Supporting Agile Software Development by Natural Language Processing - Trustworthy Eternal Systems via Evolving, Software Data and Knowledge - page 99

$UWLIDFWV

DXJPHQWHGE\

OLQNLQJ

&RGHFRPPHQWV
-XQLWWHVWGHILQLWLRQV

:LNLV

8VHUVWRU\

&RPPLWPHVVDJHV %XJUHSRUWV

7DVN 7DVNQ

,06RFLDOQHWZRUN

3URMHFW
0DQDJHPHQWWRRO

PRUHWROHVV
VWUXFWXUHG
FORVHQHVVWRFRGH

NHHSVWUDFNRIWDVNV

Fig. 6. Step 1: Linking User Stories with Artifacts

INSTANCE

ROLE GOAL BENEFIT DETAILS

NP-ROLE NP-GOAL VP-BENEFIT D1 D2

print alternating ignore whitespace

Fig. 7. Possible instance representation. The roof is a compact representation to rep-
resent tree information.

structured input data, like tree-kernel based support vector machines [5,11] could
be applied to learn a similarity function in the structured space.
Once a mapping between artifacts and user stories has been established, the
second information aggregation step is performed (cf. Figure 8): a classifier is
trained to determine the status of the user story: “to be implemented/not yet
started”, ”in progress“, “completed”. The amount of artifacts found in the first
stage, as well as related meta-data (e.g. number of lines of code associated with
a commit message, amount of JUnit tests related to the user story, status of
unit tests, number of bugs fixed, etc.) could be exploited to train a system to
classify user stories into the three categories, while further giving aggregated
information on the collected artifacts. For instance, if in the example user story
(cf. Figure 2), code comments and commit messages referring to the first task
of implementing the fancy case method are found the user story is classified
as ”in progress”. If also test cases are found with a positive reporting and no
bug reports referring to the fancy case method are found, the user story can be
labeled as ”completed”.