反詐騙語意防火牆
Anti-Scam Semantic Firewall

為什麼做這個專案 / Why this project exists

許多反詐騙工具仍停留在表面特徵辨識，例如語氣、關鍵字或格式異常。但真正重要的是：一則訊息是否具備主體、邊界、依據與責任，足以被納入人的決策過程。

Many anti-scam tools remain focused on surface signals such as tone, keywords, or abnormal formatting. This project instead asks whether a message has enough subject clarity, boundary, basis, and responsibility to influence a decision.

核心方法 / 核心判準 / Core methods and criteria

• 不是只判斷它「像不像詐騙」，而是檢查它是否具備可信的決策資格。 / The task is not only scam-likeness detection, but qualification for trust and decision entry.
• 如果主體不清、責任不明、依據不足、邊界模糊，訊息就不應被直接信任。 / If the subject, responsibility, grounds, or boundary are unclear, the message should not be trusted directly.
• 把語意責任鏈拆成可理解步驟，讓一般使用者也能形成治理式判讀。 / Break semantic responsibility into readable steps so non-experts can perform governance-style judgment.
• 把「不輕信」具體化成可操作的方法，而非抽象口號。 / Turn “do not trust too quickly” into an operational method rather than an abstract slogan.

適用場景 / Use cases

適用於一般民眾、長者、家庭成員、金融風控、客服審查、公共治理與教育宣導等需要建立「先審核、再決策」流程的場景。

Suitable for citizens, seniors, family members, financial risk control, service review, public governance, and educational contexts that require “review first, decide second” workflows.