Müşteri şikayet yönetiminin metin analitiği yöntemiyle mükemmelleştirilmesi: Sosyal medyadaki şikayetlerin sınıflandırılması ve şikayet tipi tahminlemesi

The ubiquity, ease of use, and speed of social media gave a tremendous rise to sharing consumption experiences online. The upsurge in electronic word-of-mouth created both an advantageous and dangerous field for the companies. Whereas an effectively handled online complaint may result in winning the complainant back, a poorly handled complaint may dangerously backfire on the company. Thus, excelling in online complaint management became a priority for companies. Therefore, understanding online complaints, and identifying typologies of online complaints and complainants are vital for surviving in the digital landscape. Initiated from this fact, this thesis aims to both identify content-based and interaction-based online consumer complaint types and predict complaint types according to the complaint magnitude rooted in complainants' personality traits, emotion, Twitter usage activity, as well as complaint's sentiment polarity, and feedback frequency. Accordingly, 297 thousand complaint tweets, features of over 220 thousand consumer profiles and more than 24 million user tweets were gathered from Twitter. The obtained data was analyzed via two-step machine learning approach, where the first step is the determination of the content features through text analytics. In the second step, the relationships between the clusters and the independent variables were revealed through machine learning. The validity of the revealed relationships was proven by the prediction power of the model, where the model's overall accuracy rate reaches 61.7% for the content-based clusters and 70.5% for the interaction-based clusters. This thesis contributes to both literature and practice by identifying types of online complaints that reveal diverse customer complaint behavior on Twitter, proposing a set of content and profile features that can be utilized for predicting complaint type, revealing the relationship between content features, profile features, and online complaint type, and proposing a novel and flexible approach for excelling online complaint through prioritization.