網路的興起，電子商務與社群商務開始發展，然而大量的商機中隱藏了巨量的詐欺交易事件。隨著COVID-19的期間，許多人開始習慣於網路上進行電子商務交易，其中如何準確地評估商家以進行交易十分重要。有些詐騙商家透過社會影響力欺瞞潛在客戶，透過虛假評論影響客戶的選擇。此研究利用社交網路分析和圖神經網路以偵測虛假評論，藉由評論者之間關聯程度，創建社交網路關係圖，利用圖神經網路創建預測詐欺交易的虛假評論偵測系統。本研究中使用評論的時間、星等以及評論者資訊創建社交網路關係圖，並使用兩種以上的標籤方式標記電子商務詐欺類別，及早識別詐欺現象以中斷詐欺交易鏈。研究中使用Amazon以及Yelp資料集評論，總資料集為Amazon 2338637筆、Yelp 67395筆。圖神經網路模型準確度為87%、73%於兩種不同的標籤模型方法。採用Yelp資料集特徵，評估社交影響力於社群商務上的影響力，證實社交影響力89%影響社群商務決策。在社群媒體以及網際網路蓬勃發展的時代，此研究結果證實社交影響力之於社群商務之影響、虛假評論偵測系統可有效偵測虛假評論，該研究實現更安全的電子交易環境，減少詐欺交易的發生。;With the rise of the internet, e-commerce and social commerce have flourished. There is a significant number of fraudulent transactions. During COVID-19, many people became accustomed to online e-commerce, making it crucial to accurately assess merchants before engaging in transactions. Some fraudulent sellers use social influence to deceive potential customers by posting fake reviews to sway their choices. This research employs social network analysis and graph neural networks (GNNs) to detect fraudulent reviews. A social network graph is created to predict fraudulent transactions by analyzing the relationships among reviewers. The study uses review timing, star ratings, and reviewer information to construct the social network graph, applying multiple labeling methods to categorize e-commerce fraud types. The datasets used include reviews from Amazon and Yelp, totaling 2,338,637 and 67,395 reviews, respectively. The GNN model achieved 87% and 73% accuracy using two different labeling methods. By leveraging features from the Yelp dataset, the study evaluates the impact of social influence on social commerce, demonstrating that social influence affects 89% of social commerce decisions. In an era of rapid social media and internet growth, this research confirms the significant impact of social influence on social commerce. The fraudulent review detection system effectively identifies fake reviews, contributing to a safer e-commerce environment and reducing fraudulent transactions.
