Introduction
C-IKNOW implements a network recommendation system that incorporates social motivations for why we create, maintain, and dissolve our knowledge network ties. The network data is captured by automated harvesting of digital resources using Web crawlers, text miners, tagging tools that automatically generate community-oriented metadata, and scientometric data such as co-authorship and citations. The major contribution of C-IKNOW recommender system is to personalize the search process and results based on Monge and Contractor’s Multi-Theoretical, Multi-Level (MTML)[8] which investigates social drivers for organizing networks in communities with diverse goals such as exploring new ideas and resource, exploiting existing resources and capabilities, social bonding, mobilizing for action, and rapid response (or “swarming”). These drivers may act alone or in concert and in differing amounts within and across communities.
There are two types of recommendations produced by C-IKNOW: resource recommendation and team recommendation.
Resource recommendation
Based on the knowledge network and the position of requesters in the network, the C-IKNOW recommender system produces personalized search results through two steps: identify matching entities according to query terms, their metadata, and network statistics and select the best fits according to requester’s perspectives and connections in social networks. The recommendation procedure can be interpreted as an initial identification stage based on geodesic distance (coarse identification), positive matches (medium identification) and profile similarity (fine identification) that returns the same scores for all users based on the search keyword and recommended item followed by a selection stage that incorporates information about the relationship between the user and a potential recommendation to arrive at a final score.
Team recommendation
The requester can specify a target group of persons, the number of teams, and the range of team size and C-IKNOW team recommendation will generate the best group partitions of the population to archive requesters’ goals such as maximizing diversity, similarity, and relation density in each team.