如何在lucene中文本开头找到匹配项时增强文档

Question

我想知道这怎么可能。假设我正在搜索ka，那么Karthik的得分应该比Aakash的得分要多。如何提升那些文档？。我已经试过了。

我正在尝试使用SpanFirstQuery，如下所示。但不管用。我用的是Lucene4.0

//queryString is searchText. e.g ka
//NAME, ORGANIZATION_NAME and ORGANIZATION_POSITION are indexed field names.
Map<String, Analyzer> searchAnalyzers = new HashMap<String, Analyzer>();
searchAnalyzers.put(NAME, new KeywordAnalyzer());
searchAnalyzers.put(ORGANIZATION_NAME, new KeywordAnalyzer());
searchAnalyzers.put(ORGANIZATION_POSITION, new KeywordAnalyzer());
PerFieldAnalyzerWrapper perFieldAnalyzerWrapper = new  PerFieldAnalyzerWrapper(new KeywordAnalyzer(), searchAnalyzers);
MultiFieldQueryParser multiFieldQueryParser = new MultiFieldQueryParser(Version.LUCENE_40, mSearchFields, perFieldAnalyzerWrapper); //mSearchFiels is array of fiels
multiFieldQueryParser.setDefaultOperator(QueryParser.Operator.AND);
Query query = (Utils.isEmpty(queryString)) ? new MatchAllDocsQuery() : multiFieldQueryParser.parse(QueryParser.escape(queryString)); //queryString is text to be searched

Term term = new Term(NAME, queryString);
SpanFirstQuery spanFirstQuery = new SpanFirstQuery(new SpanTermQuery(term), 5);
spanFirstQuery.setBoost(5.0f);
BooleanQuery booleanQuery = new BooleanQuery();
booleanQuery.add(spanFirstQuery, BooleanClause.Occur.SHOULD);
booleanQuery.add(query, BooleanClause.Occur.MUST);
indexSearcher.search(booleanQuery, 100);

Answer 1

我的想法是，为什么SpanFirstQuery是一个坏主意--它看起来很像一个变通方法，就性能而言，它可能工作不佳（另外，我不确定如何首先让它工作），而且还需要您存储位置（额外的空间），这并不是真的需要。

提议的解决办法：

警告-这是实验性的，可能不是生产就绪的解决方案，这仍然需要一些工作来做。

public class BoostPrefixScoringRewrite extends ScoringRewrite<BooleanQuery.Builder> {

    private final String text;

    public BoostPrefixScoringRewrite(String text) {
        // todo should be handled more carefully, since wildcard query supports other than * symbols
        this.text = text.replace("*", "");
    }

    @Override
    protected BooleanQuery.Builder getTopLevelBuilder() {
        BooleanQuery.Builder builder = new BooleanQuery.Builder();
        builder.setDisableCoord(true);
        return builder;
    }

    protected Query build(BooleanQuery.Builder builder) {
        return builder.build();
    }

    @Override
    protected void addClause(BooleanQuery.Builder topLevel, Term term, int docCount,
                             float boost, TermContext states) {
        final TermQuery tq = new TermQuery(term, states);
        if (term.text().startsWith(this.text)) {
            // experiment with the boost value
            topLevel.add(new BoostQuery(tq, 100f), BooleanClause.Occur.SHOULD);
        } else {
            topLevel.add(new BoostQuery(tq, boost), BooleanClause.Occur.SHOULD);
        }

    }

    @Override
    protected void checkMaxClauseCount(int count) {
        if (count > BooleanQuery.getMaxClauseCount())
            throw new BooleanQuery.TooManyClauses();
    }
}

注意boosting值，现在它被硬编码为100，这应该足以始终将搜索文本开头的术语放在顶部。另外，需要注意的是--如果您的术语列表会变宽，使用Boolean重写，您可能会面临TooManyClauses异常，而您需要有一个变通方法，以增加这个数字，或者以不同的方式重写这个查询。

要获得完整的测试，请看这里-https://raw.githubusercontent.com/myslionrise/information-retrieval-adventure/master/lucene5/src/main/java/org/mystic/boostbeginningwithtest.java