Reference documentation for DataFlow-KG core operators (general / temporal / multimodal / hyper-relation KG and the path & subgraph samplers shared across them). Use when generating a pipeline that needs per-operator details beyond the planner summary, or when debugging field-flow / row-expansion issues.
Reference for KGEntityBasedSubgraphSampling. Samples a textual subgraph around each entity using BFS, k-hop neighborhood, or random walk. Row-expanding. Use when: a downstream QA generator needs local KG context per entity.
Reference for KGRelationTuplePathGenerator. Enumerates k-hop undirected paths from any triple-format column. Pure-graph, no LLM. Row-expanding. Use when: feeding a path-based QA generator (general / temporal / hyper-relation).
Reference for HRKGRelationTriplePathQAGeneration. LLM-generates QA pairs over hyper-tuples (hop=1) or sampled hyper-relation paths (hop=2). Use when: producing QA datasets that exercise multi-argument facts.
Reference for HRKGTripleExtraction. Extracts hyper-relation tuples carrying multiple qualifier/attribute slots per fact. Use when: text contains multi-argument facts where a single (subject, relation, object) is insufficient.
Reference documentation for the KGEntityExtraction operator. Extracts entity surface forms from raw text using an LLM. Use when: every KG pipeline needs an entity column before triple/quadruple extraction; this operator produces it from a free-text input column.
Reference for KGRelationTripleInference. Uses an LLM to infer implicit relation triples from an existing triple list and (optionally) the source text; can merge results back into the input column. Use when: extending the knowledge graph closure beyond what was directly stated in the text.
Reference for KGRelationTriplePathQAGeneration. LLM-generates QA pairs over 1-hop triples or 2-hop paths from KGRelationTuplePathGenerator. Use when: building reasoning-flavored QA grounded in a path-shaped KG context.
Reference for KGRelationTripleSubgraphQAGeneration. LLM-generates QA pairs grounded in a sampled subgraph column. Use when: a subgraph sampler has emitted local KG context per row and QA data is needed for training or evaluation.
Reference documentation for the KGTripleExtraction operator. Extracts relation or attribute triples from text given a candidate entity list. Use when: after KGEntityExtraction has produced the entity column and the pipeline needs structured triples.
Reference for MMKGSubgraphBaseQAGeneration. VLM-generated QA pairs grounded in a textual subgraph plus its aligned vistriple and visurl. Use when: producing multimodal QA from MMKGEntityBasedSubgraphSampling output.
Reference for MMKGVisualTripleExtraction. Uses a VLM to extract visual triples linking textual entities to images in imgdict. Use when: building a multimodal KG that grounds entities in local image evidence.
Reference for TKGTupleExtraction. Extracts time-anchored 4-tuples (relation or attribute flavors) from text. Use when: the input text carries explicit timestamps / dates / time spans and the pipeline needs a temporal KG.
Reference for TKGTuplePathQAGeneration. LLM-generates time-aware QA pairs over temporal tuples (hop=1) or sampled temporal paths (hop>1). Use when: building time-anchored QA datasets — when did X happen, ordering, intervals, etc.
Reference documentation for the KGTupleNormalization operator. LLM-canonicalizes synonymous relations/attributes and de-duplicates triples in-place. Use when: after KGTripleExtraction or KGRelationTripleInference produces raw triples that may contain synonyms or directional inconsistencies.
A development context loader for the DataFlow-KG code repository. It covers knowledge-graph operators, pipelines, prompts, error diagnosis, code reviews, and knowledge-base updates.
Build production-grade DataFlow-KG operator scaffolds for generate/filter/eval/refine workflows, including operator file, module registration, and baseline tests. Trigger when users ask to create/new/scaffold KG operators or patch TYPECHECKING registration for DataFlow-KG.
Build reusable DataFlow-KG DIYPromptABC prompttemplate classes for existing KG operators. Generates both buildprompt() and buildsystemprompt() methods aligned to operator interface contracts, with two-round AskUserQuestion intake, two-stage auditable outputs, and KG-specific static acceptance checks. Trigger when…
★not rated 2 3mo agoA119 tokens
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: