| Summary: | 博士 === 國立成功大學 === 工業與資訊管理學系 === 107 === The rapid growth of e-commerce stores has led to an increasing number of online forums that allow customers to share their opinions (e.g., amazon.com, walmart.com, nike.com, and levi.com) regarding the retailers in order to facilitate their purchase decision-making process, as well as third-party product reviews (e.g., epinions.com, rateitall.com, and ZDNET.com). By allowing customers to express their experiences and rate reviews written by others, such forums greatly aid both customers in making purchase decisions and also companies in conducting promotion and marketing strategies. However, information overload due to the huge amount of reviews posted daily complicates the efforts of consumers to locate reliable information when making a purchase decision. Numerous recent studies have focused on identifying credible online reviewers of products or services by basing their rating scores on accumulated votes from the community of users. However, conventional evaluation methods consider only voting scores and fail to consider the reviewers’ expertise and behavior with respect to the source of information and the content quality of the reviews, and thus the user-driven approach has bias limitations, low coverage, and limited applicability.
Therefore, regarding to find reliable information, this study aims to propose two novel approaches based on social mechanism including trust and reputation. The first proposed approach is in term of trust in society designed using an illustrative example of mobile phones to build a trustworthy co-created recommendation model (TCo-CR) by mining unboxing forums. The model is evaluated via an empirical experiment to examine the satisfaction of study participants by using a seven-point Likert scale. The second approach is in term of reviewers’ reputation tested on different product types to locating reputable reviewer method (L2R2) based on their linguistics styles of reviews posted on online opinion sharing forums. The performances of the proposed L2R2 model are evaluated on experience and search product types using logistic regression and the Support Vector Machine (SVM) methods.
Both approaches of this study are fully implemented and tested on real-world datasets then compared with the baseline models. As the result, the proposed models outperform the baseline models and have greater customers’ satisfaction and higher estimation accuracy in evaluating the reputations of online reviewers. In addition to providing novel approaches to identifying trustable information and reputable online reviewers, the proposed approaches not only allows customers to locate their desired products and services efficiently, but also significantly contributes to the efforts of marketers in promoting and developing products and services based on trustable information by co-creating with users and reputable reviewer comments.
|
|---|
