emeierkeio/parliamentrag-camera-leg19

ParliamentRAG — Italian Chamber of Deputies, 19th legislature

Full proceedings of the Italian Chamber of Deputies (Camera dei deputati) for the 19th legislature, from the first sitting on 13 October 2022 through 6 August 2026: verbatim speech transcripts, roll-call votes with every individual ballot, parliamentary acts with EuroVoc subjects, and the deputies' group, committee and government memberships over time. The dataset is refreshed as new sittings are ingested.

The tables are the flattened form of the ParliamentRAG knowledge graph, built from the official open data of the Chamber (dati.camera.it) and the stenographic reports of each sitting. The same graph powers the live system at parliamentrag.it and the paper Who Speaks Matters: Authority-Aware Multi-View RAG over Italian Parliamentary Proceedings (arXiv:2608.13410). An RDF version of the graph is archived on Zenodo: 10.5281/zenodo.21560331.

All content is in Italian, except the session and debate recaps, which exist in both Italian and English.

Tables

Row identifiers are shared across tables: session_id (e.g. leg19_sed1), vote_id, speech_id, person_id (the dati.camera.it person URI) and act_uri join the tables together. Speeches are ordered as delivered (session date, debate order, phase order).

Individual ballot outcomes in individual_votes are favor, against, abstain, absent. For secret ballots (secret_vote = true in votes) only participation is recorded (absent, abstain): who voted for or against is not public, so those votes carry aggregate numbers only.

Example

from datasets import load_dataset

speeches = load_dataset("emeierkeio/parliamentrag-camera-leg19", "speeches", split="train")
votes = load_dataset("emeierkeio/parliamentrag-camera-leg19", "votes", split="train")

Provenance and construction

Sources:

• —dati.camera.it SPARQL endpoint (OCD ontology): deputies, mandates, groups, committees, acts, votes — CC BY-SA 4.0.
• —Stenographic reports of the Chamber (XML): speech transcripts, debate structure.
• —EuroVoc: multilingual subject thesaurus of the EU.

The build pipeline (download, parsing, entity resolution, NER linking) is open source: github.com/Emeierkeio/ParliamentRAG. This export is produced by build/export_hf.py in that repository. Session and debate recaps are LLM-generated summaries and are marked as such; everything else comes from the official record.

Known limits:

• —Chamber of Deputies only; Senate sittings are not included.
• —444 acts are placeholders (is_placeholder = true): they are referenced by debates or votes but not yet resolved to their full record (mostly number-format variants and committee versions of bills; a small share of chamber documents has no machine-readable record at the source).
• —Speaker attribution is missing for 44 speeches where the stenographic report does not identify the speaker.
• —Recaps are generated for sessions ingested after summarization was introduced (mid-2026): 60 of 705 sessions and 222 of 7,080 debates have them; the rest have null recaps.

License and attribution

CC BY-SA 4.0, inherited from the source data of the Camera dei deputati (ShareAlike: derivatives of this dataset must keep the same license). If you use this dataset, attribute both the Chamber (source data) and ParliamentRAG (derived dataset), and cite:

@inproceedings{tritella2026parliamentrag,
  title     = {Who Speaks Matters: Authority-Aware Multi-View RAG over Italian Parliamentary Proceedings},
  author    = {Tritella, Mirko and Pozzi, Riccardo and Palmonari, Matteo},
  booktitle = {The Semantic Web -- ISWC 2026},
  year      = {2026}
}