trustgraph/docs/tech-specs/extraction-provenance-subgraph.sw.md

Asili ya Utoaji: Mfumo wa Subgraph

Tatizo

Hivi sasa, utoaji wa wakati wa uondoaji huunda uelekezaji kamili kwa kila triple iliyoundwa: stmt_uri, activity_uri, na metadata inayohusiana ya PROV-O kwa kila ukweli wa maarifa. Kushughulikia sehemu moja ambayo hutoa uhusiano wa 20 hutoa triples ~220 za utoaji pamoja na triples ~20 za maarifa - mzigo wa takriban 10:1.

Hii ni ghali (uhifadhi, uwekaji wa indexi, usambazaji) na pia si sahihi kimaana. Kila sehemu hushughulikiwa na simu moja ya LLM ambayo hutoa triples zake zote katika mshono mmoja. Mfumo wa sasa wa kila triple huficha hili kwa kuunda udanganyifu wa matukio 20 ya uondoaji huru.

Zaidi ya hayo, vichakavu viwili vya nne vya uondoaji (kg-extract-ontology, kg-extract-agent) havina utoaji wowote, na hivyo kuacha pengo katika njia ya ukaguzi.

Suluhisho

Badilisha uelekezaji wa kila triple na mfumo wa subgraph: rekodi moja ya utoaji kwa kila uondoaji wa sehemu, inayoshirikiwa na triples zote zilizozalishwa kutoka sehemu hiyo.

Mabadiliko ya Dhana

Zamani	Mpya
stmt_uri (https://trustgraph.ai/stmt/{uuid})	subgraph_uri (https://trustgraph.ai/subgraph/{uuid})
statement_uri()	subgraph_uri()
tg:reifies (1:1, utambulisho)	tg:contains (1:wengi, uwezeshaji)

Muundo Unaolengwa

Triples zote za utoaji huwekwa katika grafu iliyoitwa urn:graph:source.

# Subgraph contains each extracted triple (RDF-star quoted triples)
<subgraph> tg:contains <<s1 p1 o1>> .
<subgraph> tg:contains <<s2 p2 o2>> .
<subgraph> tg:contains <<s3 p3 o3>> .

# Derivation from source chunk
<subgraph> prov:wasDerivedFrom <chunk_uri> .
<subgraph> prov:wasGeneratedBy <activity> .

# Activity: one per chunk extraction
<activity> rdf:type          prov:Activity .
<activity> rdfs:label        "{component_name} extraction" .
<activity> prov:used         <chunk_uri> .
<activity> prov:wasAssociatedWith <agent> .
<activity> prov:startedAtTime "2026-03-13T10:00:00Z" .
<activity> tg:componentVersion "0.25.0" .
<activity> tg:llmModel       "gpt-4" .          # if available
<activity> tg:ontology        <ontology_uri> .   # if available

# Agent: stable per component
<agent> rdf:type   prov:Agent .
<agent> rdfs:label "{component_name}" .

Kulinganisha Kiasi

Kwa kila sehemu inayozalisha triples tatu zilizochukuliwa:

	Zamani (kwa kila triple)	Mpya (subgraph)
tg:contains / tg:reifies	N	N
Triples za shughuli	~9 x N	~9
Triples za wakala	2 x N	2
Metadata ya taarifa/subgraph	2 x N	2
Triples tatu za jumla za asili	~13N	N + 13
Mfano (N=20)	~260	33

Upeo

Wasindikaji ambao Watasasishwa (asili iliyopo, kwa kila triple)

kg-extract-definitions (trustgraph-flow/trustgraph/extract/kg/definitions/extract.py)

Hivi sasa huita statement_uri() + triple_provenance_triples() ndani ya loop ya kila ufafanuzi.

Mabadiliko: Hamisha subgraph_uri() na activity_uri() kabla ya loop Kusanya triples za tg:contains ndani ya loop Toa kundi la shughuli/wakala/uzalishaji mara moja baada ya loop

kg-extract-relationships (trustgraph-flow/trustgraph/extract/kg/relationships/extract.py)

Mfano sawa na ufafanuzi. Mabadiliko sawa.

