This dissertation examines the predictability of user engagement on online social platforms by integrating theoretical perspectives from the literature on media and technology habits with principles of context-aware computing. It presents three studies, each targeting a different facet of technology-mediated communication, from social media use in general to more granular behaviors like active and passive use and instant messaging.
The first chapter proposes a novel approach to the study of social media habits through predictive modeling of sequential smartphone user behaviors. Using longitudinal smartphone app log data, it examines the predictability of app engagement as a way to capture a critical yet previously neglected aspect of media and technology habits: their embeddedness in repetitive behavioral sequences. The study employs Long Short-Term Memory (LSTM) and transformer neural networks to demonstrate that social media use follows predictable patterns over time and that its predictability varies substantially across individuals. T
he second chapter shifts focus to the potential of context-aware modeling as a holistic yet parsimonious and privacy-preserving approach to predicting user engagement on online social platforms. Analyzing over 100 million Snapchat sessions from nearly 80,000 users via deep LSTM neural networks, the study demonstrates the predictability of active and passive use based on past behavior and a notable improvement in predictive performance upon integrating momentary context information. Features related to connectivity status, location, temporal context, and weather were found to capture non-redundant variance in user engagement relative to features derived from histories of in-app behaviors. The findings are consistent with the idea of context-contingent, habit-driven patterns of active and passive use, highlighting the utility of contextualized representations of user behavior for predicting user engagement on online social platforms.
The third chapter investigates the predictability of attentiveness and responsiveness in instant messaging on a large online social platform. Utilizing metadata from over 19 million messages, the study examines the predictive power of a wide range of predictor groups, including message attributes, user attributes, and momentary context, as well as historical communication patterns within ego networks and dyadic relationships. The findings echo the overarching theme that habitual behaviors and contextual factors shape user engagement. However, in this case, dyad-specific messaging histories account for the overwhelming share of explained variance, underlining the socially interdependent nature of user engagement in instant messaging.
Collectively, the three studies presented in this dissertation make a theoretical contribution by establishing media and technology habits as a suitable framework for the study of user engagement and by introducing a novel perspective that emphasizes the repetitive, predictable, and context-dependent nature of media and technology habits. The research makes an important empirical contribution through the use of novel, large-scale, objective behavioral data, enhancing the ecological validity and real-world applicability of its findings. Methodologically, it pioneers the use of context-aware sequential machine learning techniques for the study of media and technology habits. The insights garnered from this research have the potential to inform the design of engaging and ethical online social platforms and mobile technologies, highlighting its practical implications for the billions of users navigating these digital environments on a daily basis.
| Identifer | oai:union.ndltd.org:columbia.edu/oai:academiccommons.columbia.edu:10.7916/dtsk-xr71 |
| Date | January 2024 |
| Creators | Peters, Heinrich |
| Source Sets | Columbia University |
| Language | English |
| Detected Language | English |
| Type | Theses |
Page generated in 0.0019 seconds