fixed join_docs.py concatenate

haystack/nodes/other/join_docs.py

@@ -1,11 +1,10 @@
-from collections import defaultdict
 import logging
+from collections import defaultdict
 from math import inf
+from typing import List, Optional
 
-from typing import Optional, List
-
-from haystack.schema import Document
 from haystack.nodes.other.join import JoinNode
+from haystack.schema import Document
 
 logger = logging.getLogger(__name__)
 
@@ -64,7 +63,7 @@ def run_accumulated(self, inputs: List[dict], top_k_join: Optional[int] = None):
         document_map = {doc.id: doc for result in results for doc in result}
 
         if self.join_mode == "concatenate":
-            scores_map = self._concatenate_results(results)
+            scores_map = self._concatenate_results(results, document_map)
         elif self.join_mode == "merge":
             scores_map = self._calculate_comb_sum(results)
         elif self.join_mode == "reciprocal_rank_fusion":
@@ -118,11 +117,22 @@ def run_batch_accumulated(self, inputs: List[dict], top_k_join: Optional[int] =
 
             return output, "output_1"
 
-    def _concatenate_results(self, results):
+    def _concatenate_results(self, results, document_map):
         """
         Concatenates multiple document result lists.
+        Return the documents with the higher score.
         """
-        return {doc.id: doc.score for result in results for doc in result}
+        list_id = list(document_map.keys())
+        scores_map = {}
+        for idx in list_id:
+            tmp = []
+            for result in results:
+                for doc in result:
+                    if doc.id == idx:
+                        tmp.append(doc)
+            item_best_score = max(tmp, key=lambda x: x.score)
+            scores_map.update({idx: item_best_score.score})
+        return scores_map
 
     def _calculate_comb_sum(self, results):
         """

releasenotes/notes/fix-joinDocuments-concatenate-56a7cdba00a7248e.yaml

@@ -0,0 +1,4 @@
+---
+enhancements:
+  - |
+    Make JoinDocuments return only the document with the highest score if there are duplicate documents in the list.

test/nodes/test_join_documents.py

@@ -54,3 +54,27 @@ def test_joindocuments_preserves_root_node():
     join_docs = JoinDocuments()
     result, _ = join_docs.run(inputs)
     assert result["root_node"] == "File"
+
+
+@pytest.mark.unit
+def test_joindocuments_concatenate_keep_only_highest_ranking_duplicate():
+    inputs = [
+        {
+            "documents": [
+                Document(content="text document 1", content_type="text", score=0.2),
+                Document(content="text document 2", content_type="text", score=0.3),
+            ]
+        },
+        {"documents": [Document(content="text document 2", content_type="text", score=0.7)]},
+    ]
+    expected_outputs = {
+        "documents": [
+            Document(content="text document 2", content_type="text", score=0.7),
+            Document(content="text document 1", content_type="text", score=0.2),
+        ]
+    }
+
+    join_docs = JoinDocuments(join_mode="concatenate")
+    result, _ = join_docs.run(inputs)
+    assert len(result["documents"]) == 2
+    assert result["documents"] == expected_outputs["documents"]