Skip to content

Class 2 Homework — Data Collection & Extraction (geomingical) - #18

Open
geomingical wants to merge 1 commit into
inference-ai-course:mainfrom
geomingical:geomingical/class2-homework
Open

geomingical wants to merge 1 commit into
inference-ai-course:mainfrom
geomingical:geomingical/class2-homework

Conversation

@geomingical

Copy link
Copy Markdown

Summary

Submission for Week 2 homework: Data Collection & Extraction.

Task 1 and Task 2 were adapted to fit a seismology research workflow instead of the arXiv/Tesseract default approach. Core skills are the same; tools and corpus differ.

  • Task 1: NHESS web scraper demo (arxiv_clean.json) + MinerU-based metadata extraction from 32-paper seismology corpus (papers_metadata.json)
  • Task 2: MinerU (layout-aware PDF parser) + PyMuPDF (native image extraction) replaces Tesseract OCR — produces clean structured Markdown, JSON layout, and native-quality figures. Pipeline documented with Mermaid flowchart (pipeline_overview.md). Sample output included for chang-2023.
  • Task 3: Whisper ASR transcription with timestamps — 1 representative sample (talks_transcripts.jsonl)
  • Task 4: N/A — MinerU parses PDF structure directly, producing clean text with no OCR artifacts or HTML noise. The 32-paper corpus is pre-selected and unique, so deduplication has no effect. Task 4 is resolved upstream by the tool choice.

Key files

Task File
Task 1 class2/hw_output/Paper_Abstract/papers_metadata.json
Task 1 demo class2/hw_output/Paper_Abstract/arxiv_clean.json
Task 2 class2/hw_output/pdf_ocr/pipeline_overview.md
Task 2 sample class2/hw_output/pdf_ocr/sample/chang-2023/
Task 3 class2/hw_output/yt_transcription/talks_transcripts.jsonl

See class2/Class 2 Homework.ipynb Cell 8 for full submission summary and modification rationale.

🤖 Generated with Claude Code

Personal modifications to Task 1 & 2 to fit seismology research workflow:

Task 1 (Paper Metadata):
- NHESS web scraper demo (arxiv_clean.json, same format as arXiv requirement)
- MinerU-based metadata extraction from 32-paper seismology corpus (papers_metadata.json)

Task 2 (PDF → Text + Images):
- MinerU layout-aware parser replaces Tesseract OCR for text-based PDFs
- PyMuPDF native image extraction replaces screenshot-based OCR
- Pipeline documented in pipeline_overview.md (Mermaid flowchart)
- Sample output included: chang-2023 (.md, content_list.json, 13 native images)
- Batch processing script: batch_process.py

Task 3 (ASR):
- Whisper transcription with timestamps (1 representative sample)

Task 4: N/A — MinerU produces clean structured output upstream,
eliminating the need for HTML cleaning, dedup, and PII removal
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant