One sentence summary
Proposed a method to advanced Palomba et al.'s work on extracting useful information from user reviews to main and evolve mobile apps.
Background
Currently, the app developer usually maintain their apps related to the user's reviews and scores. And this helps developer to located the bugs in their app. Previous research leverage IR-based fault localization techniques, but there exist some research gaps. First, the previous research focus on desktop app rather than mobile app. Second, the previous technique cannot automate the process of analysis on user reviews, which results in that the developer may ignore the new published problem.
Related work
Palomba et al. proposed CHANGEADVISOR, which clusters reviews based on similar requirement, and locates the set of source code that needs to be changed for each cluster of reviews. However, this technique relate to the reviews with detailed information, otherwise the technique could not map user's review to any class.
Method
- First, extract the informative sentences (i.e., user feedback) contained in user reviews.
- Second, use tool-SURF to automatically classify the user feedback into five categories, then cluster them using HDP which presents the best clustering performance among six popular clustering algorithms.
- Third, treat each cluster of user feedback as a query to search for the classes that should be changed in source code.
- Forth, propose a more accurate similarity metric named weight-selection-based cosine similarity to measure the similarity between an enriched version of comment cluster and the source code.
- At the end, for each cluster of user feedback, Where2Change returns a ranked list of potentially classes to be changed
Experiment
Discussion
All in all, the researcher proposed a weight selection-based cosine similarity metric to compute the similarity between the enriched cluster of user feedback and the source code. For the future work, the author plan to explore a better way to select the user feedback related to real faults and feature requests reported in issue reports. Moreover, the author are interested in developing a new method to locate source code classes to be changed for each user feedback entry rather than a cluster of user feedback.
Reading summary
What is the motivation?
Propose a framework to automate the process of analysis of user review, and fix the drawback in CHANGEADVISOR, which is no robust enough.
What is the novelty
The proposed framework implement weight selection-based cosine similarity metric.
What is the contribution
Overcome the limitation of CHANGEADVISOR.