Logistics¶
The project is a group work. Each milestone is submitted as a group via Moodle:
Shared code repository: one public code repository per group serves the whole project, with each milestone’s code committed to it cumulatively so the commit history documents the group’s progress. Milestones 1 and 2 paste its link via Moodle together with the report link, and Milestone 3 submits it together with the poster and the dataset.
Jupyter Book for Milestones 1 and 2: the milestone reports form one public Jupyter Book, hosted on its own separate repository’s GitHub Pages. Milestone 1’s report is the book’s first section and Milestone 2 adds the next section, and report figures and tables must be generated during book compilation.
Final data submission: there is no intermediate data submission. The fully processed data is hosted as a Hugging Face dataset of sharded Parquet tables, with the table set specified across the milestone statements, papers and references in Milestone 1, elements, hierarchy and bibtex in Milestone 2, and predictions in Milestone 3, while hosting follows the instructions given by the Lab Instructors. The dataset link is submitted via Moodle.
Final grading: a poster session. Each group submits one pptx file containing a single-page poster in the form factor of portrait A2 for review, and upon acceptance the group prints the poster and presents it at the poster session.
Milestones¶
Project Milestone 1 — Data Crawling: build the raw corpus, discover the assigned arXiv papers in the range provided by the Lab Instructors, crawl every version of each publication with its full-text LaTeX sources, metadata, and reference metadata, and analyze the scraped corpus.
Project Milestone 2 — Hierarchical Parsing and Standardization: structure the corpus, parse each version into a hierarchical tree, standardize the content, convert
\\bibitemreferences into BibTeX, deduplicate across versions, and analyze the processed data.Project Milestone 3 — Reference Matching Pipeline: run the machine learning pipeline, clean, label, engineer features, model, and evaluate a reference-matching system scored with Mean Reciprocal Rank 5, written MRR@5.
Milestone-specific requirements and submission formats are detailed in each milestone page.
Regulations¶
Ensure your code is well-documented with clear comments. You can use online documents for reference, but you must provide complete citations. You are free to discuss your topics with classmates, but your work must be implemented and interpreted according to your own understanding.
Any milestone will be graded zero if there is any dishonest behaviour detected.