Update connector interface with optional Owners information (#798)

backend/danswer/background/indexing/checkpointing.py

@@ -1,11 +1,11 @@
 """Experimental functionality related to splitting up indexing
-into a series of checkpoints to better handle intermmittent failures
+into a series of checkpoints to better handle intermittent failures
 / jobs being killed by cloud providers."""
 import datetime
 
 from danswer.configs.app_configs import EXPERIMENTAL_CHECKPOINTING_ENABLED
 from danswer.configs.constants import DocumentSource
-from danswer.connectors.cross_connector_utils.time_utils import datetime_to_utc
+from danswer.connectors.cross_connector_utils.miscellaneous_utils import datetime_to_utc
 
 
 def _2010_dt() -> datetime.datetime:

backend/danswer/connectors/confluence/connector.py

@@ -18,6 +18,7 @@ from danswer.connectors.interfaces import GenerateDocumentsOutput
 from danswer.connectors.interfaces import LoadConnector
 from danswer.connectors.interfaces import PollConnector
 from danswer.connectors.interfaces import SecondsSinceUnixEpoch
+from danswer.connectors.models import BasicExpertInfo
 from danswer.connectors.models import ConnectorMissingCredentialError
 from danswer.connectors.models import Document
 from danswer.connectors.models import Section
@@ -294,7 +295,9 @@ class ConfluenceConnector(LoadConnector, PollConnector):
                         source=DocumentSource.CONFLUENCE,
                         semantic_identifier=page["title"],
                         doc_updated_at=last_modified,
-                        primary_owners=[author] if author else None,
+                        primary_owners=[BasicExpertInfo(email=author)]
+                        if author
+                        else None,
                         metadata={
                             "Wiki Space Name": self.space,
                         },

backend/danswer/connectors/cross_connector_utils/miscellaneous_utils.py

@@ -0,0 +1,45 @@
+from datetime import datetime
+from datetime import timezone
+
+from dateutil.parser import parse
+
+from danswer.connectors.models import BasicExpertInfo
+from danswer.utils.text_processing import is_valid_email
+
+
+def datetime_to_utc(dt: datetime) -> datetime:
+    if dt.tzinfo is None or dt.tzinfo.utcoffset(dt) is None:
+        dt = dt.replace(tzinfo=timezone.utc)
+
+    return dt.astimezone(timezone.utc)
+
+
+def time_str_to_utc(datetime_str: str) -> datetime:
+    dt = parse(datetime_str)
+    return datetime_to_utc(dt)
+
+
+def basic_expert_info_representation(info: BasicExpertInfo) -> str | None:
+    if info.first_name and info.last_name:
+        return f"{info.first_name} {info.middle_initial} {info.last_name}"
+
+    if info.display_name:
+        return info.display_name
+
+    if info.email and is_valid_email(info.email):
+        return info.email
+
+    if info.first_name:
+        return info.first_name
+
+    return None
+
+
+def get_experts_stores_representations(
+    experts: list[BasicExpertInfo] | None,
+) -> list[str] | None:
+    if not experts:
+        return None
+
+    reps = [basic_expert_info_representation(owner) for owner in experts]
+    return [owner for owner in reps if owner is not None]

backend/danswer/connectors/cross_connector_utils/time_utils.py

@@ -1,16 +0,0 @@
-from datetime import datetime
-from datetime import timezone
-
-from dateutil.parser import parse
-
-
-def datetime_to_utc(dt: datetime) -> datetime:
-    if dt.tzinfo is None or dt.tzinfo.utcoffset(dt) is None:
-        dt = dt.replace(tzinfo=timezone.utc)
-
-    return dt.astimezone(timezone.utc)
-
-
-def time_str_to_utc(datetime_str: str) -> datetime:
-    dt = parse(datetime_str)
-    return datetime_to_utc(dt)

backend/danswer/connectors/file/connector.py

@@ -12,7 +12,7 @@ from danswer.connectors.cross_connector_utils.file_utils import detect_encoding
 from danswer.connectors.cross_connector_utils.file_utils import load_files_from_zip
 from danswer.connectors.cross_connector_utils.file_utils import read_file
 from danswer.connectors.cross_connector_utils.file_utils import read_pdf_file
-from danswer.connectors.cross_connector_utils.time_utils import time_str_to_utc
+from danswer.connectors.cross_connector_utils.miscellaneous_utils import time_str_to_utc
 from danswer.connectors.file.utils import check_file_ext_is_valid
 from danswer.connectors.file.utils import get_file_ext
 from danswer.connectors.interfaces import GenerateDocumentsOutput

