隨著電子商務的興起，網路上的產品評論也隨之成長。但是為了獲取經濟利益，廠商提供的虛假消費者評論也隨之起。為了能解決這個重要問題，有幾個研究已經嘗試設計分類器將評論分為真實與虛假。這些研究大部分使用傳統的文字挖礦甚或深度學習方法。這些方法的共同特點就是建立模型使用的字詞與訓練集的文本高度相關。但在實務上，不同領域的評論會使用差異很大的字詞來評論產品，直接導這些分類器跨領域應用時accuracy 可以低到只有55%. 就算是深度學習的方法也沒改善多少。本研究將以Stimulus-Organism-Response (S-O-R) framework 為基礎。由這理論的觀念推導出相關的字詞品類。並經由LIWC, Wordnet 以及網站取的相關字詞。根據這些字詞，一個跨領域分類器將會被設計出來。這個分類器將會被應用在三個常用領域的評論來檢驗其精準度以及跨領域的應用性。 ;E-commerce has been developed at the high pace in recent years. Accordingly, the rise of potential spamming reviews from the online services is growing quickly and attract significant concern from many organizations. Therefore, deceptive detection is one of critical issues in online businesses. Existing studies investigated deceptive detection mainly base on the technology of traditional text mining. As a result, the detection is closely related to the training corpus collected. However, reviews for different application domains utilize very different words. As the result, the precision and accuracy of these approaches were less than ideal when being applying to different domains. In this study, a cross domain detector is designed by utilizing the Stimulus–Organism–Response (S-O-R) framework to infer word categories. The proposed approach will be intensively evaluated with the three benchmark datasets comprised by previous research to compare the performance with the state of the art approaches.
