An Effective Crowdsourced Test Report Clustering Model Based on Sentence Embedding

In the crowdsourced testing industry, efficient and automated classification of true bugs from test reports can greatly reduce the cost of software testing. Most of the existing methods are based on TF-IDF or machine learning methods to vectorize the test report and then construct a classifier. However, the document vector constructed by keywords more or less ignores the description information in the document, which affects the performance of classification and detection of real defects. In order to use the description information to construct an effective report clustering model, we propose a model called RCSE to encode test report description information at the sentence level, calculate the similarity between the test reports from the feature similarity of the description sentence, then cluster the test report. We evaluated the model on 3,442 reports. The experimental results show that the clustering model based on sentence embedding has an average purity of 12.3 % and an ARI of 22.0% higher than the keyword-based model on three report datasets.