Citation analysis
Components
The network CiteB has 690 connected components with sizes 12702, 21, 20, 19, 17, 10, 9, … We limit our analysis to the largest component CiteMain. It has 5 strong components all of size 2.
To get an acyclic network we apply the preprint transformation to CiteMain. The resulting network CiteMacy has 12712 nodes and 81972 arcs. We compute the SPC weights on its arcs. The total flow is equal to 1.625 1020.
CPM path
First we determine the CPM path in this network
We save it in CPMpath.net. Using in R the function description we produce a CSV file with a list of works in this network:
setwd("C:/Users/batagelj/work/Python/WoS/BM/results/cite")
source("C:\\Users\\batagelj\\work\\Python\\WoS\\peere1\\description.R")
T <- read.csv('../../titles.csv',sep=";",colClasses="character")
T$code <- 1
dim(T)
d <- description("CPMpath.net","CPMpath.csv",T)
head(d)
that can be used in the interpretation of the network. It can be examined in Excel. We sort the list by the year.
Because of the contraction of some groups of equivalent nodes some NA fields can appear.
Taiwan approach
Network/Acyclic Network/Create (Sub)network/CPM/Global Search/Key-Route [1-150]
List of papers
For easier interpretation we produce also a CSV (use Excel as a viewer) file with details about papers contained in the CPM path and in the “Taiwan” subnetwork. See also island 10 papers.
> setwd("C:/Users/batagelj/work/Python/WoS/BM/")
> T <- read.csv('titles.csv',sep=";",colClasses="character")
> dim(T)
[1] 5698 6
> T$code <- 1
> source("C:\\Users\\batagelj\\work\\Python\\WoS\\peere1\\description.R")
> d <- description("./results/cite/CPMpath1.net","./results/cite/CPMpath.csv",T)
> head(d)
name WoSline author
1 ALBA_R(1973)3:113 459842 ALBA, RD
2 BREIGER_R(1974)53:181 462717 BREIGER, RL
3 BREIGER_R(1975)12:328 450395 BREIGER, RL
4 WHITE_H(1976)81:730 810 WHITE, HC
5 BOORMAN_S(1976)81:1384 459376 BOORMAN, SA
6 ARABIE_P(1978)17:21 1146 ARABIE, P
title
1 GRAPH-THEORETIC DEFINITION OF A SOCIOMETRIC CLIQUE
2 DUALITY OF PERSONS AND GROUPS
3 ALGORITHM FOR CLUSTERING RELATIONAL DATA WITH APPLICATIONS TO SOCIAL NETWORK ANALYSIS AND ...
4 SOCIAL-STRUCTURE FROM MULTIPLE NETWORKS .1. BLOCKMODELS OF ROLES AND POSITIONS
5 SOCIAL-STRUCTURE FROM MULTIPLE NETWORKS .2. ROLE STRUCTURES
6 CONSTRUCTING BLOCKMODELS - HOW AND WHY
journal year code
1 J MATH SOCIOL 1973 1
2 SOC FORCES 1974 1
3 J MATH PSYCHOL 1975 1
4 AM J SOCIOL 1976 1
5 AM J SOCIOL 1976 1
6 J MATH PSYCHOL 1978 1
>
> d <- description("./results/cite/taiwan.net","./results/cite/taiwan.csv",T)
> d <- description("./results/cite/island10b.net","./results/cite/island10.csv",T)
> d <- description("./results/cite/island9.net","./results/cite/island9.csv",T)
> d <- description("./results/cite/island7.net","./results/cite/island7.csv",T)
> d <- description("./results/cite/island2.net","./results/cite/island2.csv",T)
Link islands
Afterward we determine link islands of sizes [20, 200]. There are 10 islands.
10 7 9
2
3 5 6
1 4 8
From the picture we see that only island 10 7 9 and 2 have “interesting” structure.
Island 10
The island 10 on 200 nodes is unreadable. We reduced its size to 150 nodes. The maximal weight is 0.5785.
We get essentially the CPM path put in the context.
Comments based on island10.csv to be added.
Island 7
The island 7 has 74 nodes. The maximal weight is 4.9611 10-18
Papers from island 7 deal with landslides (some related to earthquakes). They are using “multi-block modeling of landslides”. The main journals are SOIL DYN EARTHQ ENG, ENG GEOL and LANDSLIDES.
Island 9
The island 9 has 44 nodes. The maximal weight is . The maximal weight is 2.416 10-14
Papers in this island deal with eartquake modeling. One among models is a “spring-block model”. The main journals are Phys Rev (A, E, lett), Physica A and J GEOPHYS RES.
Island 2
The island 2 has 33 nodes. The maximal weight is 2.462 10-19
From island.csv we see that papers from this island deal with numerical methods for computation of electromagnetic field. They use block model … Most papers are published in the journal IEEE T MICROW THEORY.
Positioning
I am trying to extend the application of Taiwan method to positioning of special topics in relation to mainstream research. Andrej added to Pajek (version 5.03) an option to construct all main paths through a given set (cluster) of works. For the network clustering literature I selected three examples:
- valued graphs (Žiberna's approach): { 3600, 3927, 9120, 10205, 10206, 11276, 11670, 3598, 11640 }
- signed graphs: { 1478, 4404, 4405, 4407, 4155, 4645, 6131, 8609 }
- geophysics: { 660, 437, 819, 1274, 1855 }
see the attached pictures.
main paths in Ziberna main paths in Signed main paths in Geophysics
I am playing with the new option
Main Paths/Global search/Through vertices in cluster
Some technical problems appeared because of very different weights in SPC nets. We can get cases where for a, b > 0 it holds a + b = a - adding a very small positive number doesn't change the result:
a <- 1
b <- 1e-20
c <- a+b
c-a
[1] 0
A temporary solution is to use the logarithms of the weights. With Andrej, we will try to find a better solution. In the case of selected nodes from GeoPhysics island we get the attached main paths network - as expected, it doesn't link to the main main path - it belongs to another (main) topic.
Citations among journals
Counting
February 2, 2018
JJ = t(WJ) * Ci * WJ
JJ(i,j) = # citations from papers published in journal i to papers published in journal j
read Cite Network/Create new network/Transform/1-mode to 2-mode = Cite2 read WJ Network/2-mode network/Transpose = WJt select Cite2 as Second network Networks/Multiply networks select WJ as Second network Networks/Multiply networks Network/Create new network/Transform/Remove/Selected vertices [1] = JJ
The work '*****' is making troubles. It is the node number 1. We delete it. Searching for link islands in [5, 50] we get 32 islands (50+13+11+2×10+3×9+5×8+2×7+4×6+13×5).
In the main island the weights are in the interval [103, 1472]. For visualization using the links width we transform the weights using the square root (and set the size of links to 0.1).
Fractional
February 2, 2018
JJf = t(WJ) * n(Ci) * WJ
JJf(i,j) = fractional contribution of citations from papers published in journal i to papers published in journal j
read Cite play macro norm1 = Cite1 Network/Create new network/Transform/1-mode to 2-mode = Cite2 read WJ Network/2-mode network/Transpose = WJt select Cite2 as Second network Networks/Multiply networks select WJ as Second network Networks/Multiply networks [yes] Network/Create new network/Transform/Remove/Selected vertices [1] Network/Create new network/Transform/Remove/Loops = JJ
