Skip to content

Detecting Behavior Anomalies in Graphical User Interfaces

License

Notifications You must be signed in to change notification settings

seunAdeks/backstage

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

24 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Backstage - Detecting Behavior Anomalies in Graphical User Interfaces

An official web-page of the Backstage project can be found here.

UI analysis phase

Prerequiesites:

  1. Java 8
  2. Maven

Building:

$ mvn initialize
$ mvn package

Usage:

The script below only works on Linux and MacOS systems. If you want to run it on Windows, just examine the runApp.sh file and run apktool and backstage.jar manually.

./runApp.sh PATH_TO_APK/myApp.apk

Results

The tool produces:

  • appSerialized.txt file in output/<name_of_apk> folder with the UI model
  • <name_of_apk>_forward_apiResults_1.xml with the mappting of callbacks to APIs

Those files are needed to obtain the mapping between UI elements and APIs later on.

Generating data for the next phase

cd scripts
./ui_extraction.sh output <output_file_prefix>
./api_extraction.sh results <output_dir>

Detecting outliers phase

Prerequiesites:

  1. R
  2. Python 3
  3. pip3
  4. Linux or MacOS

Before you start

Before running the scripts you need to follow the steps below:

Open R console:

install.packages("logging", dependencies=TRUE)
install.packages("stringr", dependencies=TRUE)
install.packages("argparse", dependencies=TRUE)
slam_link="https://cran.r-project.org/src/contrib/Archive/slam/slam_0.1-37.tar.gz"
install.packages(slam_link, repos = NULL, type="source")
install.packages("skmeans", dependencies=TRUE)
install.packages("cluster", dependencies=TRUE)
install.packages("clue", dependencies=TRUE)
install.packages("doParallel", dependencies=TRUE)
install.packages("data.table", dependencies=TRUE)
install.packages("proxy", dependencies=TRUE)

Make sure you have Python v3 and pip v3 installed and:

pip3 install argparse
pip3 install numpy
pip3 install pandas
pip3 install webcolors
pip3 install gensim
pip3 install nltk
pip3 install spacy
python3 -m spacy.en.download

Finally, open python3 console:

import nltk
nltk.download()

and download the following packages:

  • wordnet
  • stopwords
  • words

Obtaining a raw data

In order to run a mutation analysis you need to obtain a raw data with information about UI elements, their labels and APIs. Please donwload a backstage_data.zip, unzip it to the script folder and put the reference to it via $SNAP_DIR variable inside the launch.sh script.

The zip-archive contains a data folder inside. Ideally, you should move it to the scripts folder and the $SNAP_DIR is already referencing it.

Run the analysis

cd scripts
./launch.sh

Reproducing results from tables V, VI and VII from the Technical Report

Mutation procedure takes place only in the middle of the whole analysis. Thus, you don't need to rerun the whole script in order to try different mutation techniques.

Open the launch.sh file in your favourite text editor and find the line with invocation of make_mutants.R. By default we use high distance mode, but you can also make outliers based on random choice or crossover label mutations. You can specify a desired mode by using -t parameter in make_mutants.R file. Possible values are:

  • dist - stands for high distance mode
  • rand - stands for random mode
  • cross - stands for crossover label mutations

You need to run the whole analysis only once. Next time, you can skip all steps up to the place when mutations take place.

Results of mutation

You can inspect mutation results in $DATA_DIR/top_bin/results.txt folder.

About

Detecting Behavior Anomalies in Graphical User Interfaces

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Java 95.2%
  • R 3.9%
  • Other 0.9%