Knowledge bases play a vital role in the modern world, offering a systematic and structured approach to integrate various entities, concepts, rules, and relationships associated with real-world information. In this study, we aggregated the official Title IX documentation from 163 institutions, including 13 federal departments, 48 states, and 102 universities. This aggregated data forms the Title IX Knowledge Base (KB), a resource aimed at enhancing understanding and awareness of Title IX. Our approach involves analyzing topics at different levels and extracting fine-grained rules associated with them using Multi-LLM Paradigm.

We evaluated the effectiveness of our Title IX KB using a Retrieval-Augmented Generation (RAG) system, demonstrating superior performance in factuality, semantic similarity, correctness, and query-response relevance compared to naive method. These results underscore the utility of the Title IX KB in providing access to legitimate, detailed information, facilitating better policy understanding and alignment.

Moreover, this research highlights the potential use of the Title IX KB in training safe and harmless AI assistants under the Constitutional AI Framework. By fostering awareness and enhancing compliance with Title IX regulations across various institutions, the Title IX KB proves to be a valuable tool in promoting informed and equitable practices.