语义去重、去包含关系、去相似

MySimHash：

import com.hankcs.hanlp.seg.common.Term;
import com.hankcs.hanlp.tokenizer.StandardTokenizer;
import org.apache.commons.lang3.StringUtils;
import org.jsoup.Jsoup;
import org.jsoup.safety.Whitelist;

import java.math.BigInteger;
import java.util.HashMap;
import java.util.List;
import java.util.Map;

public class MySimHash {
    private String tokens; //字符串
    private BigInteger strSimHash;//字符产的hash值
    private int hashbits = 64; // 分词后的hash数;


    public MySimHash(String tokens) {
        this.tokens = tokens;
        this.strSimHash = this.simHash();
    }

    MySimHash(String tokens, int hashbits) {
        this.tokens = tokens;
        this.hashbits = hashbits;
        this.strSimHash = this.simHash();
    }


    /**
     * 清除html标签
     * @param content
     * @return
     */
    private String cleanResume(String content) {
        // 若输入为HTML,下面会过滤掉所有的HTML的tag
        content = Jsoup.clean(content, Whitelist.none());
        content = StringUtils.lowerCase(content);
        String[] strings = {" ", "\n", "\r", "\t", "\\r", "\\n", "\\t", " "};
        for (String s : strings) {
            content = content.replaceAll(s, "");
        }
        return content;
    }


    /**
     * 这个是对整个字符串进行hash计算
     * @return
     */
    private BigInteger simHash() {

        tokens = cleanResume(tokens); // cleanResume 删除一些特殊字符

        int[] v = new int[this.hashbits];

        List<Term> termList = StandardTokenizer.segment(this.tokens); // 对字符串进行分词

        //对分词的一些特殊处理 : 比如: 根据词性添加权重 , 过滤掉标点符号 , 过滤超频词汇等;
        Map<String, Integer> weightOfNature = new HashMap<String, Integer>(); // 词性的权重
        weightOfNature.put("n", 2); //给名词的权重是2;
        Map<String, String> stopNatures = new HashMap<String, String>();//停用的词性 如一些标点符号之类的;
        stopNatures.put("w", ""); //
        int overCount = 5; //设定超频词汇的界限 ;
        Map<String, Integer> wordCount = new HashMap<String, Integer>();

        for (Term term : termList) {
            String word = term.word; //分词字符串

            String nature = term.nature.toString(); // 分词属性;
            //  过滤超频词
            if (wordCount.containsKey(word)) {
                int count = wordCount.get(word);
                if (count > overCount) {
                    continue;
                }
                wordCount.put(word, count + 1);
            } else {
                wordCount.put(word, 1);
            }

            // 过滤停用词性
            if (stopNatures.containsKey(nature)) {
                continue;
            }

            // 2、将每一个分词hash为一组固定长度的数列.比如 64bit 的一个整数.
            BigInteger t = this.hash(word);
            for (int i = 0; i < this.hashbits; i++) {
                BigInteger bitmask = new BigInteger("1").shiftLeft(i);
                // 3、建立一个长度为64的整数数组(假设要生成64位的数字指纹,也可以是其它数字),
                // 对每一个分词hash后的数列进行判断,如果是1000...1,那么数组的第一位和末尾一位加1,
                // 中间的62位减一,也就是说,逢1加1,逢0减1.一直到把所有的分词hash数列全部判断完毕.
                int weight = 1;  //添加权重
                if (weightOfNature.containsKey(nature)) {
                    weight = weightOfNature.get(nature);
                }
                if (t.and(bitmask).signum() != 0) {
                    // 这里是计算整个文档的所有特征的向量和
                    v[i] += weight;
                } else {
                    v[i] -= weight;
                }
            }
        }
        BigInteger fingerprint = new BigInteger("0");
        for (int i = 0; i < this.hashbits; i++) {
            if (v[i] >= 0) {
                fingerprint = fingerprint.add(new BigInteger("1").shiftLeft(i));
            }
        }
        return fingerprint;
    }


