.get(TreeCoreAnnotations.TreeAnnotation.class);
System.out.println("语法树:");
System.out.println(tree.toString());
// this is the Stanford dependency graph of the current sentence
SemanticGraph dependencies = sentence.get(SemanticGraphCoreAnnotations.CollapsedCCProcessedDependenciesAnnotation.class);
System.out.println("依存句法:");
System.out.println(dependencies.toString());
}
// This is the coreference link graph
// Each chain stores a set of mentions that link to each other,
// along with a method for getting the most representative mention
// Both sentence and token offsets start at 1!
Map<Integer, CorefChain> graph =
document.get(CorefCoreAnnotations.CorefChainAnnotation.class);
}
}
参考文献:
[1] http://stanfordnlp.github.io/CoreNLP/index.html
[2] https://blog.sectong.com/blog/corenlp_segment.html