This study utilizes the FinBERT model (Araci, 2019) and rolling window methodology (Isakin and Pu, 2023) to analyze real estate news sentiment effects on Taiwan housing prices from 2013-2024. Using 2.1 million transactions and 8,652 news articles, we examine sentiment dynamics across multiple time horizons and regional characteristics. Results reveal a temporal reversal pattern: positive sentiment increases housing prices short-term but reverses long-term as fundamentals reassert influence. Sentiment dispersion consistently suppresses prices across all windows, confirming uncertainty effects. Non-special municipalities show stronger sentiment sensitivity than special municipalities, with Taipei exhibiting unique patterns. Thematic analysis across Policy Tools, Urban Development, Information Transparency, and Housing Affordability reveals differential impacts by domain. This study demonstrates FinBERT's applicability in real estate contexts and provides insights for policymakers and investors regarding temporal and spatial heterogeneity of sentiment effects.