基於社群媒體資料之兩階段品牌個性預測模型

Abstract

建立強大而與眾不同的品牌是企業的重要目標之一，它能夠幫助企業提升其績效和客戶忠誠度。在行銷實務上，評估品牌認知度最有效的框架是品牌個性。品牌個性被定義為「一組與品牌聯想在一起的人格特質」，它反映了消費者對企業或組織的看法和感受。過去研究發現，建立良好的品牌個性可以帶來消費者的正向反應。然而，消費者感知的品牌個性往往與公司預期的品牌個性不同，因此企業需要經常地衡量消費者所感知的品牌個性。過去經常以問卷或訪問的形式來評估消費者感受的品牌個性，但這類方法耗時費力，且無法長期監控品牌個性的變化。 隨著社交媒體的普及和廣泛使用，開始有行銷研究將社群媒體資料應用在品牌相關的研究。因此，本研究旨在透過社群媒體資料來衡量消費者感知的品牌個性，我們開發了兩階段的品牌個性預測模型，也採用 BERT 預訓練語言模型從推特資料中萃取嵌入作為特徵，同時考慮品牌個性之間的可能關係，再利用機器學習模型來預測和衡量消費者感知的品牌個性。而實驗結果顯示我們的兩階段預測模型優於基準方法，且說明了考慮品牌個性之間的關係能有效提升預測結果。希望我們提出的方法可以為品牌管理者們提供一個易於實作的方法，以便他們藉由社群平台上公開的資料來持續追蹤和管理消費者對品牌個性的看法。

Building a strong and differentiated brand is one of the essential goals for a company, which can help improve its business performance and customer loyalty. In marketing practice, the most effective framework for assessing brand perception is brand personality. Brand personality is defined as “a set of human characteristics associated with a brand,” and it reflects consumers’ perspectives and feelings about a company or an organization. Besides, prior studies indicate that a well-established brand personality can lead to more positive customer responses. However, customer-perceived brand personality may be different from company-intended brand personality. Therefore, companies need to devote efforts to assessing customer-perceived brand personality constantly. Traditionally, assessing brand personality relies on the survey-based approach, which can be time consuming, labor-intensive, and unable to constantly monitor changes in brand personality over time. With the popularity and widespread use of social media, recent marketing studies have begun to use social media data for brand-related research. Hence, in this research, we aim to measure customer-perceived brand personality with the social media data. Specifically, we develop a two-phase brand personality prediction model, which includes Base Models and Brand Personality Dependency Models. Moreover, we adopt the BERT pre-trained language model to extract embedding features from Twitter data, while considering possible relationships among brand personalities, and then utilizing machine learning methods to predict and measure customer-perceived brand personality. Our empirical evaluations demonstrate our two-phase prediction models outperform the benchmark method and also indicate that considering the relationships among brand personalities can achieve greater results. To sum, our proposed method can provide brand managers with an easy-to-implement method so that they can continuously monitor and manage customers’ views on the brand personality through the data publicly available on social media platforms.