本研究結合新聞情緒指標與 Black-Litterman 模型，提出一套具備動態調整能力之資產配置方法，並以 AI (人工智慧) 概念股為例進行實證分析。Black and Litterman (1992) 所提出之模型架構，能整合市場均衡報酬與投資者主觀觀點，適用於動態市場情境下之資產配置。研究首先運用文字探勘技術，採用黃裕烈 (2024) 建立之臺灣財經情緒字典，從逾十萬筆新聞資料中計算情緒指標，並納入結合 CAPM 因子的迴歸模型，以預測個股未來報酬。預測結果進一步轉化為 Black-Litterman 模型中之主觀觀點，並設計以判定係數 R2 為信心水準的調整機制，建構出整合市場均衡與情緒觀點之投資組合。研究期間涵蓋 2020 至 2024 年，並採用每三個月再平衡的方式進行滾動回測。實證結果顯示，結合新聞情緒之配置策略多數情境下優於市場基準，特別是集中投資於情緒敏感資產時，可大幅提升報酬表現。然而，策略亦伴隨較高波動與潛在下行風險，顯示需搭配適當風險控管以達最佳效果。本研究驗證情緒資訊於資產配置中之可行性，並拓展 Black-Litterman 模型於非量化資訊整合的應用可能性。

This study integrates news sentiment indicators with the Black-Litterman model to develop a dynamic asset allocation strategy, focusing on AI (artificial intelligence) related stocks in Taiwan. Sentiment indicators are derived through text mining using the Taiwan Financial Sentiment Dictionary (Huang, 2024) from over 100,000 financial news articles, and are incorporated into a CAPM-based return forecasting model. These forecasts serve as investor views in the Black-Litterman framework, with confidence levels based on R² values. Using a rolling window and quarterly rebalancing from 2020 to 2024, the empirical results show that sentiment-driven portfolios generally outperform the market benchmark, especially when concentrated in sentiment-sensitive stocks. However, higher returns are accompanied by increased volatility and drawdowns, highlighting the importance of risk control. The findings demonstrate the feasibility of using qualitative sentiment data in quantitative portfolio models.