Investigating model explanation of bug report assignment recommenders
Omee, Farjana Yeasmin
University of Lethbridge. Faculty of Arts and Science
Lethbridge, Alta. : Dept. of Mathematics and Computer Science
Software projects receive a lot of bug reports, and each bug report needs to be triaged. An objective of the bug report triaging process is to find an appropriate developer who can fix the reported bug. As this process can be time-consuming and requires a lot of effort, researchers have implemented recommender systems using a variety of algorithms to automate this process. Although using these recommender systems has a number of benefits, there are still many obstacles to overcome. A key obstacle is that commonly used algorithms are black-box, making it difficult for practitioners to comprehend how the models make decisions. Lack of explainability results in a lack of trust and transparency in the recommendations. This work investigates approaches that lead to visually explainable bug report assignment recommender systems. First, we developed and compared six different recommender systems using three distinct machine learning algorithms: Random Forest (RF), MLP Classifier and Bidirectional Neural Networks (BNN) and two different feature extraction techniques: TF-IDF and Word2Vec. Second, we examine the use of WordNet to improve recommender accuracy. Third, we explore the explanation of a bug report assignment recommender using the feature-based local model LIME. Finally, we assess the use of a positivenegative horizontal bar chart, feature table, and word cloud to explain the recommender systems visually. Our analytical analysis indicates that the optimum approach for developing a bug report assignment recommender system uses TF-IDF with RF and visually explains the recommendation with a word cloud and LIME as a local model.
bug reports , recommender systems , bug report triaging process , machine learning algorithms , visual explanation of algorithm decisions , Software failures--Data processing , Debugging in computer science , Recommender systems (Information filtering)--Research , Artificial intelligence , Machine learning , Algorithms , Dissertations, Academic