基于内容相似度的Elasticsearch固定分数

Question

I am working on a tool to identify similar documents and mark them as duplicated.

To do so, I am using ElasticSearch to check on the documents content so that ElasticSearch take care of managing the synomns and possible typos, however I haven't got to come up with a query that would reach my goals.

So far I came up with this query:

{
 "query":{
    "filtered":{
       "query":{
          "more_like_this":{
             "fields":[
                "description"
             ],
             "like_text":"Lorem ipsum dolor sit amet, consectetur adipiscing elit.",
             "min_term_freq":1,
             "max_query_terms":999,
             "min_doc_freq":1
          }
       }
    }
 },
 "from":0,
 "size":999,
 "search_type": "dfs_query_then_fetch",
 "sort":[
    {
       "_score":{
          "order":"desc"
       }
    }
 ]
}

But it seems like the score it gives me is quite random, I would like to have a score like 100 for contents completely equal while 0 for something that is completely different.

Answer 1

I see where you are going, but out of the box, the scoring is only going to be relevant for that particular query because it is all based on term frequencies and position. so the score is great for results for that query, but meaningless from query to query. So, I would simply wrap that in a constant score query.

If you would be down for putting each term in its own query, I can provide an example of possibly solving this with multiple constant scores ina bool query inside another bool query.