Zakherwwf/hwc-north-africa-evidence-map

Automated evidence mapping pipeline and replication archive of human–wildlife conflict research in North Africa # Automated Evidence Mapping of Human–Wildlife Conflict & Coexistence Research in North Africa -orange.svg) This repository contains the complete codebase, data assets, execution logs, and replication materials for an end-to-end **automated systematic evidence map** examining human–wildlife conflict and coexistence literature across Algeria, Egypt, Libya, Morocco, and Tunisia. All screening decisions and structured data extractions were conducted using a local multi-model Large Language Model (LLM) ensemble validated against external gold-standard benchmarks, with every extracted variable grounded in verbatim source quotes. --- ## Table of Contents 1. Project Overview 2. Repository Structure 3. Pipeline Architecture & Workflow 4. Replication Guide - Option A: Quick Analytical Reproduction (From Data) - Option B: Full End-to-End Execution (From Scratch) 5. Adapting to Other Research Questions 6. Data Dictionary & Claim Verification 7. Methodological Disclosures & Boundaries 8. Hardware & Compute Environment --- ## 1. Project Overview Evidence synthesis in conservation science often suffers from geographic and linguistic biases, high manual labor costs, and unmeasured reviewer subjectivity. This project establishes an automated, fully reproducible framework for systematic evidence mapping: * **Corpus Scale:** 95,546 candidate records retrieved across a multi-source, multi-language, multi-country search matrix (OpenAlex, CORE, Crossref). * **Deduplication & Screening:** Algorithmic deduplication followed by a local two-model LLM ensemble (`gemma4:e4b` and `llama3:latest`) operating under a recall-maximizing union rule, pre-calibrated against external benchmarks. * **Full-Text Processing:** Automated open-access retrieval and structured XML conversion via GROBID. * **Quote-Grounded Extraction:** 20 structured fields extracted per study, requiring exact verbatim string matching against source text to guarantee traceability. * **Robustness Auditing:** Seven p …