Analyzing the Complex Network of iOS App Store Recommendation Relationships

Background

Mobile applications (referred to as mobile apps) are a vital part of the modern Internet ecosystem. However, with the exponential growth in the number of mobile apps, it has become increasingly difficult for users to find desired apps in app stores and for developers to make their apps discoverable. To improve user experience, most app stores employ recommendation mechanisms based on user behavior or other algorithms. For example, the iOS App Store’s “You Might Also Like” feature displays other apps related to a particular app, forming a network of recommendation relationships.

Despite the significant impact of app recommendations on user behavior and market performance, little research has delved deeply into the characteristics of recommendation networks. Researchers aim to analyze these networks to uncover their relationship with user behavior and explore ways to improve app discovery processes or optimize market regulation. This study addresses this gap by systematically analyzing the recommendation network in the iOS App Store from the perspective of complex networks.

Paper and Author Information

The research paper, titled “Characterizing the App Recommendation Relationships in the iOS App Store: A Complex Network’s Perspective,” is a collaborative work by researchers from various institutions, including the Key Lab of High Confidence Software Technologies at Peking University, Huazhong University of Science and Technology, Alibaba Group, and Hong Kong University of Science and Technology (Guangzhou campus). The corresponding authors are Yun Ma from Peking University and Haoyu Wang from Huazhong University of Science and Technology. The paper was published in the April 2025 issue of Science China Information Sciences, Volume 68, Issue 4 (DOI: 10.1007/s11432-023-3973-1).

Research Process

Data Collection and Network Construction

To study the iOS App Store recommendation network, the research team collected a dataset containing over 1.34 million apps and more than 50 million recommendation relationships. The data was obtained as follows:

1. Seed App Collection: The researchers crawled daily real-time app rankings and the latest-released app lists from the iOS App Store (China region) between January 1, 2020, and March 31, 2021.
2. Keyword Expansion: Using App Store Optimization (ASO) keywords related to the seed apps, the team applied keyword-based searches to find more relevant apps.
3. Recursive Crawling: A breadth-first search was performed using these seed apps to iteratively gather all available recommendation relationships and app metadata.

This data was then used to construct a large-scale app recommendation network, where nodes represent apps, and directed edges represent recommendation relationships. The network’s dense connectivity and directed nature provided a solid foundation for complex network research.

Network Analysis

1. The team analyzed the overall characteristics of the recommendation network using complex network analysis methods and compared the network with other known networks such as product co-purchasing networks and social networks.
2. The distributions of node in-degrees (the frequency with which an app is recommended) and out-degrees (the frequency with which an app recommends other apps) were calculated.
3. Strongly connected components (SCCs) and weakly connected components (WCCs) were studied to understand the connectivity structure of the recommendation network.
4. Local clustering coefficients were computed to examine modularity and tight connections between apps.

Specific Research Methods

By combining app metadata and network structure characteristics, the researchers further analyzed:

1. Common Features of High-Exposure Apps: They examined app categories, user ratings, and maintenance activities (e.g., update frequency) to identify which apps were more likely to be recommended.
2. Bidirectional Recommendation Characteristics: Categories and other variables were analyzed to understand which types of apps were more likely to recommend each other.
3. Local Network Structures: Through network motif analysis, the researchers uncovered small subgraph patterns that reflect specific behaviors in the recommendation mechanism.

Building on these findings, the team designed a method based on maximum clique detection to automatically identify policy-violating apps.

Key Research Results

1. Complex Network Characteristics:

   • The iOS App Store recommendation network exhibits small-world properties, where most nodes are well-connected through a small number of intermediate nodes.
   • The network features a massive strongly connected component (SCC covering 80.08% of nodes), indicating that most apps in the recommendation system maintain a certain degree of exposure, rather than being “forgotten.”

2. Revealing Recommendation Patterns:

   • Apps in specific categories (e.g., business, education, games) are more likely to receive high recommendation rates.
   • Highly rated and frequently updated apps gain more recommendations, while low-quality apps almost never receive exposure (7.67% of nodes have zero in-degree).
   • Highly exposed apps often act as central nodes in clusters of similar or complementary apps.

3. Local Motif Analysis:

   • Motif 1 (Figure 8(a)): This pattern shows certain key apps being widely recommended to meet specific user needs.
   • Motif 2 (Figure 8(b)): Apps in the same category often tightly recommend one another, forming strongly connected cliques. This reflects functional complementarities or exclusive content differences, which lead users to install similar apps.

4. Policy-Violating App Detection:

   • By analyzing maximum cliques in the network, the researchers proposed a method for identifying policy-violating apps. This method successfully detected 43.75% of removed apps and uncovered numerous previously undetected policy violations, including fake descriptions, manipulated reviews, and low-quality apps.

Implications and Highlights

This study provides insights into the iOS App Store’s recommendation mechanism and offers significant scientific and practical value:

1. Scientific Value: The research advances knowledge of complex networks, particularly in the context of product recommendation networks, through innovative analyses of local structures and motifs.
2. Practical Value:
   • Developers can leverage the findings to enhance app visibility by focusing on popular categories, improving app quality, optimizing descriptions, and building relationships with high-exposure apps.
   • The proposed detection method provides a novel perspective for strengthening app market regulation, aiding in the identification of policy-violating apps.
3. Methodological Innovation: The study presents robust data collection techniques, analytical strategies, and novel approaches to motif analysis and policy violation detection.

This research offers a strong data-driven foundation and methodological framework for understanding and improving mobile app recommendation mechanisms. It opens up new opportunities and directions for developers, market regulators, and researchers.