SSH Citation Index Modules

Research applicability

AI modules for extracting, parsing and disambiguating bibliographic references in social-science and humanities publications.

Getting started

Read the work-in-progress warning, configure the selected extraction backend and LLM, and follow the linked PDF-to-linked-reference notebook on test documents. These are documentation-based editorial adoption suggestions, not an exercised workflow.

Maintenance

README labels the project work in progress with changing APIs and experimental/incomplete components.

License

Repository metadata identifies MIT; license/dependency obligations not audited.

Workflow & human oversight

A defined extraction/parsing/linking workflow, not autonomous scholarship. Researchers must inspect extraction completeness, bibliographic matches and ambiguous links.

Workflow level is not an evidence score; higher is not better evidence.

Approach

Documented staged API extracts PDF text and references, parses them with LLM assistance and links records through scholarly connectors; jobs are polled separately.

Limitations

README explicitly discourages production use while under development. Citation-intent classification and entity extraction are TODO, not delivered capabilities. Benchmark directories and mocked linking tests do not establish accuracy; no benchmark or live workflow was reproduced. Documented requirements.txt install command should be checked against checkout because the retrieved root file listing does not show that file.

Reproducibility

Primary repository documentation checked live. No installation, software execution, security audit, or outcome replication performed.

Adoption requirements

Python 3.10+ for development; documented API deployment requires Docker services, Redis/workers, an OpenAI-compatible vLLM endpoint and external MinerU service when used.