欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页

lucene的多种搜索

程序员文章站 2022-03-05 12:13:11
...

lucene的搜索相当强大,它提供了很多辅助查询类,各自完成一种特殊的查询,也可以相互组合使用,来完成一些复杂的操作.

 

 

public class Test...{   
       
    Analyzer analyzer = new StandardAnalyzer();   
       
    RAMDirectory directory = new RAMDirectory();   
       
    /** *//**   
     * 创建索引  
     *   
     * @throws IOException  
     */  
    public void index() throws IOException...{   
           
        IndexWriter indexWriter = new IndexWriter(directory,analyzer,true);   
           
        Document doc1 = new Document();   
           
        doc1.add(new Field("title","aaabbb",Store.YES,Index.TOKENIZED));   
           
        doc1.add(new Field("content","If you would like to help promote OpenOffice",Store.YES,Index.TOKENIZED));   
           
        doc1.add(new Field("time","2005",Store.YES,Index.TOKENIZED));   
           
        indexWriter.addDocument(doc1);   
           
        Document doc2 = new Document();   
           
        doc2.add(new Field("title","bbcc",Store.YES,Index.TOKENIZED));   
           
        doc2.add(new Field("content","sdfsdfsdfasdds",Store.YES,Index.TOKENIZED));   
           
        doc2.add(new Field("time","2007",Store.YES,Index.TOKENIZED));   
           
        indexWriter.addDocument(doc2);   
           
        indexWriter.optimize();   
           
        indexWriter.close();   
    }   
       
    // 按词条搜索   
    public void termSearcher() throws IOException...{   
           
        IndexSearcher searcher = new IndexSearcher(directory);   
           
        // 查询title中包含aaa   
        Term term = new Term("title","aaa");   
           
        Query query = new TermQuery(term);   
           
        searcher.search(query);   
       
        searcher.close();   
    }   
  
    // 短语搜索    
    public void phraseSearcher() throws IOException...{   
           
        IndexSearcher searcher = new IndexSearcher(directory);   
           
        PhraseQuery phraseQuery = new PhraseQuery();   
           
        // slop 两个项的位置之间允许的最大间隔,这里would和help中间只隔了like(to会被去掉),所以最大间隔设为1就能找到值   
        // 对于两个相连的关键词来说,无论坡度设为多少,都可以找到   
        // 对于不相连的词来说,当两个关键词相隔数小于坡度都可以找到,否则找不到   
        phraseQuery.setSlop(1);   
           
        phraseQuery.add(new Term("content","would"));   
           
        phraseQuery.add(new Term("content","help"));   
           
        // 如果将help放前面,would放后面,需要将would向后移动3个位置才能到help后面(to不算),所以要设为slop最少要设为3   
        // phraseQuery.add(new Term("content","help"));   
           
        // phraseQuery.add(new Term("content","would"));   
           
        // phraseQuery.setSlop(3);   
           
        // 短语的评分是根据项之间距离越小,评分越高,否则越小   
        Hits hits =  searcher.search(phraseQuery);   
           
        printResult(hits);   
           
        searcher.close();   
    }   
       
    // 通配符搜索 WildcardQuery      
    // 通配符包括’?’匹配一个任意字符和’*’匹配零个或多个任意字符,例如你搜索’use*’,你可能找到’user’或者’uses’:      
    public void wildcardSearcher() throws IOException...{   
           
            IndexSearcher searcher = new IndexSearcher(directory);   
               
            // 与正则一样,*代表0个或多个字母,?代表0个或一个字母   
            // WildcardQuery与QueryParser不同的是:WildcardQuery的前缀可以为*,而QueryParser不行   
            WildcardQuery query = new WildcardQuery(new Term("content","a?bbb*"));   
               
            Hits hits = searcher.search(query);   
               
            printResult(hits);   
               
            searcher.close();   
    }   
       
    // 模糊搜索 FuzzyQuery   
    public void fuzzySearcher() throws IOException...{   
               
            IndexSearcher search = new IndexSearcher(directory);   
               
            // OpenOffica虽然没被索引,但能找到相近的OpenOffice   
            FuzzyQuery query = new FuzzyQuery(new Term("content","OpenOffica"));   
               
            Hits hits = search.search(query);   
               
            printResult(hits);   
               
            search.close();   
    }   
       
    // 使用前缀PrefixQuery   
    public void divfixSearcher() throws IOException...{   
           
            IndexSearcher search = new IndexSearcher(directory);   
               
            // 全部title前缀为a   
            PrefixQuery query = new PrefixQuery(new Term("title","b"));   
               
            Hits hits = search.search(query);   
               
            printResult(hits);   
               
            search.close();   
               
