Detecting Popularity of Ideas and Individuals in Online Community
Research in the last decade has prioritized the effects of online texts and online behaviors on user information prediction. However, the previous research overlooks the overall meaning of online texts and more detailed features about users’ online behaviors. The purpose of the research is to detect the adopted ideas, the popularity of ideas, and the popularity of individuals by identifying the overall meaning of online texts and the centrality features based on user’s online interactions within an online community. To gain insights into the research questions, the online discussions on MyStarbucksIdea website is examined in this research. MyStarbucksIdea had launched since 2008 that encouraged people to submit new ideas for improving Starbuck’s products and services. Starbucks had adopted hundreds of ideas from this crowdsourcing platform. Based on the example of the MyStarbucksIdea community, a new document representation approach, Doc2Vec, synthesized with the users’ centrality features was unitized in this research. Additionally, it also is essential to study the surface-level features of online texts, the sentiment features of online texts, and the features of users’ online behaviors to determine the idea adoption as well as the popularity of ideas and individuals in the online community. Furthermore, supervised machine learning approaches, including Logistic Regression, Support Vector Machine, and Random Forest, with the adjustments for the imbalanced classes, served as the classifiers for the experiments. The results of the experiments showed that the classifications of the idea adoption, the popularity of ideas, and the popularity of individuals were all considered successful. The overall meaning of idea texts and user’s centrality features were most accurate in detecting the adopted ideas and the popularity of ideas. The overall meaning of idea texts and the features of users’ online behaviors were most accurate in detecting the popularity of individuals. These results are in accord with the results of the previous studies, which used behavioral and textual features to predict user information and enhance the previous studies' results by providing the new document embedding approach and the centrality features. The models used in this research can become a much-needed tool for the popularity predictions of future research.
Rayz, Purdue University.
Off-Campus Purdue Users:
To access this dissertation, please log in to our