Data-Driven and NLP-Based Structuring of Recommendation Framework for Urban Flooding Mitigation

<p dir="ltr">With rapid urbanization and climate change, urban flooding has become a critical challenge for cities worldwide. This research aims to learn from existing case studies and apply machine learning techniques to simplify the search and adaptability of these cases to current events. While current flood management strategies focus on hazard modeling and impact assessment using GIS, Remote Sensing (RS) data, and historical flood records, case-based flood management data remains largely underutilized. The unstructured nature of case studies further limits systematic analysis of past mitigation strategies. To address this gap, this study develops a data-driven framework that leverages case-based data to provide targeted flood mitigation solutions. By applying Natural Language Processing (NLP) and Machine Learning (ML), the framework structures textual case data to enable more efficient retrieval and comparison of adaptive strategies. Site condition embeddings are generated using various pre-trained language models, with the BGE model selected based on performance. Recommendation accuracy is further improved by combining semantic retrieval with traditional BM25 keyword matching and rule-based filters, such as climate zone classification. The resulting decision-support framework assists planners and policymakers in identifying and implementing adaptive flood solutions early in the planning process. Beyond retrieving relevant cases, the framework synthesizes strategies into categorized types, improving flood mitigation decision-making and promoting sustainable urban development.</p>