结合源码分析Solr&Lucene查询打分的工作流程

yxixi

　　基于solr4.4
　　solr中的搜索打分是在QueryComponent中进行的。
在prepare中根据查询的参数，QueryParser对查询语句进行分词，并生成Query对象树。      QParser parser = QParser.getParser(rb.getQueryString(), defType, req);      Query q = parser.getQuery();      if (q == null) {        // normalize a null query to a query that matches nothing        q = new BooleanQuery();              }　　在process方法中，进行搜索打分的过程
　　调用SolrIndexSearcher进行查询，
    SolrIndexSearcher searcher = req.getSearcher();    // normal search result    searcher.search(result,cmd);search(Query query, Filter filter, Collector results)　　SolrIndexSearcher集成lucene的IndexSearcher类，
最终调用IndexSearcher的search(Query query, Filter filter, Collector results)  public void search(Query query, Filter filter, Collector results)    throws IOException {//在这个方法中，会先创建Weight树，计算TermWeight    search(leafContexts, createNormalizedWeight(wrapFilter(query, filter)), results);  }   protected void search(List<AtomicReaderContext> leaves, Weight weight, Collector collector)      throws IOException { .........//根据weight树，构造Score对象树，以及SumScore对象树，为合并倒排表做准备//      Scorer scorer = weight.scorer(ctx, !collector.acceptsDocsOutOfOrder(), true, ctx.reader().getLiveDocs());      if (scorer != null) {        try {//根据SumScorer对象树，进行文档的合并，收集文档结果结合，并进行打分排名          scorer.score(collector);        } catch (CollectionTerminatedException e) {          // collection was terminated prematurely          // continue with the following leaf        }      }    }  }　　1、先看一下Weight对象树的生成，
　　这一部分包括query的打分计算，参见红色部分
　　
IndexSearcher.createNormalizedWeight(Query query)//重写Query对象树     query = rewrite(query);//创建weight对象树，递归计算idf       Weight weight = query.createWeight(this);计算Weight分数，     float v = weight.getValueForNormalization();//计算queryNorm     float norm = getSimilarity().queryNorm(v);    if (Float.isInfinite(norm) || Float.isNaN(norm)) {      norm = 1.0f;    }//将queryNorm的计算打分，递归调用weight     weight.normalize(norm, 1.0f); 根据Query对象树，递归的调用query对象节点的createWeight方法比如BooleanQuery对应的是BooleanWeight对象，每个BooleanWeight包含weight对象数组最终叶子节点为TermWeight对象public TermWeight(IndexSearcher searcher, TermContext termStates)      throws IOException {      assert termStates != null : &quot;TermContext must not be null&quot;;      this.termStates = termStates;      this.similarity = searcher.getSimilarity();//计算idf      this.stats = similarity.computeWeight(          getBoost(),           searcher.collectionStatistics(term.field()),           searcher.termStatistics(term, termStates));    }  public final SimWeight computeWeight(float queryBoost, CollectionStatistics collectionStats, TermStatistics... termStats) {    final Explanation idf = termStats.length == 1    ? idfExplain(collectionStats, termStats[0])    : idfExplain(collectionStats, termStats);    return new IDFStats(collectionStats.field(), idf, queryBoost);  }  public Explanation idfExplain(CollectionStatistics collectionStats, TermStatistics termStats) {    final long df = termStats.docFreq();    final long max = collectionStats.maxDoc();    final float idf = idf(df, max);    return new Explanation(idf, &quot;idf(docFreq=&quot; &#43; df &#43; &quot;, maxDocs=&quot; &#43; max &#43; &quot;)&quot;);  } 计算Weight分数    public float getValueForNormalization() throws IOException {      float sum = 0.0f;      for (int i = 0 ; i < weights.size(); i&#43;&#43;) {        // call sumOfSquaredWeights for all clauses in case of side effects        float s = weights.get(i).getValueForNormalization();         // sum sub weights        if (!clauses.get(i).isProhibited())          // only add to sum for non-prohibited clauses          sum &#43;= s;      }      sum *= getBoost() * getBoost();             // boost each sub-weight      return sum ;    }　　2、根据weight树，构造Score对象树，以及SumScore对象树，为合并倒排表做准备
