Agriculture_KnowledgeGraph

Use this repo skill when a task involves the Agriculture_KnowledgeGraph / AgriKG research demo: agricultural entity data, Neo4j graph imports and queries, the Django web demo, THULAC/KNN labels, Hudong/Wikidata/weather crawlers, or remote-supervised relation extraction.

Start by routing the task

• Graph data, Neo4j import/query, CSV schemas, hierarchy tree, or vector utilities: use graph-query-and-data-management.
• Django demo startup, routes, forms, QA, relation search pages, tagging pages, or preload side effects: use web-app-service.
• THULAC entity recognition, label ids 0-16, predicted label files, manual labels, fastText/KNN classifier prerequisites: use entity-labeling-and-ner.
• Hudong/Baike crawlers, DFS tree crawling, Wikidata property/entity/relation crawlers, weather/attribute pipelines, or generated relation CSV validation: use crawlers-and-wikidata-pipelines.
• Wikidata/Wikipedia sentence alignment, relation dataset TSV/JSON creation, deduplication, Fire preprocessing commands, or TensorFlow PCNN training: use relation-extraction-pipeline.

Read shared references when needed

• architecture-and-workflows.md maps the repository components and task order.
• installation-and-environment.md explains the un-packaged source layout, old Python/Django stack, optional services, large assets, and dependency choices.
• troubleshooting.md gives cross-cutting triage and routes symptoms to the right sub-skill.
• repo-provenance.md records the source snapshot and evidence paths used to create this skill; read it before deciding whether a checkout needs refresh.

Safe root preflight

Run the bundled checker before broad debugging:

python scripts/check_agri_kg_environment.py --help
python scripts/check_agri_kg_environment.py --repo-root /path/to/Agriculture_KnowledgeGraph

The checker is non-destructive. It imports optional packages, checks expected files when a checkout path is supplied, and can optionally probe local Neo4j/MongoDB sockets. It does not start services, crawl the network, download fastText vectors, connect with credentials, load large models, or train TensorFlow models.

Repository operating assumptions

• The repo is a legacy source checkout, not a packaged Python distribution. Use workflow-specific dependencies and working directories rather than expecting pip install -e ..
• Python 3.7 with Django 1.11-era dependencies is the safest legacy starting point; modern Python may require compatibility patches.
• Neo4j, MongoDB, network crawls, large fastText/vector files, and TensorFlow PCNN training are external prerequisites. Treat them as explicit user-approved steps, not first-line smoke checks.
• Many source modules are path-sensitive or eager at import time. Prefer bundled validators and references before importing modules that start service connections or load large files.
• Do not claim a live graph import, web app, crawl, KNN prediction, or PCNN training run passed unless that exact workflow ran in the active environment.

Quick task examples

• “Import the AgriKG CSVs into Neo4j” → graph-query-and-data-management, then its Cypher templates.
• “Django route fails before the first page loads” → web-app-service, especially preload troubleshooting.
• “Validate predict_labels.txt or explain label 6” → entity-labeling-and-ner.
• “Regenerate or validate wikidata_relation2.csv” → crawlers-and-wikidata-pipelines.
• “Convert aligned relation sentences into train/test JSON” → relation-extraction-pipeline.