Abstract:
This thesis proposes an actionable and persuasive text classification method for automatically categorizing buyer-posted questions and comments in internet auctions. This is followed by examining how a seller’s previous post-transactional reputation score and auction pre-configuration affects people’s participation in intra-transactional communication within online auction communities. A leading horizontal intermediary auction platform with a huge market share in New Zealand, is used to conduct this research. Its seller “feedback” mechanism and “ask seller a question” forum are chosen as representatives of post- and intra-transactional information disclosure. A new classification approach is proposed in this study. Given a set of scores that are assigned to key words and sequences, the approach classifies disclosure statements into relevant categories. Our approach outperforms the classic Naïve Bayes algorithm by approximately 20%. The results of multinomial logistic regression indicate that product quality, shipment and payment issues are aspects that concern buyers the most in the early stages of an auction. Subsequently, their attention is likely to be shifted to seller credibility and price or swap negotiations as listing lengths get longer. In terms of the influence of seller feedback ratings, our findings suggest that lower-rated traders are more likely to be asked questions about product description and seller credibility. Buyer concern about seller uncertainty is only alleviated if the seller has built a high reputation in past transactions. Even the medium-rated sellers are suspected of being opportunistic. Moreover, buyers are more willing to discuss transaction-related issues and raise negotiation-associated questions with sellers who have already achieved high reputation scores. Finally, the theoretical and managerial implications are elaborated.