Abstract | ||
---|---|---|
Latent Semantic Indexing (LSI) is an accepted technique for information retrieval that is used in requirements tracing to recover links between artifacts, e.g., between requirements documents and test cases. However, configuring LSI is difficult, because the number of possible configurations is huge. The configuration of LSI, which depends on the underlying dataset, greatly influences the accuracy of the results. Therefore, one of the key challenges for applying LSI is finding an appropriate configuration. Evaluating results for each configuration is time consuming, and therefore, automatically determining an appropriate configuration for LSI improves the applicability of LSI based methods.
We propose a fully automated technique to determine appropriate configurations for LSI to recover links between requirements artifacts. We evaluate our technique on six sets of requirements artifacts from industry and academia and show that the configurations selected by our approach yield results that are almost as accurate as results from configurations based on a ground truth like known links or expert knowledge. Our approach improves the applicability of LSI in industry and academia, as researchers and practitioners do not need to determine appropriate configurations manually or provide a ground truth.
|
Year | DOI | Venue |
---|---|---|
2015 | 10.1109/RET.2015.13 | RET@ICSE |
Field | DocType | ISBN |
Automated technique,Data mining,Latent semantic indexing,Information retrieval,Computer science,Agile software development,Ground truth,Test case,Tracing | Conference | 978-1-4503-5749-4 |
Citations | PageRank | References |
4 | 0.39 | 21 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Sebastian Eder | 1 | 47 | 6.27 |
Henning Femmer | 2 | 158 | 16.72 |
Benedikt Hauptmann | 3 | 87 | 9.13 |
Maximilian Junker | 4 | 62 | 8.73 |