從現代知識管理的角度而言，常見問答集(Frequently Asked Questions，FAQ)可謂 一種將原始討論文章內容經過篩選與整理，並轉換成可分享的特定表達格式的知識內涵。網路討論區大量的資訊，需要密集的人力投入，方能彙整成符合讀者需求品質的FAQ。事實上，斷詞與自動摘要等相關技術之研究與應用已相當成熟與普及，網路討論區的FAQ知識應該可自動轉換產生，讓FAQ變成討論區的基本功能之一。基於上述動機，本研究提出一網路討論區FAQ知識自動轉換的概念模式，而為探討該模式之可行性，並實作了一個FAQ自動摘要的雛型系統以進行實證。第一階段實驗評估顯示，在經過篩選與整理的五個討論區主題，壓縮比為20%時其摘要之平均召回率與精確率高於45%.，相關文獻中之表現相當；而FAQ摘要的提示性、可讀性、句數適當性及介面的輔助性的評估上，獲得很高的認同度。於第二階段實驗未經篩選的五個討論區主題時，雖然多篇回應文章的討論主題發散，使得摘要的精確率與召回率降低，但互動式介面的便利，讓使用者主觀的接受度仍高。因此，本研究兩階段實驗結果，展現了本研究所提之FAQ知識轉換的概念模式，在技術整合與使用者觀點上的初步可行性。然而，於討論區回應文章發散的特性上，仍需進一步地整合自動摘要相關技術，以改善摘要文章的結構與提示完整性。本文最後並討論自動摘要實務應用上的意涵，以及未來繼續研究的方向，以期FAQ知識轉換的概念模式可以成功地被應用於知識管理的企業組織中。

From the perspective of modern knowledge management, frequently asked questions (FAQ) is a knowledge that is screened and organized from original articles, and then transformed into a standard question-and-answer representation for easier sharing. An Internet forum contains a large volume of interactive information, which requires extensive manpower input to maintain its readability and quality. In fact, related research and applications in text retrieval or automatic abstracting have been quite matured and popular, such that automatic FAQ abstraction can be a basic function of any Internet forum software. Accordingly, we proposed a conceptual model for automating the FAQ generation, and a prototype system was implemented to explore the feasibility of this model. In the first phase, five human-adjusted Internet forums were experimented, and the recall and precision were both over 45% under 20% of compression rate, which were equivalent to the averages seen in the literature. Moreover, the evaluation on indicative of content, readability, appropriateness of sentence size, and interface assistance showed a very high acceptance. In the second phase, five unadjusted Internet forums were experimented, the recall and precision rates were low due to the divergent character of Internet articles, but the user acceptance was still high due to the friendly interface. Overall speaking, the two-phase experiment supported the technical feasibility and user acceptance of the proposed model, but the divergence of Internet articles needs to be dealt further. Some implications and future work are also discussed at the end.