CLEAR: Climate Policy Retrieval and Summarization Using LLMs

Web platforms play a critical role in climate change communication. However, despite facing increasing climate issues, residents especially those living in rural areas still struggle to access and interpret climate policies, while local governments lack structured feedback on policy effectiveness and community needs. To bridge this dual communication challenge, we present CLEAR (Climate PoLicy rEtrieval and summA Rization), a system powered by Large Language Models (LLMs) which can analyze resident queries and provide customized responses addressing their climate concerns based on relevant local government policy documents. Our system leverages a fine-tuned Llama-3.2-3B to decompose natural queries into structured components to enable precise information retrieval and summarization. The workflow of the CLEAR system consists of: (1) semantic query analysis to identify resident location and climate concerns, (2) intelligent retrieval from a curated dataset of authoritative policy documents, and (3) multi-modal policy summarization with geospatial context. When there is no corresponding policy for the specific climate concern, the CLEAR system can help to generate contextualized feedback to relevant local governments. In this paper, through real-world implementation and demonstration, we show how CLEAR can effectively bridge the information gap between Australian rural communities and local governments through LLM-based policy retrieval and summarization. Our source code is available at https://github.com/wuyoscar/CLEAR.