Who Speaks Matters: Authority-Aware Multi-View RAG...

Who Speaks Matters: Authority-Aware Multi-View RAG... | AI Research

Key Takeaways

Who Speaks Matters: Authority-Aware Multi-View RAG over Italian Parliamentary Proceedings introduces ParliamentRAG, a system designed to improve how users ac...
Parliamentary proceedings are a primary record of democratic deliberation, yet their volume and fragmentation make multi-perspective access difficult for citizens, journalists, and researchers.
We present ParliamentRAG, a RAG system for the Italian Chamber of Deputies that addresses these risks jointly.
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 system achieves higher coverage across political groups (0.97 vs.

Paper AbstractExpand

Parliamentary proceedings are a primary record of democratic deliberation, yet their volume and fragmentation make multi-perspective access difficult for citizens, journalists, and researchers. Applying Retrieval-Augmented Generation (RAG) to parliamentary transcripts introduces three specific risks: dominance of the most frequent speakers, inability to weight speakers according to topical expertise, and citation misattribution in politically sensitive text. We present ParliamentRAG, a RAG system for the Italian Chamber of Deputies that addresses these risks jointly. Its core contribution is 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 system achieves higher coverage across political groups (0.97 vs. 0.95), perfect quotation faithfulness (1.00 vs. 0.95), and stronger expert preferences on source-related dimensions, while NotebookLM remains stronger on prose-oriented dimensions.

Who Speaks Matters: Authority-Aware Multi-View RAG over Italian Parliamentary Proceedings introduces ParliamentRAG, a system designed to improve how users access and synthesize information from Italian parliamentary transcripts. The researchers address the tendency of standard Retrieval-Augmented Generation (RAG) systems to favor frequent speakers, ignore topical expertise, and misattribute quotes in sensitive political contexts.

Addressing RAG Limitations in Politics

Standard RAG systems often struggle with the specific nature of parliamentary data, which is fragmented and voluminous. The authors identify three primary risks when applying these systems to political records: the dominance of speakers who appear most frequently, the failure to account for a speaker's specific expertise on a topic, and the potential for inaccurate citation in politically sensitive discussions. ParliamentRAG aims to mitigate these issues by integrating speaker authority into the retrieval and generation process.

The Authority-Aware Model

The core of the system is a topic-dependent authority model. Instead of treating all speech chunks equally, the system calculates a speaker's authority based on the specific query provided by the user. This model uses interpretable data points, including the speaker’s profession, their educational background, and their history of previous interventions in the chamber. When a user submits a query, the system retrieves relevant speech segments, identifies experts across different parliamentary groups, and generates a summary that synthesizes these diverse perspectives with verified supporting quotations.

Performance and Evaluation

The researchers evaluated ParliamentRAG against Google NotebookLM using 15 policy topics. The evaluation protocol combined automated metrics with a blind A/B human assessment conducted by six domain experts. According to the study, ParliamentRAG achieved a coverage score of 0.97 across political groups compared to 0.95 for NotebookLM. Furthermore, the system reached a perfect score of 1.00 for quotation faithfulness, whereas the comparison system scored 0.95.

Comparative Strengths

The evidence suggests that ParliamentRAG is more effective at handling source-related tasks, as indicated by stronger preferences from the domain experts involved in the study. However, the authors note that NotebookLM remains more effective in terms of prose-oriented dimensions. This suggests a trade-off between the structural, authority-based accuracy of ParliamentRAG and the stylistic fluency of general-purpose RAG systems.