    /**
     * 对单个的分词进行hash计算;
     * @param source
     * @return
     */
    private BigInteger hash(String source) {
        if (source == null || source.length() == 0) {
            return new BigInteger("0");
        } else {
            /**
             * 当sourece 的长度过短，会导致hash算法失效，因此需要对过短的词补偿
             */
            while (source.length() < 3) {
                source = source + source.charAt(0);
            }
            char[] sourceArray = source.toCharArray();
            BigInteger x = BigInteger.valueOf(((long) sourceArray[0]) << 7);
            BigInteger m = new BigInteger("1000003");
            BigInteger mask = new BigInteger("2").pow(this.hashbits).subtract(new BigInteger("1"));
            for (char item : sourceArray) {
                BigInteger temp = BigInteger.valueOf((long) item);
                x = x.multiply(m).xor(temp).and(mask);
            }
            x = x.xor(new BigInteger(String.valueOf(source.length())));
            if (x.equals(new BigInteger("-1"))) {
                x = new BigInteger("-2");
            }
            return x;
        }
    }

    /**
     * 计算海明距离,海明距离越小说明越相似;
     * @param other
     * @return
     */
    int hammingDistance(MySimHash other) {
        BigInteger m = new BigInteger("1").shiftLeft(this.hashbits).subtract(
                new BigInteger("1"));
        BigInteger x = this.strSimHash.xor(other.strSimHash).and(m);
        int tot = 0;
        while (x.signum() != 0) {
            tot += 1;
            x = x.and(x.subtract(new BigInteger("1")));
        }
        return tot;
    }


    public double getSemblance(MySimHash s2 ){
        double i = (double) this.hammingDistance(s2);
        return 1 - i/this.hashbits ;
    }

    public static void main(String[] args) {

        String s1 = " 云计算厂商服务器采购需求不及预期 ";
        String s2 = " 服务器需求不及预期 ";
        String s3 = " 中金：上调比亚迪H股目标价60%，高端车“汉”持续放量可期 ";
        String s4 = " 中金：上调比亚迪H股目标价60%至138港元，高端车“汉”持续放量可期 ";
        long l3 = System.currentTimeMillis();
        MySimHash hash1 = new MySimHash(s1, 64);
        MySimHash hash2 = new MySimHash(s2, 64);
        MySimHash hash3 = new MySimHash(s3, 64);
        MySimHash hash4 = new MySimHash(s4, 64);
        System.out.println("======================================");
        System.out.println(  hash1.hammingDistance(hash2) );
//        System.out.println(  hash2.hammingDistance(hash3) );
        System.out.println(  hash3.hammingDistance(hash4) );
        System.out.println(  hash1.getSemblance(hash4) );
//        System.out.println(  hash2.getSemblance(hash3) );
        System.out.println(  hash2.getSemblance(hash4) );
        long l4 = System.currentTimeMillis();
        System.out.println(l4-l3);
        System.out.println("======================================");



    }
}

FilterInclusion：

import com.hankcs.hanlp.seg.common.Term;
import com.hankcs.hanlp.tokenizer.StandardTokenizer;
import org.apache.commons.lang3.StringUtils;

import java.util.ArrayList;
import java.util.List;
import java.util.stream.Collectors;


/**
 * created on 2020/10/15
 */
public class FilterInclusion {
    public List<Term> getSegList(String sen) {
        List<Term> termList = StandardTokenizer.segment(sen);
        return termList;
    }

    //    计算句前重复
    public double getSimilarity(String s1, String s2) {
        List ls1 = getSegList(s1);
        List ls2 = getSegList(s2);
        int len = ls1.size() <= ls2.size() ? ls1.size() : ls2.size();
        double count = 0;
        double similarity = 0;
        for (int i = 0; i < len; i++) {
            List ls3 = ls2.subList(0, len);
            if (ls3.contains(ls1.get(i))) {
                count += 1;
            }
        }
        similarity = count / (len);
//        考虑句长
//        similarity = count / (Math.log(ls1.size()+ls2.size()));
        return similarity;
    }

//      计算字符重复次数
//    public double getSimiliarity1(String s1,String s2){
//        double dis = StringUtils.getFuzzyDistance(s1, s2, Locale.CHINA);
//        return dis;
//    }

    //    通过海明距离计算相似度
    public double getSimilarity1(String s1, String s2) {
        double similarity = 0;
        MySimHash hash1 = new MySimHash(s1, 64);
        MySimHash hash2 = new MySimHash(s2, 64);
        similarity = hash1.getSemblance(hash2);
        return similarity;
    }

    public List drop_duplicate(List stringList) {
        System.out.println("stringList:"+stringList.size());
        List indexList = new ArrayList<Integer>();

        //两两过滤
//        for (int i = stringList.size()-1; i >=0; i--) {
//            if (i  == 0) {
//                break;
//            }
//            String str1 = (String) stringList.get(i);
//            String str2 = (String) stringList.get(i - 1);
//            double similarity1 = getSimilarity(str1, str2);
//            double similarity2 = getSimilarity1(str1, str2);
//            if (similarity1 > 0.5 || similarity2>0.65) {
//                if (str1.length() >= str2.length()) {
//                    stringList.remove(i-1);
//                    indexList.add(i-1);
//                } else {
//                    stringList.remove(i);
//                    indexList.add(i);
//                }
//            }
//        }

        //全部过滤
        for (int i = stringList.size()-1; i >=0; i--) {
            if (i == 0) {
                break;
            }
            for (int j = stringList.size()-1; j>=0; j--) {
                if (i==j)continue;
                String str1 = (String) stringList.get(i);
                String str2 = (String) stringList.get(j);
                double similarity1 = getSimilarity(str1, str2);
                double similarity2 = getSimilarity1(str1, str2);
                if (similarity1 > 0.5 || similarity2>0.65) {
                    if (str1.length() >= str2.length()) {
                        indexList.add(j);
                    } else {
//                        stringList.remove(i);
                        indexList.add(i);
                    }
                }
            }
        }
        indexList = (List) indexList.stream().distinct().collect(Collectors.toList());
        System.out.println("indexList:"+indexList.toString());
        return indexList;
    }
    public static void main(String[] args) {
        FilterInclusion test = new FilterInclusion();
        List<String> stringList = new ArrayList<String>();
        stringList.add("农产品和原材料价格波动、磷酸一铵价格大幅下跌风险、成本测算的局限性、环保政策风险、竞争加剧&复合肥销量不及预期风险");
        stringList.add("成本测算局限性");
        stringList.add("新能源汽车需求不及预期");
        stringList.add("新能源汽车销量不及预期");
        stringList.add("疫情导致下半年招生不及预期，核心高管流失，市场竞争激烈");
        stringList.add("行业需求增长不及预期");
        stringList.add("因疫情造成经营业绩下降的风险");
        stringList.add("铅锌产品价格持续大幅下滑");
        stringList.add("公司资源勘探低于预期，铅锌金属价格下跌，全球流动性收紧黄金价格下跌");
//        stringList.add("广州港：前9月预计完成货物吞吐量37874万吨，同比增长1.4%");
//        stringList.add("广州港：前9月预计完成货物吞吐量同比增长1.4%");
//        stringList.add("下游需求不及预期");
//        stringList.add("下游需求不及预期，市场竞争加剧");
//        stringList.add("中金：上调比亚迪H股目标价60%，高端车“汉”持续放量可期");
//        stringList.add("中金：上调比亚迪H股目标价60%至138港元，高端车“汉”持续放量可期");
//        stringList.add("云计算厂商服务器采购需求不及预期");
//        stringList.add("服务器需求不及预期");

        System.out.println(stringList);
        System.out.println(test.drop_duplicate(stringList));
    }
}

直接调用FilterInclusion类，句子以ArrayList输入，返回要删除语句的索引