Phrasing in Dynamic Requirements Trace Retrieva

Dynamic trace retrieval provides an alternate option to traditional traceability methods such as matrices, hyperlinks, and manual link construction. Instead of relying upon manually constructed and maintained traces, links are generated dynamically on an ´as-needed´ basis using information retrieval techniques. Prior work in this area has indicated that in order to retrieve between 90% to 95% of the correct traces, only low precision levels can be obtained, which means that analysts must spend time filtering out unwanted links. This paper describes a method for improving the precision of trace results through incorporating the use of phrases detected and constructed from requirements using a part-of-speech tagger. A project glossary is also used to find additional phrases and weight the contributions of key phrases and terms. The approach is implemented in a probabilistic trace retrieval tool and evaluated through a series of experiments. The results show that phrasing can significantly increase the accuracy of the dynamic trace retrieval tool by generally increasing precision, and also by moving good trace links towards the top of the candidate links list