Improving Quality of the Solution for the Team Formation Problem in Social Networks Using SCAN Variant and Evolutionary Computation
Keywords - Genetic Algorithms(GA), Cultural Algorithms(CA), Structural Clustering Algorithm for Networks(SCAN), Weighted Structural Clustering Algorithm for Networks(WSCAN) for Team Formation Problem(TFP), Hybrid Genetic Algorithm, Hybrid Cultural Algorithm, Evolutionary Computation, Optimization
Social Network Analysis helps to visualize and understand the roles and relationships that ease or impede the collaboration and sharing of the information and knowledge in an organization. In this research work, we will focus on the Team Formation Problem (TFP) which is an open problem where we need to identify an ideal team, with members of complementary talent or skills, to solve any given task. Current research suggests that TFP solutions have been attempted with evolutionary computation approach using Cultural Algorithms (CA) and Genetic Algorithms(GA). However, SCAN (Structural Clustering Algorithm for Networks) variants such as WSCAN (Weighted Structural Clustering Algorithm for Networks) demonstrate a high capability to find solutions for another type of network problems. In this thesis, we first propose to use WSCAN-TFP algorithm to deal with the problem of team formation in social networks, and we our findings indicate that WSCAN-TFP algorithm worked faster than the evolutionary algorithms counterparts but was of lower performance compared to CAs and GAs. Next, we propose two hybrid solutions by combining GA and CA with a modified WSCAN-TFP algorithm. To test the performance of our proposed approaches, we define multiple quality criteria based on communication cost (CC), average fitness score (AFS) and average processing time. We used big datasets from DBLP nodes network with sizes 50K and 100K. The results show that our proposed methods HGA and HCA can find the near-optimal solutions faster with minimum communication cost with the improvement of approx. 66% and approx. 57% in average fitness in comparison to existing GA and CA methods respectively.