Django搜索栈中Elasticsearch MLT搜索失效：如何修改min_term_freq和min_doc_freq？

Hey there! Let's sort out your More Like This (MLT) search issue in your Django project—since you're using Elasticsearch 2, django-haystack, and drf-haystack, here's exactly how to tweak those min_term_freq and min_doc_freq defaults, plus some extra tips to get MLT working smoothly:

You have two main ways to override these default values, depending on whether you want a global change or per-request control:

Option 1: Set Parameters Per Request (in your drf-haystack View) #

If you want to adjust MLT parameters dynamically for specific API calls, modify the SearchQuerySet directly in your view. Here's how to do it:

from haystack.query import SearchQuerySet
from drf_haystack.viewsets import HaystackViewSet
from .serializers import YourSearchSerializer

class YourSearchViewSet(HaystackViewSet):
    serializer_class = YourSearchSerializer
    index_models = [YourModel]  # Your model linked to the Haystack index

    def get_queryset(self):
        sqs = super().get_queryset()
        
        # Check if the request is asking for MLT results
        if self.request.query_params.get('mlt') and self.request.query_params.get('doc_id'):
            target_doc = SearchQuerySet().get(id=self.request.query_params['doc_id'])
            # Override MLT parameters here
            sqs = sqs.more_like_this(
                target_doc,
                min_term_freq=1,  # Set your desired minimum term frequency
                min_doc_freq=1     # Set your desired minimum document frequency
                # Add other MLT params here if needed, like max_query_terms=10
            )
        return sqs

Trigger MLT via an API request like GET /search/?mlt=true&doc_id=your-document-id.

Option 2: Global Override via Custom Elasticsearch Backend #

For a project-wide default change, create a custom Elasticsearch backend that replaces the default MLT parameters:

First, create a haystack_backends.py file in your project directory:

from haystack.backends.elasticsearch_backend import ElasticsearchSearchBackend, ElasticsearchSearchEngine

class CustomElasticsearchBackend(ElasticsearchSearchBackend):
    def __init__(self, connection_alias, **connection_options):
        super().__init__(connection_alias, **connection_options)
        # Override the default MLT parameters here
        self.default_mlt_params = {
            'min_term_freq': 1,
            'min_doc_freq': 1,
            # You can add other defaults here, e.g., max_query_terms=15, stop_words="_english_"
        }

class CustomElasticsearchSearchEngine(ElasticsearchSearchEngine):
    backend = CustomElasticsearchBackend

Then update your settings.py to use this custom backend:

HAYSTACK_CONNECTIONS = {
    'default': {
        'ENGINE': 'your_project_name.haystack_backends.CustomElasticsearchSearchEngine',
        'URL': 'http://localhost:9200/',
        'INDEX_NAME': 'your_index_name',
    },
}

• Double-Check Your Haystack Index: Ensure your SearchIndex includes all fields you want MLT to analyze. Use a template for the document=True field to combine multiple model fields:

  from haystack import indexes
  from .models import YourModel
  
  class YourModelIndex(indexes.SearchIndex, indexes.Indexable):
      text = indexes.CharField(document=True, use_template=True)
  
      def get_model(self):
          return YourModel
  

  Create a template file at templates/search/indexes/your_app/yourmodel_text.txt that lists the fields to include:

  {{ object.title }}
  {{ object.description }}
  {{ object.content }}
  

• Test MLT Directly in Elasticsearch: Rule out Haystack/drf-haystack issues by running an MLT query directly against Elasticsearch with curl:

  curl -X GET "http://your-es-host:9200/your_index/_search" -d '{
      "query": {
          "more_like_this": {
              "fields": ["text"],
              "like": {"_id": "your-document-id"},
              "min_term_freq": 1,
              "min_doc_freq": 1
          }
      }
  }'
  

  If this returns results, the problem is in your Django setup; if not, adjust parameters or check your index data.

• Verify Elasticsearch Field Types: In Elasticsearch 2.x, fields used for MLT should be of type string (not keyword). Haystack usually handles this automatically, but if you have custom mappings, confirm they're set correctly.

• Tweak Other MLT Parameters: If results are still off, experiment with additional parameters like max_query_terms (limits the number of terms used in the MLT query), stop_words (exclude common words), or min_word_length (ignore short words). You can add these to either the per-request more_like_this call or the global default_mlt_params.

内容的提问来源于stack exchange，提问作者ehsan ahmadi