Who Speaks Matters: Authority-Aware Multi-View RAG over Italian Parliamentary Proceedings
ParliamentRAG is a RAG system for the Italian Chamber of Deputies that uses a topic-dependent authority model to weight speakers by profession, education, and previous interventions. It retrieves relevant speech chunks, identifies topic-relevant experts across parliamentary groups, and generates a summary with supporting quotations. It was evaluated against Google NotebookLM on 15 policy topics using automated metrics and blind A/B human evaluation by six domain experts.
Parliamentary proceedings are a primary record of democratic deliberation, yet their volume and fragmentation make multi-perspective access difficult. Applying RAG to parliamentary transcripts introduces risks: dominance of frequent speakers, inability to weight speakers by topical expertise, and citation misattribution. ParliamentRAG addresses these risks jointly with a topic-dependent authority model that estimates each speaker's authority as a function of the current query, combining interpretable components such as profession, education, and previous interventions. Given a user query, the system retrieves relevant speech chunks, identifies topic-relevant experts across parliamentary groups, and generates a summary synthesizing their perspectives, accompanied by supporting quotations. ParliamentRAG is evaluated against Google NotebookLM on 15 policy topics via a two-level protocol combining automated metrics and blind A/B human evaluation by six domain experts.
The core technical contribution is a topic-dependent authority model that estimates speaker authority as a function of the current query, combining interpretable components such as profession, education, and previous interventions. This model is integrated into a multi-view RAG pipeline that retrieves speech chunks, identifies topic-relevant experts across parliamentary groups, and generates a synthesized summary with supporting quotations. The evaluation protocol combines automated metrics and blind A/B human evaluation by six domain experts across 15 policy topics, comparing against Google NotebookLM.
This work demonstrates a practical approach to making large, fragmented public records more accessible through authority-aware retrieval and generation. It highlights the importance of mitigating speaker dominance and citation misattribution in politically sensitive text, which is relevant for applications in journalism, civic technology, and government transparency. The comparison with Google NotebookLM suggests a growing market for specialized RAG systems tailored to institutional data.
ParliamentRAG offers value to citizens, journalists, and researchers by enabling multi-perspective access to parliamentary proceedings with authority-weighted summaries and reliable citations. For institutions, it could reduce the cost of manual analysis and improve transparency. For AI vendors, it demonstrates a specialized RAG application that could be productized for government and civic technology markets.
Observable next signals include publication of detailed evaluation results, potential open-sourcing of the ParliamentRAG system, and adoption or pilot studies by other parliamentary bodies or public institutions. Further research may extend the authority model to other domains such as legal proceedings, corporate meetings, or academic discourse.