Home

Awesome

jitpack Stars License Apache 2.0 GitHub issues Wechat Group

Similarity

similarity, compute similarity score between text strings, Java written.

similarity,相似度计算工具包,可用于文本相似度计算、情感倾向分析等,Java编写。

similarity是由一系列算法组成的Java版相似度计算工具包,目标是传播自然语言处理中相似度计算方法。similarity具备工具实用、性能高效、架构清晰、语料时新、可自定义的特点。

Feature

similarity提供下列功能:

在提供丰富功能的同时,similarity内部模块坚持低耦合、模型坚持惰性加载、词典坚持明文发布,使用方便,帮助用户训练自己的语料。

Usage

引入Jar包

Maven

<repositories>
  <repository>
    <id>jitpack.io</id>
    <url>https://jitpack.io</url>
  </repository>
</repositories>
<dependency>
  <groupId>com.github.shibing624</groupId>
  <artifactId>similarity</artifactId>
  <version>1.1.6</version>
</dependency>

Gradle

gradle的引入:jitpack

使用示例

import org.xm.Similarity;
import org.xm.tendency.word.HownetWordTendency;

public class demo {
    public static void main(String[] args) {
        double result = Similarity.cilinSimilarity("电动车", "自行车");
        System.out.println(result);

        String word = "混蛋";
        HownetWordTendency hownetWordTendency = new HownetWordTendency();
        result = hownetWordTendency.getTendency(word);
        System.out.println(word + "  词语情感趋势值:" + result);
    }
}

功能演示

1. 词语相似度计算

文本长度:词语粒度

推荐使用词林相似度:org.xm.Similarity.cilinSimilarity,是基于同义词词林的相似度计算方法

example: src/test/java/org.xm/WordSimilarityDemo.java

package org.xm;

public class WordSimilarityDemo {

    public static void main(String[] args) {
        String word1 = "教师";
        String word2 = "教授";
        double cilinSimilarityResult = Similarity.cilinSimilarity(word1, word2);
        double pinyinSimilarityResult = Similarity.pinyinSimilarity(word1, word2);
        double conceptSimilarityResult = Similarity.conceptSimilarity(word1, word2);
        double charBasedSimilarityResult = Similarity.charBasedSimilarity(word1, word2);

        System.out.println(word1 + " vs " + word2 + " 词林相似度值:" + cilinSimilarityResult);
        System.out.println(word1 + " vs " + word2 + " 拼音相似度值:" + pinyinSimilarityResult);
        System.out.println(word1 + " vs " + word2 + " 概念相似度值:" + conceptSimilarityResult);
        System.out.println(word1 + " vs " + word2 + " 字面相似度值:" + charBasedSimilarityResult);
    }
}

word_sim result

2. 短语相似度计算

文本长度:短语粒度

推荐使用短语相似度:org.xm.Similarity.phraseSimilarity,本质是通过两个短语具有的相同字符,和相同字符的位置计算其相似度的方法

example: src/test/java/org.xm/PhraseSimilarityDemo.java

public static void main(String[] args) {
    String phrase1 = "继续努力";
    String phrase2 = "持续发展";
    double result = Similarity.phraseSimilarity(phrase1, phrase2);

    System.out.println(phrase1 + " vs " + phrase2 + " 短语相似度值:" + result);
}

phrase sim result

3. 句子相似度计算

文本长度:句子粒度

推荐使用词形词序句子相似度:org.xm.similarity.morphoSimilarity,一种既考虑两个句子相同文本字面,也考虑相同文本出现的前后顺序的相似度方法

example: src/test/java/org.xm/SentenceSimilarityDemo.java

public static void main(String[] args) {
    String sentence1 = "中国人爱吃鱼";
    String sentence2 = "湖北佬最喜吃鱼";

    double morphoSimilarityResult = Similarity.morphoSimilarity(sentence1, sentence2);
    double editDistanceResult = Similarity.editDistanceSimilarity(sentence1, sentence2);
    double standEditDistanceResult = Similarity.standardEditDistanceSimilarity(sentence1,sentence2);
    double gregeorEditDistanceResult = Similarity.gregorEditDistanceSimilarity(sentence1,sentence2);

