MABTriage: Multi Armed Bandit Triaging Model Approach

Abstract

Recommendation of bugs to appropriate developers about whom we have very less or no information is a challenging problem faced in many open source developers community. In most of the reported works, this bug-triaging problem is handled through popular machine learning algorithms. However, in the absence of sufficient information of either a developer or a bug, it is difficult to build, train and test a conventional machine-learning model. One of the possible solutions in such a scenario is a reinforcement-learning model. In this paper, we propose an approach called MABTriage, to help a triager assign bugs to developers under uncertainty. To the best of our knowledge, it is the first work that has formulated bug-triaging process as a MAB problem. Experiments conducted on five publicly available open source datasets have shown that MABTriage approach performed better than a random selection. We have also evaluated the performance of six MAB algorithms -Greedy, -Decay, Softmax, Thompson Sampling, Optimistic Agent and UCB based on cumulative rewards. Results have shown that all five performed well in comparison to random selection.