如何获取ElasticSearch中高亮字段的单独评分？

需求说明 #

现有查询可在索引的4个字段（title、description、subtitle、chapter）中搜索并返回高亮结果，但需要获取每个文档中高亮字段的单独评分，而非整个文档的评分。

业务场景：前端固定展示title字段，需从description、subtitle、chapter三个字段中筛选出与搜索关键词最相关的字段进行展示，因此依赖字段级评分判断相关性。

已尝试无效方案 #

• 使用explain=true，结果过于复杂，无法便捷提取字段评分
• 尝试sub_searches，未生效

当前查询代码 #

$filesQuery['body'] = [
    'size' => 5,
    'query' => [
        'function_score' => [
            'query' => [
                'bool' => [
                    'must' => [
                        [
                            'bool' => [
                                'should' => [
                                    [
                                        'match' => [
                                            'title' => [
                                                'query' => $request['query'],
                                                'boost' => 4
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'title' => [
                                                'value' => '*' . $request['query'] . '*',
                                            ],
                                        ],
                                    ],
                                    [
                                        'match' => [
                                            'description' => [
                                                'query' => $request['query'],
                                                'boost' => 3
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'description' => [
                                                'value' => '*' . $request['query'] . '*',
                                            ],
                                        ],
                                    ],
                                    [
                                        'match' => [
                                            'chapters_v1.chapter' => [
                                                'query' => $request['query'],
                                                'boost' => 3
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'chapters_v1.chapter' => [
                                                'value' => '*' . $request['query'] . '*',
                                            ],
                                        ],
                                    ],
                                    [
                                        'match' => [
                                            'subtitle' => [
                                                'query' => $request['query'],
                                                'boost' => 1
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'subtitle' => [
                                                'value' => '*' . $request['query'] . '*',
                                            ],
                                        ],
                                    ],
                                ],
                            ],
                        ],
                        [
                            'term' => [
                                'team_id' => $teamId,
                            ],
                        ],
                    ],
                ],
            ],
        ],
    ],
    'collapse' => [
        'field' => 'file_id',
    ],
    'aggs' => [
        'total_files' => [
            'cardinality' => [
                'field' => 'file_id',
            ],
        ],
    ],
    'highlight' => [
        'fields' => [
            'title' => [
                'type' => 'plain',
                "fragment_size" => 70,
                "no_match_size" => 70,
            ],
            'description' => [
                'type' => 'plain',
                "fragment_size" => 55,
                "no_match_size" => 55,
            ],
            'subtitle' => [
                'type' => 'plain',
                'number_of_fragments' => 1,
                'fragment_size' => 30,
            ],
            'chapters_v1.chapter' => [
                'type' => 'plain',
                'number_of_fragments' => 1,
                'fragment_size' => 30,
            ],
        ],
    ],
    'track_total_hits' => true,
];

可行解决方案 #

方案思路：命名查询 + 脚本提取字段得分 #

通过给每个字段的查询（match/wildcard）添加唯一命名，再通过script_fields解析查询的_explanation，提取每个字段对应的查询得分总和，得到字段级评分。

