Urban planning increasingly faces the dual challenge of heterogeneous data sources and the complex realities of fast-growing cities, particularly in Africa. Conventional decision-support systems often rely on structured indicators, neglecting the richness of unstructured textual data such as citizen feedback, local reports, and narratives. This paper proposes a hybrid architecture for urban planning decision support, combining data warehousing, data mining, and natural language processing (NLP) to integrate and analyze multi-source information. The architecture is structured around a centralized data warehouse consolidating Subnational Water Access (SWA) indicators, geospatial attributes, and unstructured urban narratives. To achieve this, we apply UMAP to perform dimensionality reduction as part of the pattern discovery phase, after which we utilize BERTopic to extract meaningful topic clusters. The resulting topics, which reflect public sentiment on water access, are then spatially aligned with subnational water access indicators to enable comparative analysis across urban districts. We validate this framework using a case study in Nouakchott, Mauritania, where discrepancies between piped water availability and perceived infrastructure pressure are mapped and interpreted. The results show a strong semantic coherence among the extracted topics (coherence value score =0.95) and demonstrate the utility of spatialized topic modeling for highlighting latent urban inequalities. Our contribution enables more inclusive, contextaware, and data-driven decision-making in urban planning.