[Paper Review] Mobile Social Big Data: WeChat Moments Dataset, Network Applications, and Opportunities

Yuanxing Zhang, Zhuqi Li+5 more|arXiv (Cornell University)|Oct 21, 2017

Human Mobility and Location-Based Analysis13 references3 citations

TL;DR

This paper introduces WeChatNet, a large-scale mobile social big data dataset from WeChat Moments involving over 25 million users and 246 million link reposts. It proposes novel analytics for information diffusion, network traffic prediction, and population distribution using real-world user behavior, achieving over 90% accuracy in traffic load prediction and enabling accurate offline population projection via Dirichlet Process modeling.

ABSTRACT

In parallel to the increase of various mobile technologies, the mobile social network (MSN) service has brought us into an era of mobile social big data, where people are creating new social data every second and everywhere. It is of vital importance for businesses, government, and institutes to understand how peoples' behaviors in the online cyberspace can affect the underlying computer network, or their offline behaviors at large. To study this problem, we collect a dataset from WeChat Moments, called WeChatNet, which involves 25,133,330 WeChat users with 246,369,415 records of link reposting on their pages. We revisit three network applications based on the data analytics over WeChatNet, i.e., the information dissemination in mobile cellular networks, the network traffic prediction in backbone networks, and the mobile population distribution projection. Meanwhile, we discuss the potential research opportunities for developing new applications using the released dataset.

Motivation & Objective

• To understand how online social behaviors in mobile social networks like WeChat Moments affect underlying network performance and real-world offline behaviors.
• To address the lack of publicly available, large-scale datasets for mobile social network (MSN) research, particularly those reflecting real user interactions with access control and privacy policies.
• To enable data-driven research on network applications such as traffic prediction, population distribution modeling, and KOL detection using real-world mobile social data.
• To release the WeChatNet dataset to foster innovation in mobile social big data analytics and support new applications in marketing, privacy, and public policy.

Proposed method

• Collected a real-world dataset (WeChatNet) from WeChat Moments, capturing 25,133,330 users and 246,369,415 link reposting records with strict privacy and access control policies.
• Proposed an online KOL (Key Opinion Leader) detection method that is independent of the number of a user’s friends, leveraging reposting behavior for influence estimation.
• Applied Dirichlet Process Mixture Models (DPMM) to model the spatial distribution of the floating population based on message diffusion patterns.
• Used temporal and spatial clustering of reposting activities to predict backbone network traffic with over 90% accuracy, enabling near-optimal resource allocation.
• Visualized geo-homophily effects across diffusion graphs during key events (e.g., Spring Festival) to analyze behavioral shifts in social networks.
• Leveraged group chat dynamics and private content sharing to model social tie strength and information propagation under restricted access policies.

Experimental results

Research questions

• RQ1 How can online reposting behaviors in WeChat Moments be leveraged to predict network traffic loads in backbone networks with high accuracy?
• RQ2 To what extent can message diffusion patterns in WeChat Moments reflect and predict real-world population distribution, especially during major events like the Spring Festival?
• RQ3 How do access control policies (e.g., no stranger access, private content) in WeChat Moments shape information diffusion and social tie strength compared to open platforms like Twitter?
• RQ4 What are the implications of using dynamic diffusion graphs for marketing strategies, particularly in relation to user influence and extrinsic rewards?
• RQ5 How can the WeChatNet dataset be used to detect spam, rumors, or privacy leaks in mobile social networks?

Key findings

• The proposed traffic prediction model achieved a prediction accuracy rate exceeding 90%, enabling near-optimal network resource allocation based on online behavior.
• The Dirichlet Process Mixture Model successfully captured the distribution of the floating population, allowing accurate projection of offline population distribution using online message diffusion data.
• Visualizations revealed significant geo-homophily effects in WeChat Moments, with distinct diffusion patterns observed before, during, and after the Spring Festival.
• The online KOL detection method demonstrated effectiveness in identifying influential users without relying on the size of their friend network, enhancing scalability and fairness.
• The dataset revealed that private content sharing and mutual-following relationships in WeChat strengthen social ties more than one-way following, influencing information spread dynamics.
• The study identified that group chats in WeChat serve as a key channel for connecting strangers and enabling rapid information diffusion, especially in localized contexts.