LSTM Based Automated GUI Testing For Android Applications

With the continuous growth of smartphone holding rate and the further development of mobile network facilities around the world,mobile applications play an irreplaceable role in the life and production of modern people.Nowadays,ensuring the correctness and reliability of mobile applications is a key issue of social concern.However,the proven software testing methods have become less and less practicable to satisfy the growing demand for APP testing on the market.Automated testing is a major development orientations for testing techniques.At the same time,existing automated testing methods meet difficulty in systematic and efficient testing because the abstraction methods for app features are inaccurate and generated test cases cannot function as expected.This thesis proposes an Android GUI automated testing method that build nondeterministic GUI models with dynamic GUI abstraction methods and systematically generate model-based high-quality test cases by LSTM model.The three main contributions of this thesis are as following:1.This thesis proposes a method for modeling Android application using nondeterministic finite automata(NFA)with dynamic GUI abstraction methods based on action feedback,and defines test coverage criteria based on the model.The method make possible systematic model-based testing by offering appropriate GUI abstraction methods and stabilize model structure by tolerate nondeterministic events.2.This thesis proposes an approach to determine the behavior of nondeterministic actions in nondeterministic GUI models using LSTM models,which can be used in model-based test case generation to guide the generation of high-quality test cases whose execution on-device is the same as model-based generation.The method solves the problems that generated test cases cannot trigger test functions on the device effectively.3.This thesis implements an automated testing tool for Android App using proposed deterministic GUI model and LSTM-based test cases generation method,which can execute systematic automated testing on Android apps and generate comprehensible and reproducible error reports.Comparing with three state-of-the-art Android automated testing tools in several representative open source apps,the proposed methods can effectively explore various features and find bugs in app under test.