backend/danswer/connectors/guru/connector.py

@@ -8,7 +8,7 @@ import requests
 from danswer.configs.app_configs import INDEX_BATCH_SIZE
 from danswer.configs.constants import DocumentSource
 from danswer.connectors.cross_connector_utils.html_utils import parse_html_page_basic
-from danswer.connectors.cross_connector_utils.time_utils import time_str_to_utc
+from danswer.connectors.cross_connector_utils.miscellaneous_utils import time_str_to_utc
 from danswer.connectors.interfaces import GenerateDocumentsOutput
 from danswer.connectors.interfaces import LoadConnector
 from danswer.connectors.interfaces import PollConnector

backend/danswer/connectors/models.py

@@ -27,6 +27,22 @@ class Section(BaseModel):
     link: str | None
 
 
+class BasicExpertInfo(BaseModel):
+    """Basic Information for the owner of a document, any of the fields can be left as None
+    Display fallback goes as follows:
+    - first_name + (optional middle_initial) + last_name
+    - display_name
+    - email
+    - first_name
+    """
+
+    display_name: str | None = None
+    first_name: str | None = None
+    middle_initial: str | None = None
+    last_name: str | None = None
+    email: str | None = None
+
+
 class DocumentBase(BaseModel):
     """Used for Danswer ingestion api, the ID is inferred before use if not provided"""
 
@@ -38,9 +54,9 @@ class DocumentBase(BaseModel):
     # UTC time
     doc_updated_at: datetime | None = None
     # Owner, creator, etc.
-    primary_owners: list[str] | None = None
+    primary_owners: list[BasicExpertInfo] | None = None
     # Assignee, space owner, etc.
-    secondary_owners: list[str] | None = None
+    secondary_owners: list[BasicExpertInfo] | None = None
     # `title` is used when computing best matches for a query
     # if `None`, then we will use the `semantic_identifier` as the title in Vespa
     title: str | None = None

backend/danswer/connectors/productboard/connector.py

@@ -10,7 +10,7 @@ from retry import retry
 
 from danswer.configs.app_configs import INDEX_BATCH_SIZE
 from danswer.configs.constants import DocumentSource
-from danswer.connectors.cross_connector_utils.time_utils import time_str_to_utc
+from danswer.connectors.cross_connector_utils.miscellaneous_utils import time_str_to_utc
 from danswer.connectors.interfaces import GenerateDocumentsOutput
 from danswer.connectors.interfaces import PollConnector
 from danswer.connectors.interfaces import SecondsSinceUnixEpoch

backend/danswer/direct_qa/answer_question.py

@@ -23,7 +23,6 @@ from danswer.direct_qa.interfaces import StreamingError
 from danswer.direct_qa.models import LLMMetricsContainer
 from danswer.direct_qa.qa_utils import get_chunks_for_qa
 from danswer.document_index.factory import get_default_document_index
-from danswer.expert_recommendation.heuristics_based import extract_experts
 from danswer.indexing.models import InferenceChunk
 from danswer.search.models import QueryFlow
 from danswer.search.models import RerankMetricsContainer
@@ -34,7 +33,6 @@ from danswer.search.search_runner import chunks_to_search_docs
 from danswer.search.search_runner import full_chunk_search
 from danswer.search.search_runner import full_chunk_search_generator
 from danswer.secondary_llm_flows.answer_validation import get_answer_validity
-from danswer.server.models import ExpertsResponse
 from danswer.server.models import LLMRelevanceFilterResponse
 from danswer.server.models import NewMessageRequest
 from danswer.server.models import QADocsResponse
@@ -244,10 +242,6 @@ def answer_qa_query_stream(
     # immediately see some results
     top_chunks = cast(list[InferenceChunk], next(search_generator))
 
