Solr MLT过滤器如何选择查询值以标识相似文档？

如何解决Solr MLT过滤器如何选择查询值以标识相似文档？

我正在尝试对Solr执行MLT查询，我的查询看起来像这样

{
 "params": {
        "wt": "json","fl": "_id,type,score","start": 0,"rows": 10,"bq": [
            "(*:* AND official_content_b:true)^5","(*:* -viewed_ss:5d376a397d091d605e40eaa4)^5"
        ],"bf": [
            "product(2,popularity_all,recip(ms(Now,timestamp_created_at),1,2592000000,2592000000))","product(2,popularity_recent)","product(sub(def(facet_rating,3),sqrt(metric_rated))"
        ],"_route_": "test.com!","q": "{!edismax}{!mlt boost=true minwl=3 mintf=1 qf='title'}test.com!5bba8e1c66bbaa2423b58c9d","fq": [
            "relationship_domain:test.com AND facet_type:(Item) AND -archived_b:true"
         ],"sort": "score desc,id desc",}
}

原始文档包含以下值：文档中的标题字段，描述和列表字段不存在

"title":["Bobby s Test Video"]

当我在solr6.6中执行此查询并调试结果时，会看到以下结果

test.com!522a209b3f0c5d4fa40000e5": "\n112.74925 = sum of:\n  106.94 = sum of:\n    106.94 = sum of:\n      8.021624 = weight(title:bob in 654408) [BM25Similarity]

它正在其他文档中搜索值“ bob”以标识相似的文档

我将相同的数据索引到solr7.7中，当我执行上述查询时，我看到以下结果

"test.com!560c20e48cc7ae2200001311": "\n44.143784 = sum of:\n  6.1496787 = sum of:\n    6.1496787 = sum of:\n      6.1496787 = weight(title:test in 202) [BM25Similarity]

在solr 6.6和solr 7.7集群中，Lucene版本都设置为6.5.1，为什么这两个不同的solr集群从同一字段中选择不同的单词以查找相似的文档？