AppGen: Mobility-aware App Usage Behavior Generation for Mobile Users (2412.07267v1)

Abstract: Mobile app usage behavior reveals human patterns and is crucial for stakeholders, but data collection is costly and raises privacy issues. Data synthesis can address this by generating artificial datasets that mirror real-world data. In this paper, we propose AppGen, an autoregressive generative model designed to generate app usage behavior based on users' mobility trajectories, improving dataset accessibility and quality. Specifically, AppGen employs a probabilistic diffusion model to simulate the stochastic nature of app usage behavior. By utilizing an autoregressive structure, AppGen effectively captures the intricate sequential relationships between different app usage events. Additionally, AppGen leverages latent encoding to extract semantic features from spatio-temporal points, guiding behavior generation. These key designs ensure the generated behaviors are contextually relevant and faithfully represent users' environments and past interactions. Experiments with two real-world datasets show that AppGen outperforms state-of-the-art baselines by over 12% in critical metrics and accurately reflects real-world spatio-temporal patterns. We also test the generated datasets in applications, demonstrating their suitability for downstream tasks by maintaining algorithm accuracy and order.

• The paper introduces AppGen, a novel generative model that synthesizes realistic mobile app usage data from mobility using probabilistic diffusion and urban context.
• The model employs probabilistic diffusion, an autoregressive structure, and an urban knowledge graph, achieving over 12% improvement in key metrics over state-of-the-art baselines.
• Practically, AppGen provides stakeholders like app developers and network operators with high-quality synthetic data, mitigating data privacy and collection challenges.

AppGen: Mobility-aware App Usage Behavior Generation for Mobile Users

The paper introduces AppGen, an innovative autoregressive generative model that aims to synthesize mobile app usage behavior based on users’ mobility trajectories. AppGen is designed to address the inherent challenges associated with the collection and sharing of mobile app usage data—challenges that include high collection costs and stringent privacy regulations. Given these challenges, AppGen offers a method for generating synthetic app usage data while maintaining fidelity to real-world usage patterns and respecting privacy concerns.

Model Overview

AppGen leverages a probabilistic diffusion model to simulate the stochastic nature of app usage behavior. By adopting an autoregressive structure, the model captures the intricate sequential relationships between different app usage events. A key component of AppGen is its ability to leverage latent encoding to extract semantic features from spatio-temporal points, which guides the generation of user behaviors that are contextually relevant. To enhance realism, AppGen employs an urban knowledge graph to depict relationships between urban elements such as base stations, regions, business areas, and points of interest.

Technical Contributions

1. Probabilistic Diffusion Modeling: AppGen models the stochastic nature of human app usage, which is inherently uncertain and influenced by numerous contextual factors. The use of a probabilistic diffusion model enables AppGen to iteratively refine and generate realistic app usage patterns from a Gaussian noise distribution. Its state-of-the-art nature is exemplified by significant performance improvements over baseline models.
2. Autoregressive Structure: AppGen uses an autoregressive approach to generate each app in a sequence one at a time. This allows the model to effectively capture dependencies across app usage events. The autoregressive mechanism is enhanced by a conditional module that considers the influence of historical and present spatio-temporal factors on current app usage.
3. Urban Knowledge Graph Utilization: The inclusion of an urban knowledge graph is pivotal for representing the spatial contextual relationships in an urban environment. This feature facilitates the extraction of spatial information relevant to mobile user behavior and contributes to the realistic synthesis of app usage data.

Numerical Evaluation

AppGen's performance was evaluated against state-of-the-art baselines using real-world datasets. The results showcase its superior efficacy, with improvements exceeding 12% in critical metrics such as Jensen–Shannon divergence and Spearman's Rank Correlation Coefficient. These results indicate that the generated app usage behavior aligns closely with real-world patterns.

Implications and Future Work

Practically, AppGen enables stakeholders such as app developers, network operators, and smartphone manufacturers to access high-quality synthetic app usage datasets for various applications while safeguarding user privacy. Theologically, this work contributes to the broader understanding of generative models applied in the context of mobile computing.

Future development could explore the integration of more complex contextual factors and enhance the robustness of the generative process. Additionally, the framework could be adapted to other domains where user interaction patterns are influenced by spatio-temporal contexts.

In summary, AppGen stands out as a significant contribution to mobile app usage behavior modeling—aiding stakeholders in overcoming data accessibility and privacy challenges while maintaining high fidelity to real-world data distributions.