現今網際網路，群眾易透過網路獲取所需之知識，且群眾評論易影響讀者之觀點，當群眾討論之過程中，部分群眾濫用評論發文，以發表不客觀之評論與攻擊性之評論，將影響群眾之討論品質，於不客觀及攻擊性之評論中，部分評論沒有給予群眾客觀及合理之意見（如批判性思考之評論），並且沒有直接表達真實語意，使讀者無法正確理解真實語意（如反諷性之評論），進而使讀者產生錯誤資訊，其中，反諷評論強調個人真實想法，以利用與真實語意相反之字面表達意見，易導致讀者較難釐清正確之語意，而誤解成與字面相同語意，而導致群眾產生錯誤之資訊。 有鑑於此，本研究建立一套「批判性思考、反諷性陳述改寫及推薦模式」，首先，本研究將針對批判性思考之特徵分析評論，以獲取較客觀之評論；其次，本研究將探討評論中反諷之表達模式並彙整，以辨識評論中反諷語句；最後，本研究將評論中之反諷語句進行改寫，以改寫為與實際相同之語句，再以網路理論為基礎，分析評論中領袖意見做為讀者參考之推薦。 最後，為確認本研究方法論於實務應用之可行性，以驗證本系統之績效，於「批判性思考模組」中，藉由「Yahoo奇摩新聞」做為驗證資料，並於績效驗證指標評估中，可獲得召回率為95%、正確率為50%及F值為65%；於「反諷性陳述解析模組」中，藉由論壇中撰寫者發表評論時標記「反諷」之評論做為驗證資料，可獲得整體平均召回率、正確率及F值為74%之績效水平；於「反諷評論改寫及推薦模組」中，藉由改寫滿意度量表評估改寫之反諷評論，並根據量表獲得數據得知各問項分析結果皆為顯著。 綜上所述，本研究乃提出「批判性思考、反諷性陳述改寫及推薦模式」及系統，藉由本系統輔助管理者評估評論內容，並改寫反諷語句，進而推薦較具參考性之評論，以改善讀者獲取錯誤之資訊。 Because the public can exchange with one another via internet to acquire the information they need and the information generated from comments posted by the public will mutually affect their perspectives, the knowledge acquired by the public to generate the same idea will be affected by people posting it. However, some of the posters use subjective expressions to post comments to mislead the public and affect their self-identity (e.g. critical thinking comments). Some of the posters may use ironic statements for expression of sarcasm or humor, which leads to the situation where the public cannot clarify the actual meanings expressed by ironic statements and further generate erroneous information. Consequently, this paper develops a Model for Critical Thinking Comments and Ironic Statements Rewriting and Recommendation based on discussing issues and comments in the forums to judge the categories of opinions according to the contents of opinions from the public, rewrite the opinions that are more incomprehensible, and further determine the leading opinion to recommend opinions with reference value to the public. Furthermore, this paper develops a Web-based system accordingly and uses a real-world case “YAHOO News” for case verification to confirm the feasibility of the methodology. The verification results show that the recall rate is 95%, the correct rate is 50%, and the F value is 65% for critical thinking comments determination and the average recall and accuracy rates can reach 74%. Generally speaking, the system performance grows continuously with the periods and training load, and eventually reaches stable and good performance level.
