A Search-Based Approach to Geographical Data Generation for Testing Location-Based Services

Xiaoying Bai,Kejia Hou,Jun Huang,Mingli Yu
DOI: https://doi.org/10.1002/9781119357056.ch19
2018-01-01
Abstract:Location-Based Service (LBS) has been a key infrastructure service for mobile applications like navigation and recommendation. Once hosted on a cloud platform, the service can support a large number of location-related queries from diversified devices and applications. The quality of the service, such as correctness, precision, and timeliness of the location information, is critical to those dependent applications and the ecosystem of mobile software built around it. Testing of the open services is thus necessary. However, LBS testing is challenging due to the enormous input domain of geographical areas. It is impossible to exhaustively enumerate all the possible inputs, and the corresponding expected outputs as well. How to design test inputs to achieve a certain level of confidence? How to validate test results when there are a large number of test cases? As manual testing is expensive, requiring extensive time to develop and computation resources to execute, how to devise an automatic approach enhancing test efficiency and effectiveness? To cope with the difficulties, this research proposes a testing framework to support automatic LBS testing from two aspects: geographical test data generation and query result verification, taking geocoding and reverse geocoding as illustrative LBS functions. To generate geographical data, a simulated-annealing algorithm is designed to formulate the problem as an iterative optimization search process. It is based on a defect clustering heuristic assumption. That is, defects tend to cluster in certain areas. Hence, areas with detected defects deserve more effort in the follow-up testing. A Bayes classifier is used to predict the defect probability of geographical areas, and to guide geolocation data generation in the search process. To automate the validation of query results, it applies a voting mechanism to compare results across different platforms providing the same LBS functions. Experiments were exercised on four LBS platforms. The results showed considerable improvements in defect detection with reduced cost.
What problem does this paper attempt to address?