修改后的查询代码 #

$filesQuery['body'] = [
    'size' => 5,
    'explain' => true, // 必须开启，才能访问_explanation
    'query' => [
        'function_score' => [
            'query' => [
                'bool' => [
                    'must' => [
                        [
                            'bool' => [
                                'should' => [
                                    // 给title的两个查询命名
                                    [
                                        'match' => [
                                            'title' => [
                                                'query' => $request['query'],
                                                'boost' => 4,
                                                '_name' => 'title_match'
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'title' => [
                                                'value' => '*' . $request['query'] . '*',
                                                '_name' => 'title_wildcard'
                                            ],
                                        ],
                                    ],
                                    // 给description的两个查询命名
                                    [
                                        'match' => [
                                            'description' => [
                                                'query' => $request['query'],
                                                'boost' => 3,
                                                '_name' => 'description_match'
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'description' => [
                                                'value' => '*' . $request['query'] . '*',
                                                '_name' => 'description_wildcard'
                                            ],
                                        ],
                                    ],
                                    // 给chapter的两个查询命名
                                    [
                                        'match' => [
                                            'chapters_v1.chapter' => [
                                                'query' => $request['query'],
                                                'boost' => 3,
                                                '_name' => 'chapter_match'
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'chapters_v1.chapter' => [
                                                'value' => '*' . $request['query'] . '*',
                                                '_name' => 'chapter_wildcard'
                                            ],
                                        ],
                                    ],
                                    // 给subtitle的两个查询命名
                                    [
                                        'match' => [
                                            'subtitle' => [
                                                'query' => $request['query'],
                                                'boost' => 1,
                                                '_name' => 'subtitle_match'
                                            ],
                                        ],
                                    ],
                                    [
                                        'wildcard' => [
                                            'subtitle' => [
                                                'value' => '*' . $request['query'] . '*',
                                                '_name' => 'subtitle_wildcard'
                                            ],
                                        ],
                                    ],
                                ],
                            ],
                        ],
                        [
                            'term' => [
                                'team_id' => $teamId,
                            ],
                        ],
                    ],
                ],
            ],
        ],
    ],
    'collapse' => [
        'field' => 'file_id',
    ],
    'aggs' => [
        'total_files' => [
            'cardinality' => [
                'field' => 'file_id',
            ],
        ],
    ],
    'highlight' => [
        'fields' => [
            'title' => [
                'type' => 'plain',
                "fragment_size" => 70,
                "no_match_size" => 70,
            ],
            'description' => [
                'type' => 'plain',
                "fragment_size" => 55,
                "no_match_size" => 55,
            ],
            'subtitle' => [
                'type' => 'plain',
                'number_of_fragments' => 1,
                'fragment_size' => 30,
            ],
            'chapters_v1.chapter' => [
                'type' => 'plain',
                'number_of_fragments' => 1,
                'fragment_size' => 30,
            ],
        ],
    ],
    // 新增script_fields，提取每个字段的得分
    'script_fields' => [
        'title_score' => [
            'script' => [
                'source' => '
                    def score = 0;
                    for (def exp : _explanation.details) {
                        if (exp.description.contains("title_match") || exp.description.contains("title_wildcard")) {
                            score += exp.value;
                        }
                    }
                    return score;
                '
            ]
        ],
        'description_score' => [
            'script' => [
                'source' => '
                    def score = 0;
                    for (def exp : _explanation.details) {
                        if (exp.description.contains("description_match") || exp.description.contains("description_wildcard")) {
                            score += exp.value;
                        }
                    }
                    return score;
                '
            ]
        ],
        'chapter_score' => [
            'script' => [
                'source' => '
                    def score = 0;
                    for (def exp : _explanation.details) {
                        if (exp.description.contains("chapter_match") || exp.description.contains("chapter_wildcard")) {
                            score += exp.value;
                        }
                    }
                    return score;
                '
            ]
        ],
        'subtitle_score' => [
            'script' => [
                'source' => '
                    def score = 0;
                    for (def exp : _explanation.details) {
                        if (exp.description.contains("subtitle_match") || exp.description.contains("subtitle_wildcard")) {
                            score += exp.value;
                        }
                    }
                    return score;
                '
            ]
        ],
    ],
    'track_total_hits' => true,
];

方案说明 #

1. 命名查询：给每个字段的match和wildcard查询添加_name属性，方便后续在脚本中识别对应字段的查询得分。
2. 开启explain：必须设置'explain' => true，才能在脚本中访问查询的_explanation结构，从中提取各子查询的得分。
3. script_fields提取得分：通过Painless脚本遍历_explanation.details，累加对应字段的所有查询（match+wildcard）的得分，得到该字段的总评分。
4. 结果使用：返回的每个文档会包含fields字段，其中title_score、description_score等就是对应字段的单独评分，前端可根据这些值筛选最相关的字段展示。

优化建议 #

• 如果wildcard查询的权重需要调整，可以在查询中添加boost属性，脚本会自动累加带boost的得分。
• 如果不需要_explanation返回结果，可以在获取数据后忽略该字段，仅保留script_fields中的评分。

内容的提问来源于stack exchange，提问作者kishan maharana