TL;DR
This paper introduces "Goose," a mobile social application deployed in Germany and China, to study user social behaviors in real-world mobile social networks. They found that user activity patterns mirror daily life, social interactions follow a heavy-tail distribution and the Pareto principle, and information diffuses rapidly through epidemic strategies.
What Problem Does This Paper Address?
The proliferation of smartphones and social networks has led to a surge in mobile social applications. However, there's a significant lack of empirical data on how users actually interact within these applications in real-world, deployed environments. Understanding these human interaction and social behaviors is crucial for improving the design and quality of service for these applications. Existing studies often rely on simulations or limited datasets, failing to capture the nuances of spontaneous, opportunistic interactions that characterize mobile social networks.
Methods
The core of this research is the development and deployment of a mobile social application called "Goose." Goose is built on the Nokia Symbian platform and supports microblogging and messaging. Its architecture comprises three main components: a contact manager, a forwarding manager, and a network manager, utilizing both Bluetooth and GSM for communication.
The researchers employed an experimental methodology involving the deployment of Goose on two university campuses: the University of Goettingen (UGoe) in Germany and Nanjing University (NJU) in China. Volunteers used the application for 15 days, during which their interaction data was collected. The data collection focused on anonymized interaction traces, including message types, timestamps, sender/receiver IDs (hashed), and message size, while explicitly excluding message content to protect user privacy.
For analysis, they investigated user activity patterns, session counts, duration, mobility, and message statistics. They specifically analyzed user encounters (when devices are within Bluetooth range) and user interactions (messages received), applying concepts like heavy-tail distributions and the Pareto principle. Information diffusion was studied using an epidemic strategy for message broadcasting.
Datasets
The paper introduces a newly collected dataset from the 15-day deployment of the Goose application on two university campuses: the University of Goettingen (UGoe) in Germany and Nanjing University (NJU) in China. This dataset comprises anonymized user interaction traces, including message types, timestamps, hashed sender/receiver IDs, and message size. The data was collected from 12 users in UGoe and 15 users in NJU. This dataset differs from existing benchmarks by capturing real-world, spontaneous interactions within a deployed mobile social network environment, rather than simulated or controlled settings.
Key Insights and Ideas
- Human Activity Mirrors Daily Life: User activity patterns in mobile social applications closely align with typical daily work schedules, with slightly longer active durations and periodic usage. This suggests that mobile social apps are integrated into users' existing routines rather than creating entirely new behavioral patterns.
- Social Interactions Follow Heavy Tails and Pareto: User encounters in mobile social networks exhibit a heavy-tail distribution, meaning a few users encounter many others, while most users encounter very few. Similarly, user interactions follow the Pareto principle (80/20 rule), where a small subset of users (around 20%) forms the majority of close connections. This highlights the existence of a core group of highly active and connected users within these networks.
- Information Spreads Like a Virus: Information diffusion, particularly through broadcast messages, follows an epidemic strategy. This means information spreads very rapidly initially, reaching about 50% of the population in a short period, and then slows down, eventually reaching around 80% of the population over a longer duration. This rapid diffusion has implications for applications like real-time alerts or news dissemination.
- Communication Paths are Short: Communication paths between mobile nodes in this network are generally short, mostly within 6 hops. This is consistent with the "small-world" phenomenon observed in many networks.
Key Results
- User activity duration in UGoe was from 9 am to 10 pm, while in NJU it extended to 12 am, showing slightly longer active periods in NJU.
- The study found that approximately 20% of users make close connections to the other users (Pareto principle in interactions).
- Information diffusion using an epidemic strategy reached 50% of the informed population in a short term and approached 80% in a long term.
- Most messages could be delivered within 6 hops, verifying the small-world phenomenon.
- Broadcast messages, like status updates, were delivered in an average of 2 hops.
- Approximately 40% of messages were not delivered due to sparse connections in DTNs, indicating DTN routing is best suited for non-urgent messages.
Limitations and Future Work
Limitations:
- Dataset Size: The authors acknowledge that the dataset size is limited due to budget and platform constraints, with only 27 users participating in total. This may limit the generalizability of the findings.
- Platform Specificity: The Goose application was developed for Nokia Symbian, which might not represent the full spectrum of mobile operating systems and user experiences.
- Encounter vs. Interaction: The IMote experiment mentioned in the related works highlights that user encounters don't always equate to real human interaction, a nuance that might be present in the Goose data as well, though the paper focuses on collected interaction data.
Future Work:
- Enhance Goose Functionality: The authors plan to continue developing Goose, including features for group communication and integration with online social networks.
- Deeper Information Diffusion Analysis: Further research will focus on studying the characteristics of information diffusion in mobile social networks.
- Broader Deployment: Expanding the deployment to more diverse environments and a larger user base would provide more robust and generalizable insights.
Summary updated