問答網站之問答互補性分析研究

Abstract

隨著網路科技的發展，問答網站逐漸成為重要的知識分享平台。在問答網站中，使用者可以提出任何種類的問題，然後等待其他使用者的問答。然而這樣的問答機制可能會產生一些潛在的問題。從問答知識中可發現有些問答彼此之間是類似或是重複的，導致使用者在搜尋相關知識時會獲得重複的資訊。此外，使用者對於答案內容中不明瞭或有興趣的資訊，常須進一步搜尋其相關問答知識。 為了改善上述的困難，本研究提出一個分析問答知識間相關互補性的方法。所提的方法考慮了目標問答知識與其他問答知識之間的問題相關性、答案創新性以及答案關聯性，更進一步將互補型態分為兩大類：創新互補與衍伸互補，以描述不同問答間不同的互補關係。我們提出決策樹分類法來確認問答間的互補關係，並利用分類結果，將問答知識間的互補關係建構成互補問答網路。本研究並提出一互補問答網路分析方法，進一步找出間接互補的問答知識。最後本研究以奇摩知識家問答網站做為實驗評估的資料來源，實驗結果顯示本研究所提出的方法確實能有效的為目標問答找出相對應的互補問答知識。

With the rapid development of Internet and Web 2.0 technology, Question & Answering (Q&A) websites have become an essential knowledge sharing platform. As the number of posting questions and answers increases rapidly throughout time, the massive amount of question-answering knowledge creates the problem of information overload. QA website, such as Yahoo! Answer, is an open platform for users to freely ask or answer questions, and there exists redundant QAs. Moreover, users may need to know further details of some information presented in a QA’s answer by using related keywords in a QA’s answer to search more extended complementary information. This research proposes a novel approach to identify complementary QAs of a target QA. We define two types of complementation - partial complementation and extended complementation. We adopt a decision-tree classification approach to build classification model and predict complementary-relationships between QAs based on three measures, question similarity, answer novelty, and answer correlation. A complementary QA network is constructed based on the complementary relationships between QAs. We propose a novel complementary QA network analysis to identify complementary QAs of target QA by searching the QA network to find direct and indirect-complementary QAs of target QA. We conduct experiments on a dataset collected from Yahoo! Answers in Taiwan. The experiment result shows that our proposed approach is effective in identifying complementary QAs. The proposed complementary QA network analysis can identify more indirect-complementary QAs which cannot be identified by the decision-tree classification approach.