| 摘要: | 現今虛擬論壇自由且便利之發言平台，透過發言規範與論壇管理員之審核，論壇使用者即可輕易地發表各項文章，然而，因使用者逐漸增加下，多數虛擬論壇開始設有管理者用以審核違規之文章，但因大量文章注入虛擬論壇中，導致論壇管理者難以逐一審核並回請修正所有之文章，且對於發文者而言，可能於無法意識中寫入之違規字詞，於此，文章撰寫者發文後將隨即觸犯論壇規範，並需再針對違規文章進行修改，使得發文者再發文之意願逐漸低下。有鑑於上述問題，本研究乃發展「提升論壇知識利用價值之論壇文章情感解析及語句結構重組模式」，並劃分為「文章表達情緒判定模組」與「發文者文章語句重組模組」兩核心模組，以論壇文章為分析之基礎，發展一套適用於論壇之方法論。於前者中乃以文章代表事件分析、情緒詞彙隸屬係數建立、相似語句分析、情緒機率與穩定值解析等技術，以推論文章之情緒類別；於後者中，本研究則以整合語意相似分析、評閱分數分析、候選填充句建立、以及多重組合句建立等技術，於最終將可重組文章之語句架構。承接於本研究發展之模式，為確認本方法論於實務應用中之可行性，本研究乃建構一套以網際網路為基礎之提升論壇知識利用價值之論壇文章情感解析及語句結構重組系統。此外，為驗證本系統績效，本研究以論壇文章真實資料作為驗證資料之樣本，針對「文章表達情緒判定」與「發文者文章語句重組」進行獨立驗證，確保兩相議題間之驗證結果不相互影響。另一方面，本研究亦以案例為導向探討整合兩核心模組，所建構之模式與系統，於實務情境中之應用與管理意涵。
At present, the virtual forums are free and convenient speaking platforms, according to the speaking specifications and through the examination of forum administrators, the forum users can publish various articles easily. However, as the number of users increases, most of virtual forums have administrators to examine the violative articles, but there are so many articles imported into the virtual forums, it is difficult for the forum administrators to check them one by one and return all the articles to be revised. On the other hand, the publishers may write violative words unconsciously, so the article writers infringe the forum specifications with the publication, and shall revise the violative articles, so that the publishers' willingness to publish articles declines gradually. For the above problems, this paper develops "A Forum Article Sentiment Analysis and Statement Structure Reorganization Model for Increasing the Forum Knowledge Utilization Value", divided into two kernel modules, which are "Article Expressed Emotion Determination Module" and "Publisher's Article Statement Reorganization Module". A methodology applicable to forums is developed by using the forum articles as the basis of analysis. The former one deduces the emotion type of articles by analyzing representative events of articles, creating emotion word membership coefficient, analyzing similar statements and analyzing emotion probability and stable value. In the latter one, this study uses integrated semantic similarity analysis, review score analysis, candidate filler sentence establishment and multiple combined sentences, the article's statement structure can be reorganized. In order to confirm the feasibility of this methodology in practical application, this paper builds a Web-based system. Furthermore, a real-world case is applied to evaluate the proposed model. To sum up, this paper analyzes the articles which may contain extreme words and reorganizes statements for this type of articles. For the forum administrators, the violative articles can be extracted rapidly from the articles with specific emotions and the violative statements will be reorganized. For the article writers, the statements of violative articles can be revised automatically, so as to save the time and manpower for revising violative articles, and the probability of violation can be known immediately. |
