使用使用者生成內容預測消費者感受之品牌個性

Abstract

品牌個性是「一組與品牌聯想在一起的人格特質」，也是消費者將品牌擬人化後的印象。相較於功能面的產品屬性，品牌個性可以滿足消費者象徵性的自我表達上的需求。消費者會喜歡與真實的、理想中的自己個性相符的品牌，用來展現自己。研究發現，品牌個性與自我概念之間的一致性會導向正面的消費者反應，例如：增加品牌偏好度、品牌忠誠度、購買意願、良好的口碑等等。因此，品牌管理者將品牌個性視為品牌策略中不可或缺的關鍵，藉此與同市場中的競爭對手做出區別，並建立無形的品牌獨特性，鎖定特定客戶群和制定定位策略。品牌管理者會試圖衡量公司以及品牌企圖建立的品牌個性，是否有確實地被消費者感受到。過去衡量消費者感受到之品牌個性，大都是以問卷或訪問的形式來做評估，但這類方法不僅費時費力，也難於長期頻繁地去監控不同地域的消費者對品牌個性感受的程度差異以及行銷活動或負面公司醜聞對於品牌個性的影響。 而隨著消費者愈來愈習慣在不同產品評論和社群媒體平台上發表對品牌的經驗與意見，這些使用者生成內容正不斷地蓬勃累積。因此本研究提出一個套用機器學習的方法，對大量使用者生成內容進行文字探勘，包括產品評論與推特推文，從文字中擷取出隱含的消費者感受與資訊，藉此預測與評估消費者感受之品牌個性。我們的實驗結果也顯示我們提出的方法能有效地評估消費者感受之品牌個性。希望此方法可以提供品牌管理者們一個易於實作、低成本、即時、有效的方式來與問卷調查的結果做互補和相互驗證，進而得到更具可靠性的評估結果。

Brand personality is defined as “a set of human characteristics associated with a brand,” which personifies a brand with human personality traits. In contract to functional product-related attributes, brand personality serves as symbolic usage for customers’ need of self-expression; hence customers prefer brands with brand personality that matches their actual or ideal selves. Furthermore, prior studies also suggest that greater congruence between brand personality and self-concept leads to positive customer responses. Accordingly, brand managers view brand personality as a key of brand positioning strategy and seek to ensure the consistency between company-intended brand personality and customer-perceived brand personality. Traditionally, marketers often use the survey-based approach to assess customers’ perceptions of brand personality. However, this traditional approach constantly suffers from several limitations. Nowadays, in order to understand customers’ opinions, many researchers have utilized user-generated content (UGC) available in product review platforms or social media on the Internet. Therefore, this research aims to develop a social-media-based approach for predicting and measuring customer-perceived brand personality by exploiting machine learning methods and utilizing UGC data. Our empirical evaluations suggest our social-media-based brand personality prediction method attains a satisfactory effectiveness and outperforms the benchmark methods. In addition, our proposed method with tweets as the data source achieves greater effectiveness than that with product reviews as the data source does. Our proposed method provides an easy-to-implement, low-cost and real-time approach that is a complement to the traditional survey-based approach.