ANALYSIS OF DESIGN ELEMENTS IN THE MACHINE-PLATFORM-CROWD TRANSFORMATION
thesisposted on 2021-07-28, 11:05 authored by Yipu DengYipu Deng
Digital transformation greatly affects all segments of our society. There are three powerful trends unleashed by the digital revolution: machine, platform, and crowd. The first trend emphasizes that machine learning can either complements or supplements human capabilities, which leads to data-driven decision making. The second trend shows that value creation is moving from physical products to platforms (e.g., Uber and Airbnb) where network effects can have a great impact. The third trend is about the emergence of online crowds. Several good examples are crowdfunding platforms like Indiegogo and collaborative platforms such as Wikipedia. My research work studies these three trends from different aspects.
In the first project, we investigated how professional reviewers influence subsequent non-editor reviewers in their writing behaviors. Restaurants that receive editorial reviews are found to have reviewers who not only post more frequently, but also give lengthier and more neutral feedback. Further investigation of the mechanism finds that in terms of the topics, sentiment, and readability, following reviews of restaurants that receive editorial reviews become increasingly similar to their editorial reviews, indicating that a herding effect is the main driver of the shift in later reviews. In this study, we not only look at quantitative review characteristics such as rating and review length, but also extract qualitative review characteristics embedded in review text using Natural Language Processing (NLP) techniques (e.g., Topic modeling and Sentiment analysis).
In the second project, we studied how AI-based shelf monitoring can help manufacturers with their shelf management efforts. In general, we've discovered that AI-powered shelf monitoring boosts product sales. We further reveal that the positive effect shall be attributed to independent retailers rather than chained retailers. More broadly, the finding further suggests that AI-powered monitoring is more scalable, allowing manufacturers to cope more effectively with more heterogeneous objects. In this study, we analyzed shelf photos using deep learning (e.g., image recognition). Furthermore, we conducted a qualitative study (i.e., interviews) as a supplement attempt to uncover the underlying mechanism behind the interesting phenomenon found in our field experiment.
In the third project, we tried to understand the dynamic contribution patterns caused by backers’ multiple roles and fundraisers’ strategic behaviors. We show that projects described by more subjective content (i.e., title and introduction) significantly repel potential donors. We further show that fundraisers’ contribution to their own projects might increase donor’ intention to donate and has no significant impact on reward pledging of subsequent backers. Above that, we find a positive interplay between donation and reward pledge, suggesting a cross-channel peer influence that will facilitate the fundraising progress.