trustgraph-flow 2.7.2__tar.gz → 2.7.3__tar.gz

{trustgraph_flow-2.7.2 → trustgraph_flow-2.7.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: trustgraph-flow
-Version: 2.7.2
+Version: 2.7.3
 Summary: TrustGraph provides a means to run a pipeline of flexible AI processing components in a flexible means to achieve a processing pipeline.
 Author-email: "trustgraph.ai" <security@trustgraph.ai>
 Project-URL: Homepage, https://github.com/trustgraph-ai/trustgraph

{trustgraph_flow-2.7.2 → trustgraph_flow-2.7.3}/trustgraph/extract/kg/ontology/entity_normalizer.py

@@ -23,8 +23,11 @@ def normalize_entity_name(entity_name: str) -> str:
     # Replace spaces and underscores with hyphens
     normalized = re.sub(r'[\s_]+', '-', normalized)
 
-    # Remove any characters that aren't alphanumeric, hyphens, or periods
-    normalized = re.sub(r'[^a-z0-9\-.]', '', normalized)
+    # Remove any characters that aren't word characters (incl. non-ASCII
+    # letters such as CJK), hyphens, or periods. \w is Unicode-aware for
+    # str patterns in Python 3, so non-ASCII entity names are preserved
+    # rather than stripped (which would collapse them onto a single URI).
+    normalized = re.sub(r'[^\w\-.]', '', normalized)
 
     # Remove leading/trailing hyphens
     normalized = normalized.strip('-')
@@ -52,8 +55,10 @@ def normalize_type_identifier(type_id: str) -> str:
     # Replace slashes, colons, and spaces with hyphens
     normalized = re.sub(r'[/:.\s_]+', '-', normalized)
 
-    # Remove any remaining non-alphanumeric characters except hyphens
-    normalized = re.sub(r'[^a-z0-9\-]', '', normalized)
+    # Remove any remaining characters that aren't word characters (incl.
+    # non-ASCII letters such as CJK) or hyphens. \w is Unicode-aware for
+    # str patterns in Python 3, preserving non-ASCII type identifiers.
+    normalized = re.sub(r'[^\w\-]', '', normalized)
 
     # Remove leading/trailing hyphens
     normalized = normalized.strip('-')

trustgraph_flow-2.7.3/trustgraph/flow_version.py

@@ -0,0 +1 @@
+__version__ = "2.7.3"

{trustgraph_flow-2.7.2 → trustgraph_flow-2.7.3}/trustgraph/retrieval/graph_rag/graph_rag.py

@@ -27,6 +27,7 @@ from trustgraph.provenance import (
     set_graph,
     GRAPH_RETRIEVAL, GRAPH_SOURCE,
     TG_CONTAINS, PROV_WAS_DERIVED_FROM,
+    DC_TITLE, RDFS_LABEL,
 )
 
 # Module logger
@@ -500,7 +501,9 @@ class Query:
             edge_uris: List of (s, p, o) URI string tuples
 
         Returns:
-            List of unique document titles
+            (doc_edges, sources): document metadata edges as (s, p, o)
+            string tuples, and per-document source references as
+            {"uri", "title"} dicts sorted by uri
         """
         # Step 1: Find subgraphs containing these edges via tg:contains
         subgraph_tasks = []
@@ -535,7 +538,7 @@ class Query:
                 subgraph_uris.add(str(triple.s))
 
         if not subgraph_uris:
-            return []
+            return [], []
 
         # Step 2: Walk prov:wasDerivedFrom chain to find documents
         current_uris = subgraph_uris
@@ -569,7 +572,7 @@ class Query:
             current_uris = next_uris - doc_uris
 
         if not doc_uris:
-            return []
+            return [], []
 
         # Step 3: Get all document metadata properties
         SKIP_PREDICATES = {
@@ -577,12 +580,14 @@ class Query:
             "http://www.w3.org/1999/02/22-rdf-syntax-ns#type",
         }
 
