Utilizing Supporting Evidence to Improve Dynamic Requirements Traceability

Author(s): Cleland-Huang, J., Settimi, R., Duan, C., Zou, X.
Venue: International Conference on Requirements Engineering
Date: 2005

Sample Size: 3

Quality
3

Link: http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=1531035

This paper studies three strategies for enhancing the precision of dynamic requirements traceability by incorporating supporting information into a probabilistic retrieval algorithm. The strategies studied include hierarchical modeling, logical clustering of artifacts, and semi-automated pruning of the probabilistic network. Hierarchical modeling captures information stored in the hierarchical nature of artifacts, in which words used to describe a higher-level element capture the general meaning of its sub-elements. Logical clustering is based on observations that links to to occur in clusters. If a link exists to a document, then there is a greater chance that links should exist to other documents in its logical cluster. Pruning the probabilistic network reduces the search space for a particular query. Each enhancement was compared against the basic probabilistic retrieval algorithm on three datasets.

The hierarchical modeling and logical clustering enhancement strategies had little effect on the precision. On the Ice Breaker System (IBS) dataset, each strategy tended to improve precision by about 10% over the basic algorithm when applied only to the low-confidence links. However, neither strategy showed significant improvement when applied to the whole IBS dataset or when applied to the other two datasets. The probability pruning enhancement shows an improvement in precision of about 1-3% for all three datasets when the probability value is set to 50%. When the probability value is set to 0%, it shows significant improvement in precision but also significant loss of recall.

0