| dc.contributor | 傳播學院 | |
| dc.creator (作者) | 陳聖智 | |
| dc.creator (作者) | Poopradubsil, Thamolwan;Chen, Sheng-Chih;Latcharote, Panon;Thaipisutikul, Tipajin | |
| dc.date (日期) | 2026-08 | |
| dc.date.accessioned | 24-Aug-2026 15:13:14 (UTC+8) | - |
| dc.date.available | 24-Aug-2026 15:13:14 (UTC+8) | - |
| dc.date.issued (上傳時間) | 24-Aug-2026 15:13:14 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=184607 | - |
| dc.description.abstract (摘要) | Timely and context-sensitive climate policy recommendation requires intelligent systems that can seam-lessly integrate evolving environmental trends with domain-specific knowledge. However, existing tools rely predominantly on static keyword-based searches, lacking the ability to combine structured time-series signals with unstructured policy texts or to perform semantic reasoning across modalities. This limitation hinders policymakers from formulating rapid, evidence-based responses to emerging climate challenges, especially in resource-limited and data-constrained contexts. In this work, we present SEEK-Policy (Semantic Embeddings for Environment and Knowledge-based Policy) an open and reproducible cross-modal retrieval framework that directly aligns multivariate time-series data with climate policy documents using semantic embeddings and retrieval-augmented generation (RAG). The framework introduces three key modules: TimeTranscriber, which transforms multivariate time-series signals into concise, policy-relevant summaries; ChunkAlign, which performs fine-grained retrieval by mapping these summaries to semantically aligned segments within large corpora of climate policy documents; and PolicySynthesizer, which composes these segments into coherent, interpretable recommendations. To support open science, we develop POLiMATCH, a dataset linking over 1,200 climate policy documents with 10 years of structured time-series data across multiple countries and sectors, enabling rigorous evaluation of cross-modal retrieval tasks. Empirical results demonstrate that SEEK-Policy consistently outperforms strong baselines, including a Siamese network with contrastive learning and a role-based multi-agent summarization approach, achieving improvements of approximately 22% in Hit@1 and 8% in MRR@5 over the best-performing baseline. Beyond retrieval performance, SEEK-Policy provides an interpretable pipeline that links quantitative environmental indicators to climate policy documents, aiming to reduce the technical burden on decision-makers in resource-constrained settings. The framework and POLiMATCH dataset are openly available to support further research in cross-modal climate policy retrieval. | |
| dc.format.extent | 101 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | PeerJ Computer Science, Vol.12, e4054, pp.1-42 | |
| dc.title (題名) | SEEK-Policy: a data-driven cross-modal retrieval framework for climate policy recommendation | |
| dc.type (資料類型) | article | |
| dc.identifier.doi (DOI) | 10.7717/peerj-cs.4054 | |
| dc.doi.uri (DOI) | https://doi.org/10.7717/peerj-cs.4054 | |
