| # Source code notebooks |
|
|
| This folder contains the notebooks used to construct and analyze Community Notes BR. |
| They are provided as reproducibility and inspection material for the public dataset. |
|
|
| The notebooks were originally developed in Google Colab/local notebook sessions. |
| Outputs and execution counters were stripped before publication, and embedded |
| credentials were removed. To rerun the full pipeline, users must configure their |
| own credentials where needed and have access to the upstream Community Notes |
| public data, optional X API hydration, and GPU resources for the embedding/topic |
| modeling/NER stages. |
|
|
| ## Notebook order |
|
|
| 1. `01_etl_snapshot.ipynb` |
| Downloads/organizes the upstream Community Notes snapshot and materializes |
| typed analytical tables. |
|
|
| 2. `02_portuguese_filter.ipynb` |
| Builds the Portuguese subset using language identification and joins ratings |
| and latest note status. |
|
|
| 3. `03_topic_modeling_macrothemes.ipynb` |
| Generates semantic embeddings, topic clusters, LLM-constrained labels, and |
| macrothemes. |
|
|
| 4. `04_named_entity_recognition.ipynb` |
| Extracts named entities with GLiNER, regex extractors, canonical dictionaries, |
| and overlap resolution. |
|
|
| 5. `05_matrix_factorization_scoring.ipynb` |
| Reproduces the Community Notes matrix-factorization scoring fields over the |
| Portuguese subset. |
|
|
| 6. `06_helpfulness_classification.ipynb` |
| Contains exploratory predictive modeling and interpretability experiments |
| over note helpfulness/consensus. |
|
|
| 7. `07_udpipe_bosque_enrichment_colab.ipynb` |
| Optional Google Colab notebook that creates an automatic Universal |
| Dependencies syntax layer with UDPipe Bosque and uploads it to |
| `syntax_ud_bosque/` in the existing Hugging Face dataset repository. |
|
|
| ## Notes |
|
|
| - Tweet text is not redistributed. Only tweet identifiers appear in the dataset. |
| - The notebooks are intended as transparent research code, not as a packaged |
| one-command production pipeline. |
| - The released Parquet files are the authoritative dataset artifacts. |
| - Optional derived layers generated by external models may have separate |
| license metadata from the CC0 core dataset. |
|
|