    }   
  
    // 范围搜索 RangeQuery   
    public void rangeSearcher() throws IOException...{   
           
        IndexSearcher search = new IndexSearcher(directory);   
           
        // RangeQuery query = new RangeQuery(beginTime, endTime, false);    
        // 开始时间,结束时间,最后一个参数表示是否包含边界条件本身,如果为false   
        RangeQuery query = new RangeQuery(new Term("time","2005"),new Term("time","2007"),true);   
           
        Hits hits = search.search(query);   
           
        printResult(hits);   
           
        search.close();   
    }   
       
       
    // 与或搜索BooleanQuery   
    //BooleanClause用于表示布尔查询子句关系的类,包括:BooleanClause.Occur.MUST,BooleanClause.Occur.MUST_NOT,BooleanClause.Occur.SHOULD。有以下6种组合:   
  
    //1.MUST和MUST:取得连个查询子句的交集。   
  
    //2.MUST和MUST_NOT:表示查询结果中不能包含MUST_NOT所对应得查询子句的检索结果。   
  
    //3.MUST_NOT和MUST_NOT:无意义,检索无结果。   
  
    //4.SHOULD与MUST、SHOULD与MUST_NOT:SHOULD与MUST连用时,无意义,结果为MUST子句的检索结果。与MUST_NOT连用时,功能同MUST。   
  
    //5.SHOULD与SHOULD:表示“或”关系,最终检索结果为所有检索子句的并集。   
  
    public void booleanSearcher() throws IOException, ParseException...{   
           
        IndexSearcher search = new IndexSearcher(directory);   
           
        QueryParser qp1 = new QueryParser("title",new StandardAnalyzer());   
               
        Query query1 = qp1.parse("aa*");   
               
        QueryParser qp2 = new QueryParser("title",new StandardAnalyzer());   
               
        Query query2 = qp2.parse("bb*");   
               
        BooleanQuery query = new BooleanQuery();   
               
        // 搜索结果的title的前双缀可以是aa,或bb   
        query.add(query1, BooleanClause.Occur.SHOULD);   
               
        // BooleanClause.Occur.MUST 必须   
        // BooleanClause.Occur.MUST_NOT 必须不是    
        query.add(query2, BooleanClause.Occur.SHOULD);   
           
        Hits hits = search.search(query);   
               
        printResult(hits);   
           
        search.close();   
           
    }   
       
  
    // 多关键的搜索 PhrasePrefixQuery   
    public void phrasePrefixSearcher() throws IOException...{   
           
        IndexSearcher search = new IndexSearcher(directory);   
           
        PhrasePrefixQuery query = new PhrasePrefixQuery();   
           
        // 这里两项都有可能首先被匹配   
        query.add(new Term[]...{new Term("content","would"),new Term("content","can")});   
           
        // 只有一项必须匹配   
        query.add(new Term("content","help"));   
           
        // If you would like to help promote OpenOffice   
        // can I help you   
        // slop因子的作用域为查询中的所有短语   
        query.setSlop(1);   
           
        // 匹配第一项为 would 或 can 第二项为help   
        // solp设置为1    
        // If you would like to help promote OpenOffice  除去if to 外,would与help的距离=1   
        // can I help you 的距离也=1  所以可以搜索出两条数据   
           
        Hits hits = search.search(query);   
           
        printResult(hits);   
           
        search.close();   
    }   
       
    // 在多个域上查询 MultiFieldQueryParser   
    public void multiFieldSearcher() throws IOException, ParseException...{   
           
        IndexSearcher search = new IndexSearcher(directory);   
           
        // 默认情况下的方式为Occur.SHOULD   
        // titile可以匹配bb,content可以匹配you   
//      MultiFieldQueryParser.parse(new String[]{"bb","you"},new String[]{"title","content"}, analyzer);   
           
        // titile必须匹配bb,content不能匹配   
        Query query = MultiFieldQueryParser.parse( new String[]...{"bb","you"},new String[]...{"title","content"},new BooleanClause.Occur[]...{Occur.MUST,Occur.MUST_NOT}, analyzer);   
           
        // title中必须包含bb  content不能有bb   
        // Query query = MultiFieldQueryParser.parse( "bb*",new String[]{"title","content"},new BooleanClause.Occur[]{BooleanClause.Occur.MUST,BooleanClause.Occur.MUST_NOT}, analyzer);   
           
        Hits hits = search.search(query);   
           
        printResult(hits);   
           
        search.close();   
    }   
       
    public void printResult(Hits hits) throws IOException...{   
        for(int i = 0; i < hits.length(); i++)...{   
            Document d = hits.doc(i);   
            System.out.println(d.get("title"));   
            System.out.println(d.get("content"));   
            System.out.println(d.get("time"));   
        }   
    }   
}  

 

相关标签: lucene