25.11.2017
v članku
Loet Leydesdorff na strani 8 opiše morda zanimiv način, kako priti do povezanih omrežij. Dobljena produktna matrika je simetrična - dvodelni podmatriki sta ena transponiranka druge. Manjša omrežja lahko dobimo s postavljanjem pragov na vrednosti v matriki. Sam imam več naborov bibliografskih omrežij s pari dela X avtorji in dela X ključne besede.
Using WoS2Pajek we can transform the bibliographic data obtained from Web of Science into a collection of networks: WA - works ✕ authors, WK - works ✕ keywords, WJ - works ✕ journals, and a one-mode citation network Ci. Since they all share a common set works they are linked.
From this example we see a problem of blockmodeling of “stacked networks” - the partition of the common set is restricting the partitions of the other sets.
In the complete blockmodeling problem of linked networks (on two sets) the set of nodes consists of two disjoint sets V = U ∪ W, U ∩ W = ∅. We assume that we know the (network) matrices UU, UW, and WW. We are searching for partitions π of U and ρ of W that minimizes a criterion function P(π,ρ) measuring their blockmodeling quality.
Try the case: UU = AW * Ci * WA, WW = KW * Ci * WK, UW = KW * WA on (some subnetwork of) SN5 network. Select the right normalization. See