A Longitudinal Study of Apps Removed from IOS App Stores
introduce
This is a paper "A Longitudinal Study of Removed Apps in iOS App Store" from WWW 2021. The authors are as follows: Fuqi Lin, Haoyu Wang, Liu Wang, Xuanzhe Liu
Article content
Longitudinal measurement of deleted apps in the iOS app store.
Pre-knowledge
ASO: app search optimization, application search optimization, by optimizing the keywords of the application metadata (name, description), indicating the prevalence rate in the search results
Measurement methods
- Identify which apps were removed, and the associated dates, by comparing datasets from January 2019 to April 2020
- Describe the overall condition of the removed app
- Analyze the actual reason behind the removal of popular apps
- Designed an app removal prediction model based on machine learning techniques
Dataset source
1. Obtain daily deleted applications from the IOS application store every day, and know which applications are deleted and when they are deleted by comparing two consecutive data sets
2. Application details: {[Application target information: application name, ID, developer name, application price], [application evaluation], [application popularity: market ranking, ASO capability]}, where the data of ASO capability comes from cooperative companies.
measuring angle
Angle 1. Overall Situation
1. General trend of deleted apps
1) Overall distribution:
2) Types of deleted apps:
2. Popularity of deleted apps
1) App popularity
2) Number of ratings by users
3. Developers of removed apps
1) Proportion of removed apps per developer
2) Most aggressive developers
4. Removed app lifecycle
Measuring update date and removal date, release time and removal time, removal date and relaunch date
Angle II. Reasons why popular apps were removed
1. Reasons for deletion
2. Behavior patterns of app comments
1) Duplicate comments
Ratio of repeated reviews via deleted popular apps and common apps
2) 5-star positive reviews
3) Abnormal users
3. Behavior pattern of ASO key rules
4. Behavior pattern angle of description
3. Detect deleted applications
Utilize machine learning to design automatic prediction of suspicious applications
Logistic Regression, Support Vector Machine (aka, SVM), K-Nearest Neighbors (aka, KNN), Random Forest, and Gradient Boosting Decision Tree (aka, GBDT). (Based on https://scikit-learn.org/stable/ , https://lightgbm.readthedocs.io/en/latest/)
compare two similar papers:
Beyond Google Play: A Large-Scale Comparative Study of Chinese Android App Markets (2018IMC): Only malware removal was measured
Why are Android Apps Removed From Google Play? A Large-scale Empirical Study (2018MSR):
Apps to be removed by Malware, Privacy Risky Software, Fake Apps, Spam Apps, Ad Blocking Apps, COPPA Violating Apps ( Does the developer raise a clear privacy notice, does the application description comply, does the third-party library share sensitive information); no predictive model is built