+        sorted_doc_uris = sorted(doc_uris)
+
         metadata_tasks = [
             self.rag.triples_client.query(
                 s=uri, p=None, o=None, limit=50,
                 collection=self.collection,
             )
-            for uri in doc_uris
+            for uri in sorted_doc_uris
         ]
 
         metadata_results = await asyncio.gather(
@@ -590,18 +595,22 @@ class Query:
         )
 
         doc_edges = []
-        for result in metadata_results:
-            if isinstance(result, Exception) or not result:
-                continue
-            for triple in result:
-                p = str(triple.p)
-                if p in SKIP_PREDICATES:
-                    continue
-                doc_edges.append((
-                    str(triple.s), p, str(triple.o)
-                ))
+        sources = []
+        for uri, result in zip(sorted_doc_uris, metadata_results):
+            title = ""
+            if not isinstance(result, Exception) and result:
+                for triple in result:
+                    p = str(triple.p)
+                    if p in SKIP_PREDICATES:
+                        continue
+                    doc_edges.append((
+                        str(triple.s), p, str(triple.o)
+                    ))
+                    if p == DC_TITLE or (p == RDFS_LABEL and not title):
+                        title = str(triple.o)
+            sources.append({"uri": uri, "title": title})
 
-        return doc_edges
+        return doc_edges, sources
 
 class GraphRag:
     """
@@ -740,15 +749,13 @@ class GraphRag:
             if edge_id(s, p, o) in uri_map
         ]
 
-        source_documents = await q.trace_source_documents(
-            selected_edge_uris,
-        )
-
-        if isinstance(source_documents, Exception):
-            logger.warning(
-                f"Document tracing failed: {source_documents}"
+        try:
+            source_documents, sources = await q.trace_source_documents(
+                selected_edge_uris,
             )
-            source_documents = []
+        except Exception as e:
+            logger.warning(f"Document tracing failed: {e}")
+            source_documents, sources = [], []
 
         # Build focus explainability data with cross-encoder metadata
         selected_edges_with_reasoning = []
@@ -866,4 +873,4 @@ class GraphRag:
             "model": last_model,
         }
 
-        return resp, usage
+        return resp, usage, sources

{trustgraph_flow-2.7.2 → trustgraph_flow-2.7.3}/trustgraph/retrieval/graph_rag/rag.py

@@ -6,7 +6,7 @@ Input is query, output is response.
 
 import logging
 
-from ... schema import GraphRagQuery, GraphRagResponse, Error
+from ... schema import GraphRagQuery, GraphRagResponse, Error, Source
 from ... schema import Triples, Metadata
 from ... provenance import GRAPH_RETRIEVAL
 from . graph_rag import GraphRag
@@ -22,6 +22,9 @@ logger = logging.getLogger(__name__)
 default_ident = "graph-rag"
 default_concurrency = 1
 
+def source_refs(sources):
+    return [Source(uri=s["uri"], title=s["title"]) for s in sources]
+
 class Processor(FlowProcessor):
 
     def __init__(self, **params):
@@ -229,7 +232,7 @@ class Processor(FlowProcessor):
                     )
 
                 # Query with streaming and real-time explain
-                response, usage = await rag.query(
+                response, usage, sources = await rag.query(
                     query = v.query, collection = v.collection,
                     entity_limit = entity_limit, triple_limit = triple_limit,
                     max_subgraph_size = max_subgraph_size,
@@ -245,7 +248,7 @@ class Processor(FlowProcessor):
 
             else:
                 # Non-streaming path with real-time explain
-                response, usage = await rag.query(
+                response, usage, sources = await rag.query(
                     query = v.query, collection = v.collection,
                     entity_limit = entity_limit, triple_limit = triple_limit,
                     max_subgraph_size = max_subgraph_size,
@@ -267,6 +270,7 @@ class Processor(FlowProcessor):
                         in_token=usage.get("in_token"),
                         out_token=usage.get("out_token"),
                         model=usage.get("model"),
+                        sources=source_refs(sources),
                     ),
                     properties={"id": id}
                 )
@@ -281,6 +285,7 @@ class Processor(FlowProcessor):
                     in_token=usage.get("in_token"),
                     out_token=usage.get("out_token"),
                     model=usage.get("model"),
+                    sources=source_refs(sources),
                 ),
                 properties={"id": id}
             )

{trustgraph_flow-2.7.2 → trustgraph_flow-2.7.3}/trustgraph_flow.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: trustgraph-flow
-Version: 2.7.2
+Version: 2.7.3
 Summary: TrustGraph provides a means to run a pipeline of flexible AI processing components in a flexible means to achieve a processing pipeline.
 Author-email: "trustgraph.ai" <security@trustgraph.ai>
 Project-URL: Homepage, https://github.com/trustgraph-ai/trustgraph

trustgraph_flow-2.7.2/trustgraph/flow_version.py

@@ -1 +0,0 @@
-__version__ = "2.7.2"