Extracting meaningful insights from multilingual tourism reviews poses significant challenges due to cultural terminology and linguistic diversity. This paper introduces FewTopNER, a framework that combines few-shot learning with topic modeling to recognize tourism-specific entities in Moroccan travel discourse. Unlike traditional approaches requiring extensive labeled data, our method leverages minimal training examples while incorporating thematic context to improve recognition of culturally embedded terms. We evaluate our approach on a comprehensive TripAdvisor dataset across five languages, focusing on accommodation, cuisine, regional, and transportation entities. Our framework demonstrates competitive performance with controlled degradation from generic baselines, while topic modeling integration proves particularly beneficial for culturally specific entities compared to neutral ones. Cross-lingual experiments show effective knowledge transfer between languages, and systematic error analysis identifies key challenges including orthographic variations and code-switching patterns. Our contributions include the first comprehensive multilingual tourism NER dataset for Morocco, systematic cross-lingual evaluation, and empirical validation of topic-aware few-shot learning in domain-specific contexts. The results establish practical foundations for automated multilingual tourism analytics and cross-cultural recommendation systems.