Elasticsearch应用Suggestion的代码示例

最新推荐文章于 2025-02-06 17:57:31 发布

mania_yan

最新推荐文章于 2025-02-06 17:57:31 发布

阅读量630

点赞数 8

分类专栏：搜索文章标签： elasticsearch

本文链接：https://blog.youkuaiyun.com/yyw794/article/details/135065286

版权

搜索专栏收录该内容

11 篇文章

订阅专栏

文章详细介绍了如何在SpringdataElasticsearch客户端中使用phraseSuggestion功能进行搜索建议，包括代码片段和处理可能的错误建议的方法。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

为什么会有这篇文章？

ES的java有2个主流客户端，官方的和spring data的。

不管哪一个，关于suggestion的代码示例都非常少。

这里就是为了弥补这个空白。

基于Spring data 的ES客户端

phrase suggestion的代码示例

public SearchResponse phraseSuggest(String indexName, final String text, final boolean isPinyin){
    //log.info("phrase suggest: {}", text);

    final String field = isPinyin? "ner_pinyin.trigram": "ner.trigram";
    //final int maxEdit = isPinyin? 2: 1;
    final int maxEdit = 2;

    QueryBuilder queryBuilder = QueryBuilders.matchPhraseQuery("{{field_name}}", "{{suggestion}}");

    Map<String, Object> params = new HashMap<>();
    params.put("field_name", "ner");

    final PhraseSuggestionBuilder phraseSuggestionBuilder = SuggestBuilders.phraseSuggestion(field)
            .addCandidateGenerator(new DirectCandidateGeneratorBuilder(field)
                    .suggestMode("always")  //如果是搜索，建议为always。如果是纠错，可能missing更合适
                    .maxEdits(maxEdit) //default 2
                    .prefixLength(0)  //default 1
                    .minWordLength(2)  //default 4
                    .maxInspections(5)  //default 5, it times shards_size. bigger but slower
                    .size(100) //default 5
            )
            .text(text)
            .maxErrors(maxEdit) //TODO:
            .confidence(1)  //default 1.0f  //如果是0，或比较低的值，会返回自己
            .gramSize(3)
            //.realWordErrorLikelihood(0.95f)
            .shardSize(100)  //这个设置太小，有些结果会无法召回，例如：深圳市龙岗区龙岗街道宝平路，无法召回 宝坪路
            .size(100)
            .highlight("<em>", "</em>")
            .collateQuery(queryBuilder.toString())
            .collateParams(params)
            .collatePrune(true);
    final SuggestBuilder suggestBuilder = new SuggestBuilder().addSuggestion(SUGGESTION_FIELD, phraseSuggestionBuilder);

    final SearchSourceBuilder searchSourceBuilder = new SearchSourceBuilder();
    searchSourceBuilder.suggest(suggestBuilder);
    final SearchRequest searchRequest = new SearchRequest();
    searchRequest.indices(indexName);
    searchRequest.source(searchSourceBuilder);
    try{
        SearchResponse searchResponse = this.client.search(searchRequest, RequestOptions.DEFAULT);
        return searchResponse;
    }catch(Exception e){
        log.error("{}", e.toString());
        return null;
    }


}

public List<String> handlePossibleMistakeSuggestions(String inputQuery, List<SearchResponse> searchResponses){
    List<String> suggestResults = new ArrayList<>();
    //Map<String, Float> suggestionsWithScore = new HashMap<>();
    for(SearchResponse searchResponse: searchResponses) {
        searchResponse.getSuggest().getSuggestion(SUGGESTION_FIELD).getEntries().forEach(entry -> entry.forEach(option -> {
            // 不带高亮标签的返回结果
            final String esRawPhraseSuggest = option.getText().toString();
            /*
            // 带高亮标签的返回结果
            final String esHighlightSuggest = String.valueOf(option.getHighlighted());
            List<String> highLightWords = extractHighLightWords(esHighlightSuggest);
            if (highLightWords.size()>1) {
                log.warn("{} has more than one highlight", esHighlightSuggest);
            }

             */
            Float score = option.getScore();
            boolean collateMatch = option.collateMatch();
            log.info("phrase suggest({}): {}({})",collateMatch, esRawPhraseSuggest, score);
            String newSuggestion = ingoreNumberFixed(inputQuery, esRawPhraseSuggest);
            if(!suggestResults.contains(newSuggestion)){
                suggestResults.add(newSuggestion);
            }

        }));
    }
    
    return suggestResults;
}