    System.out.println(sentence1 + " vs " + sentence2 + " 词形词序句子相似度值:" + morphoSimilarityResult);
    System.out.println(sentence1 + " vs " + sentence2 + " 优化的编辑距离句子相似度值:" + editDistanceResult);
    System.out.println(sentence1 + " vs " + sentence2 + " 标准编辑距离句子相似度值:" + standEditDistanceResult);
    System.out.println(sentence1 + " vs " + sentence2 + " gregeor编辑距离句子相似度值:" + gregeorEditDistanceResult);
}

sentence sim result

4. 段落文本相似度计算

文本长度:段落粒度(一段话,25字符 < length(text) < 500字符)

推荐使用词形词序句子相似度:org.xm.similarity.text.CosineSimilarity,一种考虑两个段落中相同的文本,经过切词,词频和词性权重加权,并用余弦计算相似度的方法

example: src/test/java/org.xm/similarity/text/CosineSimilarityTest.java

@Test
public void getSimilarityScore() throws Exception {
        String text1 = "对于俄罗斯来说,最大的战果莫过于夺取乌克兰首都基辅,也就是现任总统泽连斯基和他政府的所在地。目前夺取基辅的战斗已经打响。";
        String text2 = "迄今为止,俄罗斯的入侵似乎没有完全按计划成功执行——英国国防部情报部门表示,在乌克兰军队激烈抵抗下,俄罗斯军队已经损失数以百计的士兵。尽管如此,俄军在继续推进。";
        TextSimilarity cosSimilarity = new CosineSimilarity();
        double score1 = cosSimilarity.getSimilarity(text1, text2);
        System.out.println("cos相似度分值:" + score1);

        TextSimilarity editSimilarity = new EditDistanceSimilarity();
        double score2 = editSimilarity.getSimilarity(text1, text2);
        System.out.println("edit相似度分值:" + score2);
        }
cos相似度分值:0.399143
edit相似度分值:0.0875

5. 基于义原树的情感分析

example: src/test/java/org/xm/tendency/word/HownetWordTendencyTest.java

@Test
public void getTendency() throws Exception {
    HownetWordTendency hownet = new HownetWordTendency();
    String word = "美好";
    double sim = hownet.getTendency(word);
    System.out.println(word + ":" + sim);
    System.out.println("混蛋:" + hownet.getTendency("混蛋"));
}

tendency result

本例是基于义原树的词语粒度情感极性分析,关于文本情感分析有pytextclassifier,利用深度神经网络模型、SVM分类算法实现的效果更好。

6. 近义词推荐

example: src/test/java/org/xm/word2vec/Word2vecTest.java

@Test
public void testHomoionym() throws Exception {
    List<String> result = Word2vec.getHomoionym(RAW_CORPUS_SPLIT_MODEL, "武功", 10);
    System.out.println("武功 近似词:" + result);
}

@Test
public void testHomoionymName() throws Exception {
    String model = RAW_CORPUS_SPLIT_MODEL;
    List<String> result = Word2vec.getHomoionym(model, "乔帮主", 10);
    System.out.println("乔帮主 近似词:" + result);

    List<String> result2 = Word2vec.getHomoionym(model, "阿朱", 10);
    System.out.println("阿朱 近似词:" + result2);

    List<String> result3 = Word2vec.getHomoionym(model, "少林寺", 10);
    System.out.println("少林寺 近似词:" + result3);
}

word2vec train

word2vec result

Word2vec词向量训练用的java版word2vec训练工具Word2VEC_java,训练语料是小说天龙八部,通过词向量实现得到近义词。 用户可以训练自定义语料,也可以用中文维基百科训练通用词向量。

Todo

文本相似性度量

Contact

<img src="docs/wechat.jpeg" width="200" />

License

授权协议为 The Apache License 2.0,可免费用做商业用途。请在产品说明中附加similarity的链接和授权协议。

Contribute

项目代码还很粗糙,如果大家对代码有所改进,欢迎提交回本项目,在提交之前,注意以下两点:

之后即可提交PR。

Reference