由許多網路評論家共同撰寫的產品評論使企業有參考依據去改善其業務策略，並賦予口碑新的價值。其中企業積極地想了解的即是對於購買意圖的影響力，過去文獻多是以評論的相關面向去討論間接影響力，因此我們將直搗核心，補足過去並未深入探討的這部分。本研究將預測評論的影響力評估視為分類問題，並採用四項重要的理論構面作為擷取變數的基礎，分別為幫助度、可信度、資訊質量以及專業度，除了分析單一變數相關性，我們也運用屬性篩選算法去檢視各種變數組合，並提出一項集成式學習架構，用以預測產品評論的購買意圖影響程度，此外與其他著名的幾項分類演算法相比，我們提出的模型表現皆為最佳。最後，我們證實了結合評論的四個重要構面，才能達到較完整影響力的預測。;Product reviews, co-authored by many Internet reviewers, can help consumers make purchasing decisions and give businesses a basis for improving their business strategies. Among them, the most important thing for companies to find out actively is the influence on purchase intention. In the past, most of the literature discussed the indirect influence based on the relevant aspects of the review. Thence, we home in on the core of issue and complement the part of the past literature that has not been explored in depth. This study treats the influence evaluation of predictive reviews as a classification issue, and use four important theoretical aspects as the basis framework for extracting variables. Which are helpfulness, credibility, information quality, and professionalism. In addition to analyzing the correlation of a single variable, we also use attribute filtering algorithms to examine various combinations of variables. Besides, we propose an ensemble learning architecture to predict the degree of purchase intention influence of product reviews. Furthermore, compared with other well-known classification algorithms, our proposed model performs best. In the end, we confirmed the four important facets of the review in order to reach a more complete influence forecast.
