應用深度學習技術於網路虛假評論偵測

Applying Deep Learning Techniques for Fake Review Detection

網際網路蓬勃發展使得電子商務成為消費者重要的採購媒介。消費者為了取得商品的資訊，會到重要的購物論壇或是討論群組閱讀其他消費者的評論心得。這也使得網路的評論對消費者的採購決策有很大影響力和重要性。企業花錢聘用特定的寫手撰寫對自己有利的評論，不肖廠商更聘用寫手散播不利對手的評論。這些虛假評論會誤導消費者也會傷害商品製造商。過去研究都指出這些虛假評論真假難辨。本研究將採用深度學習技術與傳統文字探勘的技術來比較識別虛假評論的內容的效果，資料前處理用傳統與深度學習的技術，機器學習使用了多種傳統與深度學習的模型，來建構識別虛假評論的分類器，本研究實驗將使用過去學者所提出的台灣知名論壇虛假評論真實資料集。

E commerce becomes an important channel for consumers to purchase product. Online reviews are an important information resource for consumes before making a purchase. Users always browse online forum that are posted to share post-purchase experiences of products and services. However, the fake reviews in the online forum are harmful to consumers who might buy misrepresented products. Consumers can't identify authentic and fake reviews. This study proposed a novel framework to detect fake reviews which integrated several techniques. There are traditional text mining techniques to deal with textual data including bag-of-words, latent semantic analysis and word2vec for word representation. Next, we used machine learning to train the model to detect fake review, including SVM, deep neural network (DNN), convolutional neural network (CNN) and long short-term memory (LSTM). Finally, we evaluated the performance in a real dataset.