-    expert_emails = extract_experts(top_chunks)
-    expert_response = ExpertsResponse(experts=expert_emails).dict()
-    yield get_json_line(expert_response)
-
     top_docs = chunks_to_search_docs(top_chunks)
     initial_response = QADocsResponse(
         top_documents=top_docs,

backend/danswer/document_index/vespa/index.py

@@ -47,6 +47,9 @@ from danswer.configs.constants import SOURCE_LINKS
 from danswer.configs.constants import SOURCE_TYPE
 from danswer.configs.constants import TITLE
 from danswer.configs.model_configs import SEARCH_DISTANCE_CUTOFF
+from danswer.connectors.cross_connector_utils.miscellaneous_utils import (
+    get_experts_stores_representations,
+)
 from danswer.document_index.document_index_utils import get_uuid_from_chunk
 from danswer.document_index.interfaces import DocumentIndex
 from danswer.document_index.interfaces import DocumentInsertionRecord
@@ -240,8 +243,8 @@ def _index_vespa_chunk(chunk: DocMetadataAwareIndexChunk) -> None:
         EMBEDDINGS: embeddings_name_vector_map,
         BOOST: DEFAULT_BOOST,
         DOC_UPDATED_AT: _vespa_get_updated_at_attribute(document.doc_updated_at),
-        PRIMARY_OWNERS: document.primary_owners,
+        PRIMARY_OWNERS: get_experts_stores_representations(document.primary_owners),
-        SECONDARY_OWNERS: document.secondary_owners,
+        SECONDARY_OWNERS: get_experts_stores_representations(document.secondary_owners),
         # the only `set` vespa has is `weightedset`, so we have to give each
         # element an arbitrary weight
         ACCESS_CONTROL_LIST: {acl_entry: 1 for acl_entry in chunk.access.to_acl()},

backend/danswer/expert_recommendation/heuristics_based.py

@@ -1,35 +0,0 @@
-from collections import defaultdict
-
-from danswer.indexing.models import InferenceChunk
-from danswer.utils.text_processing import is_valid_email
-
-# What is the minimum cumulative score for a user to be considered an Expert
-# If a set of 50 results is shown, user needs a cumulative doc score of 2.5 to be an expert
-_EXPERT_SCORE_RATIO = 2.5 / 50
-# How much should a score be discounted if the user is not the primary owner
-_SECONDARY_OWNER_DISCOUNT = 0.5
-
-
-def extract_experts(
-    chunks: list[InferenceChunk], score_ratio: float = _EXPERT_SCORE_RATIO
-) -> list[str]:
-    target_score = score_ratio * len(chunks)
-
-    expert_scores: dict[str, float] = defaultdict(float)
-
-    for chunk in chunks:
-        if chunk.primary_owners:
-            for p_owner in chunk.primary_owners:
-                if chunk.score:
-                    expert_scores[p_owner] += chunk.score
-
-        if chunk.secondary_owners:
-            for s_owner in chunk.secondary_owners:
-                if chunk.score:
-                    expert_scores[s_owner] += _SECONDARY_OWNER_DISCOUNT * chunk.score
-
-    return [
-        owner
-        for owner, score in expert_scores.items()
-        if score >= target_score and is_valid_email(owner)
-    ]

backend/danswer/indexing/indexing_pipeline.py

@@ -5,6 +5,9 @@ from typing import Protocol
 from sqlalchemy.orm import Session
 
 from danswer.access.access import get_access_for_documents
+from danswer.connectors.cross_connector_utils.miscellaneous_utils import (
+    get_experts_stores_representations,
+)
 from danswer.connectors.models import Document
 from danswer.connectors.models import IndexAttemptMetadata
 from danswer.db.document import get_documents_by_ids
@@ -50,8 +53,8 @@ def upsert_documents_in_db(
             document_id=doc.id,
             semantic_identifier=doc.semantic_identifier,
             first_link=first_link,
-            primary_owners=doc.primary_owners,
+            primary_owners=get_experts_stores_representations(doc.primary_owners),
-            secondary_owners=doc.secondary_owners,
+            secondary_owners=get_experts_stores_representations(doc.secondary_owners),
             from_ingestion_api=doc.from_ingestion_api,
         )
         doc_m_batch.append(db_doc_metadata)

backend/danswer/server/models.py

@@ -202,10 +202,6 @@ class SearchFeedbackRequest(BaseModel):
     search_feedback: SearchFeedbackType
 
 
-class ExpertsResponse(BaseModel):
-    experts: list[str]
-
-
 class RetrievalDocs(BaseModel):
     top_documents: list[SearchDoc]