Social Link Prediction

Rezvanian, Alireza; Moradabadi, Behnaz; Ghavipour, Mina; Khomami, Mohammad Mehdi Daliri; Meybodi, Mohammad Reza

doi:10.1007/978-3-030-10767-3_6

Cited by 3 publications

(18 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Since the attributes of vertices are generally hidden, we use another type of similarity named structural similarity that is based on the network structure and does not need global topological information, thereby needing less computational time. These indices are typically based on the number of common neighbours and mainly discussed in link prediction studies in online social networks [40]. Here, we introduce seven local similarity indices, including common neighbours (CN), Salton, Jaccard, Sorensen, Hub promoted index (HPI), Hub depressed index (HDI), and Leicht–Holme–Newman (LHN) index as the following description: (i) CN index [40].…”

Section: Methodsmentioning

confidence: 99%

“…These indices are typically based on the number of common neighbours and mainly discussed in link prediction studies in online social networks [40]. Here, we introduce seven local similarity indices, including common neighbours (CN), Salton, Jaccard, Sorensen, Hub promoted index (HPI), Hub depressed index (HDI), and Leicht–Holme–Newman (LHN) index as the following description: (i) CN index [40]. CN is one of the simplest measures for computing similarity between vertex

v_{i}

and vertex

v_{j}

by counting directed overlapping neighbours.…”

Section: Methodsmentioning

confidence: 99%

“…

s_{x y}^{CN} = |Γ (v_{i}) \cap Γ (v_{j})|

where let

normalΓ ()(, v_{i})

is the set of neighbouring vertices of the vertex

v_{i}

. Intuitively, two similar vertices have many common neighbours. (ii) Salton index [40]. This index also called the cosine similarity, which normalises the result of CN index by considering the degrees of vertices

s_{x y}^{Salton} = \frac{(||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{\sqrt{k_{v_{i}} \times k_{v_{j}}}}

where

k_{v_{i}}

and

k_{v_{j}}

are the degrees of vertex

v_{i}

and vertex

v_{j}

, respectively. (iii) Jaccard index [40].…”

Section: Methodsmentioning

confidence: 99%

“…This index also called the cosine similarity, which normalises the result of CN index by considering the degrees of vertices

s_{x y}^{Salton} = \frac{(||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{\sqrt{k_{v_{i}} \times k_{v_{j}}}}

where

k_{v_{i}}

and

k_{v_{j}}

are the degrees of vertex

v_{i}

and vertex

v_{j}

, respectively. (iii) Jaccard index [40]. This index is known as a popular similarity concept for many domains and it defines as

s_{x y}^{Jacard} = \frac{(||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{(||, normalΓ false(v_{i} false) \cup normalΓ false(v_{j} false))}

This index is similar to the common neighbour index; the difference is that in the Jaccard index the vertices are more similar if they have a large number of common neighbours and a small number of non‐common neighbours. (iv) Sorensen index [40]. This index is similar to CN index; however the numerator of common neighbours is divided by the sum the degrees of vertices as follows:

s_{x y}^{Sorensen} = \frac{2 (||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{k_{v_{i}} + k_{v_{j}}}

(v) HPI […”

Section: Methodsmentioning

confidence: 99%

“…HPI is defined as follows:

s_{x y}^{HPI} = \frac{(||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{\min (}{, k_{v_{i}}, k_{v_{j}})}

(vi) Hub depressed index (HDI) [40]. Similar to the HPI, another modification can also be considered for high‐degree vertices as follows:

s_{x y}^{HDI} = \frac{(||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{\max (}{, k_{v_{i}}, k_{v_{j}})}

(vii) LHN index [40]. The LHN index, similar to all introduced similarity index, considers the number of common neighbours for the numerator and normalises the result by dividing on multiplying degrees of vertices as follows:

s_{x y}^{LHN} = \frac{(||, normalΓ false(v_{i} false) \cap normalΓ false(v_{j} false))}{k_{v_{i}} \times k_{v_{j}}}

…”

Section: Methodsmentioning

confidence: 99%

See 4 more Smart Citations

AntLP: ant‐based label propagation algorithm for community detection in social networks

Hosseini

Rezvanian

2020

CAAI trans. intell. technol.

Self Cite

View full text Add to dashboard Cite

In social network analysis, community detection is one of the significant tasks to study the structure and characteristics of the networks. In recent years, several intelligent and meta‐heuristic algorithms have been presented for community detection in complex social networks, among them label propagation algorithm (LPA) is one of the fastest algorithms for discovering community structures. However, due to the randomness of the LPA, its performance is not suitable for the general purpose of network analysis. In this study, the authors propose an improved version of the label propagation (called AntLP) algorithm using similarity indices and ant colony optimisation (ACO). The AntLP consists of two steps: in the first step, the algorithm assigns weights for edges of the input network using several similarity indices, and in the second step, the AntLP using ACO tries to propagate labels and optimise modularity measure by grouping similar vertices in each community based on the local similarities among the vertices of the network. In order to study the performance of the AntLP, several experiments are conducted on some well‐known social network datasets. Experimental simulations demonstrated that the AntLP is better than some community detection algorithms for social networks in terms of modularity, normalised mutual information and running time.

show abstract