Wasindikaji ambao Watasasishwa ili Kuongeza Asili (sasa hayapo)

kg-extract-ontology (trustgraph-flow/trustgraph/extract/kg/ontology/extract.py)

Hivi sasa hutoa triples bila asili. Ongeza asili ya subgraph kwa kutumia mfano sawa: subgraph moja kwa kila sehemu, tg:contains kwa kila triple iliyochukuliwa.

kg-extract-agent (trustgraph-flow/trustgraph/extract/kg/agent/extract.py)

Hivi sasa hutoa triples bila asili. Ongeza asili ya subgraph kwa kutumia mfano sawa.

Mabadiliko ya Maktaba ya Asili iliyoshirikiwa

trustgraph-base/trustgraph/provenance/triples.py

Badilisha triple_provenance_triples() na subgraph_provenance_triples() Kazi mpya inakubali orodha ya triples zilizochukuliwa badala ya moja Inazalisha tg:contains moja kwa kila triple, kundi la shughuli/wakala lililoshirikiwa Ondoa triple_provenance_triples() ya zamani

trustgraph-base/trustgraph/provenance/uris.py

Badilisha statement_uri() na subgraph_uri()

trustgraph-base/trustgraph/provenance/namespaces.py

Badilisha TG_REIFIES na TG_CONTAINS

Hayajajumuishwa katika Upeo

kg-extract-topics: wasindikaji wa mtindo wa zamani, hawatumiki kwa sasa katika mtiririko wa kawaida kg-extract-rows: hutoa mistari si triples, mfumo tofauti wa asili Asili ya wakati wa swali (urn:graph:retrieval): suala tofauti, tayari hutumia mfumo tofauti (swali/uchunguzi/lengo/muhtasari) Asili ya hati/ukurasa/sehemu (dekoda ya PDF, kichunguzi): tayari hutumia derived_entity_triples() ambayo ni kwa kila kitu, si kwa kila triple — hakuna suala la ziada

Maelezo ya Utendaji

Upangaji Upya wa Loop ya Msindikaji

Kabla (kwa kila triple, katika uhusiano):

for rel in rels:
    # ... build relationship_triple ...
    stmt_uri = statement_uri()
    prov_triples = triple_provenance_triples(
        stmt_uri=stmt_uri,
        extracted_triple=relationship_triple,
        ...
    )
    triples.extend(set_graph(prov_triples, GRAPH_SOURCE))

Baada ya (mfumo mdogo):

sg_uri = subgraph_uri()

for rel in rels:
    # ... build relationship_triple ...
    extracted_triples.append(relationship_triple)

prov_triples = subgraph_provenance_triples(
    subgraph_uri=sg_uri,
    extracted_triples=extracted_triples,
    chunk_uri=chunk_uri,
    component_name=default_ident,
    component_version=COMPONENT_VERSION,
    llm_model=llm_model,
    ontology_uri=ontology_uri,
)
triples.extend(set_graph(prov_triples, GRAPH_SOURCE))

Saini Mpya ya Msaidizi

def subgraph_provenance_triples(
    subgraph_uri: str,
    extracted_triples: List[Triple],
    chunk_uri: str,
    component_name: str,
    component_version: str,
    llm_model: Optional[str] = None,
    ontology_uri: Optional[str] = None,
    timestamp: Optional[str] = None,
) -> List[Triple]:
    """
    Build provenance triples for a subgraph of extracted knowledge.

    Creates:
    - tg:contains link for each extracted triple (RDF-star quoted)
    - One prov:wasDerivedFrom link to source chunk
    - One activity with agent metadata
    """

Mabadiliko Makubwa

Hii ni mabadiliko makubwa kwa mfumo wa asili ya data. Asili ya data haijatolewa, kwa hivyo hakuna uhamishaji unaohitajika. Msimbo wa zamani wa ⟦CODE_0⟧ / tg:reifies unaweza kuondolewa kabisa. Msimbo statement_uri unaweza kufutwa kabisa.