This document presents a method for identifying key stakeholders in large-scale software projects. It involves building a social network of stakeholders through recommendations and then eliciting requirements from stakeholders. Stakeholder influence is determined using social network measures and ratings of requirement importance. A genetic algorithm is used to search for weights that can better determine real stakeholder influence compared to existing measures alone. The method was evaluated on a real-world dataset and was able to prioritize stakeholders and requirements more accurately than social network measures.
19. Step 5: Prioritise requirements
n
ImportanceR = ∑ ProjectInfluenceS × RatingS
S=1
Use social network measures, e.g.,
• Betweenness centrality
• PageRank
• Out‐degree centrality
• In‐degree centrality
S.L. Lim & A. Finkelstein (2011) StakeRare: Social Networks and CollaboraLve Filtering for
Large‐Scale Requirements ElicitaLon. IEEE TransacLons on SoCware Engineering (TSE).
20. Step 5: Prioritise requirements
0.81
0.70
0.58
0.56
0.49 0.48
S.L. Lim & A. Finkelstein (2011) StakeRare: Social Networks and CollaboraLve Filtering for
Large‐Scale Requirements ElicitaLon. IEEE TransacLons on SoCware Engineering (TSE).
21. Step 5: Prioritise requirements
0.81 n
ImportanceR = ∑ ProjectInfluenceS × RatingS
0.70 S=1
0.58
0.56
€ 0.49 0.48
S.L. Lim & A. Finkelstein (2011) StakeRare: Social Networks and CollaboraLve Filtering for
Large‐Scale Requirements ElicitaLon. IEEE TransacLons on SoCware Engineering (TSE).
22. Use a genetic algorithm
to search for
real influence
GA to search for weights
S.L. Lim, M. Harman & A. Susi. Searching for Key Stakeholders in Large‐Scale
SoCware Projects (submiVed).
29. Data Set
• ~150 requirements
• 68 stakeholders recommended other
stakeholders
• 76 stakeholders provided ratings
• actual ranked list of requirements based
on post project knowledge
30. Findings
• Existing social network measures can be
used to prioritise stakeholders….but they
are not optimal and may miss out key
stakeholders (GA can always improve
them).
• Evolution corrected assumptions made by
the measures that don’t hold for the
stakeholder.
31. Findings
• The GA found many good solutions
– A good set of requirements can be constructed
from many different subsets of stakeholders
• Some stakeholders hold unique knowledge
(always selected by the GA), but the majority
of stakeholders share similar knowledge
(replaceable)
• The concept of who is a “key stakeholder”
depends on which other stakeholders have
already been identified.