Filter is slower than that not in a skewed dataset

{
  "from": 0,
  "size": 10,
  "query": {
    "bool": {
      "filter": [
        {
          "terms": {
            "platform": [
              "DPA"
            ],
            "boost": 1
          }
        }
      ],
      "must": [
        {
          "match": {
            "title": {
              "query": "牛奶",
              "minimum_should_match": "1"
            }
          }
        }
      ]
      
    }
  }
}	  

I built a index which has more than 80m documents, in which 90% is on platform "DPA".
My problem is: when I use a filter query, the time used is much more than that not.
I doubt if this situation is expected, and if it is, how should I improve my search time when I need to filter the DPA docs?