Scorer scorer = weight.scorer(ctx, !collector.acceptsDocsOutOfOrder(), true, ctx.reader().getLiveDocs());BooleanWeight递归调用节点weight.score创建score对象   public Scorer scorer(AtomicReaderContext context, boolean scoreDocsInOrder,        boolean topScorer, Bits acceptDocs)        throws IOException {      List<Scorer> required = new ArrayList<Scorer>();      List<Scorer> prohibited = new ArrayList<Scorer>();      List<Scorer> optional = new ArrayList<Scorer>();      Iterator<BooleanClause> cIter = clauses.iterator();      for (Weight w  : weights) {        BooleanClause c =  cIter.next();        Scorer subScorer = w.scorer(context, true, false, acceptDocs);         required.add(subScorer);          return new BooleanScorer2(this, disableCoord, minNrShouldMatch, required, prohibited, optional, maxCoord);       }//在创建BooleanScore2的过程中，计算coordBooleanQuery$BooleanWeight,coord，    public float coord(int overlap, int maxOverlap) {      return maxOverlap == 1 ? 1F : similarity.coord(overlap, maxOverlap);    }//最终调用TermWeight.scorer方法，创建score对象 public Scorer scorer(AtomicReaderContext context, boolean scoreDocsInOrder,        boolean topScorer, Bits acceptDocs) throws IOException {      assert termStates.topReaderContext == ReaderUtil.getTopLevelContext(context) : &quot;The top-reader used to create Weight (&quot; &#43; termStates.topReaderContext &#43; &quot;) is not the same as the current reader's top-reader (&quot; &#43; ReaderUtil.getTopLevelContext(context);      final TermsEnum termsEnum = getTermsEnum(context);      if (termsEnum == null) {        return null;      }//Term对应的docs      DocsEnum docs = termsEnum.docs(acceptDocs, null);      assert docs != null;//TermScorer负责doc的打分      return new TermScorer(this, docs, similarity.simScorer(stats, context));    }   TermScorer(Weight weight, DocsEnum td, Similarity.SimScorer docScorer) {    super(weight);    this.docScorer = docScorer;    this.docsEnum = td;  }　　
3、根据SumScorer对象树，进行文档的合并，收集文档结果结合，并进行打分排名          scorer.score(collector);  public void score(Collector collector) throws IOException {    assert docID() == -1; // not started    collector.setScorer(this);    int doc;//在nextDoc的过程中合并document，合并倒排表是按照树的结构进行，先合并子树，子树与子树合并，一直到根    while ((doc = nextDoc()) != NO_MORE_DOCS) {//收集doc，并打分，根据文档的打分，放入优先级队列（最小堆）中      collector.collect(doc);    }  }//整个Score以及SumScorer对象数的打分计算，最终会落到叶子节点TermScorer上TermScorer： @Override  public float score() throws IOException {    assert docID() != NO_MORE_DOCS;    return docScorer.score(docsEnum.docID(), docsEnum.freq());    }
//打分计算公式：tf * norm * weightValue = tf * norm *queryNorm * idf^2 * t.getBoost()TFIDFSimilarity$TFIDFSimScorer    @Override    public float score(int doc, float freq) {//weight是在创建weight阶段的query分词的打分，   //这一部分计算打分公式的蓝色部分，再乘以weight       final float raw = tf(freq) * weightValue; // compute tf(f)*weight，weight=queryNorm * idf^2 * t.getBoost()      return norms == null ? raw : raw * decodeNormValue(norms.get(doc));  // normalize for field, norm部分    }　　
　　
　　
　　
　　
　　
　　
　　
　　
　　
         版权声明：本文为博主原创文章，未经博主允许不得转载。