[Issue #74] Add opt-in "enhanced" strategy for schema-based graph extraction - #78

Open

YanhaoLv wants to merge 20 commits into

hugegraph:mainfrom

YanhaoLv:issue74-schema-based-extraction

Commits on Jul 7, 2026

feat(llm): add extract_strategy/include_debug plumbing (default baseline)
YanhaoLv
committed
948052b
feat(llm): introduce GraphSchemaIndex for enhanced extraction
YanhaoLv
committed
968ede7
feat(llm): add StructuredWarning registry for enhanced extraction
YanhaoLv
committed
946185c
feat(llm): add CandidateGraphParser for enhanced extraction
YanhaoLv
committed
2396ce3
feat(llm): add SchemaAwareNormalizer for chunk-level graph normalization
YanhaoLv
committed
9f07e7c
feat(llm): add DocumentGraphAssembler for cross-chunk graph assembly
YanhaoLv
committed
dfbb9d3
feat(llm): add GraphQualityGate metrics for enhanced extraction
YanhaoLv
committed
eb71d62
feat(llm): wire enhanced strategy into PropertyGraphExtract
YanhaoLv
committed
81f1dea
feat(llm): add GraphExtractionEvaluator + baseline-vs-enhanced benchmark
YanhaoLv
committed
d26cbc6
test(llm): expand benchmark to 14 scenarios + latency/call metrics; add usage doc
YanhaoLv
committed
6cfc898
feat(llm): add live DeepSeek benchmark driver + effect-report live section
YanhaoLv
committed
126d4ac
test(llm): add s15/s16 domain-limit scenarios; bake +5% design threshold
YanhaoLv
committed
f304ac4
feat(llm): add public actor-corpus builder (Wikipedia + Wikidata Entity API)
YanhaoLv
committed
10ee00a
test(llm): pin public actor corpus (8 actors, 65 vertices, 57 edges)
YanhaoLv
committed
6e4cd29
feat(llm): live benchmark supports --corpus and --runs with mean/std aggregation
YanhaoLv
committed
ef985ed
docs(llm): replace hand-authored live results with public-corpus benchmark
YanhaoLv
committed
2a4c3c7
scripts(llm): drop legacy hand-authored fallback from live benchmark driver
YanhaoLv
committed
8afc32c
docs(llm): commit raw live-benchmark archive for reviewer cross-checking
YanhaoLv
committed
336d6c0
docs(llm): expand effect report — how-to-reproduce section + +4.8% rationale
YanhaoLv
committed
b601bca
docs(llm): drop bundled live archive JSON (available on request)
YanhaoLv
committed
c9b51b3

Navigation Menu

[Issue #74] Add opt-in "enhanced" strategy for schema-based graph extraction - #78

[Issue #74] Add opt-in "enhanced" strategy for schema-based graph extraction#78
YanhaoLv wants to merge 20 commits into
hugegraph:mainfrom
YanhaoLv:issue74-schema-based-extraction

Commits on Jul 7, 2026

feat(llm): add extract_strategy/include_debug plumbing (default baseline)

feat(llm): introduce GraphSchemaIndex for enhanced extraction

feat(llm): add StructuredWarning registry for enhanced extraction

feat(llm): add CandidateGraphParser for enhanced extraction

feat(llm): add SchemaAwareNormalizer for chunk-level graph normalization

feat(llm): add DocumentGraphAssembler for cross-chunk graph assembly

feat(llm): add GraphQualityGate metrics for enhanced extraction

feat(llm): wire enhanced strategy into PropertyGraphExtract

feat(llm): add GraphExtractionEvaluator + baseline-vs-enhanced benchmark

test(llm): expand benchmark to 14 scenarios + latency/call metrics; add usage doc

feat(llm): add live DeepSeek benchmark driver + effect-report live section

test(llm): add s15/s16 domain-limit scenarios; bake +5% design threshold

feat(llm): add public actor-corpus builder (Wikipedia + Wikidata Entity API)

test(llm): pin public actor corpus (8 actors, 65 vertices, 57 edges)

feat(llm): live benchmark supports --corpus and --runs with mean/std aggregation

docs(llm): replace hand-authored live results with public-corpus benchmark

scripts(llm): drop legacy hand-authored fallback from live benchmark driver

docs(llm): commit raw live-benchmark archive for reviewer cross-checking

docs(llm): expand effect report — how-to-reproduce section + +4.8% rationale

docs(llm): drop bundled live archive JSON (available on request)