Identifying Most Relevant Node Path To Increase Connection Probability In Graph Network

Abhiram Gandhe & Parag Deshpande
International Journal of Data Engineering (IJDE), Volume (6) : Issue (1) : 2015 1
Identifying Most Relevant Node Path To Increase Connection
Probability In Graph Network
Abhiram Gandhe abhiram.gandhe@gmail.com
Computer Science and Engineering
Visvesvaraya National Institute of Technology
Nagpur, India
Parag Deshpande psdeshpande@cse.vnit.ac.in
Computer Science and Engineering
Visvesvaraya National Institute of Technology
Nagpur, India
Abstract
In social networks, one of the most challenging problems is to find the best way to establish a
relationship between two nodes. Different attributes (Topological, Non-Topological) can be used
to define friendship score between two nodes which indicates the strength of a relationship. Non-
Topological attributes can be used to define the strength of a relationship even if two nodes are
not connected. The concept of friendship score to define the strength of a relationship between
two nodes transforms social network into a complete graph where each node is connected to
every other node and where friendship score is used as link attribute. The information on already
existing connections in social media network and graph which is formed based on friendship
score can be used to find out best way of connecting two different nodes even if no path is in
existence in social media network between these nodes.
In this paper, we propose a novel way of estimating friendship score using non-topological
attributes based on available information in social media network and algorithm to find out best
way of connecting two nodes in the form of chain of reference. The chain of reference between
node X1 and Xn is a path X1->X2->….->Xn-1->Xn where each link Xi->Xj is having high
friendship score. The chain of reference indicates how X1 can be connected to Xn even if no path
exists between X1 and Xn in social media network.
Keywords: Friendship Link, Online Social Network, Graph Network, Node Path, Reference
Chain.
1. INTRODUCTION
Online Social Networks (OSN) like Facebook, Google+, Twitter, LinkedIn, etc. are very dynamic.
With the growing influence of the social network and their far-fetched reach, every individual is
trying to get the maximum out of his social network. OSN have million to billions of users
communicating and connecting with each other on a daily basis. Every individual is on a lookout
for good information of his social network and more and more connect on the go.
While the connections in the network grow and more relevant and interesting individuals get
added to the network, it is always a demand for tools to look for the best and most relevant ways
of connecting with the people of interest. The friend recommendations in today’s OSN are
restrictive in the way they suggest connects.
In most of the OSNs, connects are suggested by:
a) Identifying two unconnected individuals with maximum mutual friends
b) Mutual friends are asked to suggest connects between unconnected friends

c) Unconnected individuals having multiple short length paths in the graph are suggested a
connect
d) Unconnected individuals commenting on the same conversation, multiple times are
suggested a connect
There is no way an individual to find out the best way or a recommended path to a distant
individual who is not in his network or in FOAF (Friend of a Friend) network [5]. LinkedIn has a
function to “get introduced”, but that is only limited to FOAF network. Anything above the second
level contact doesn’t have a way to get introduced. If person X who is a Software Developer in
Bangalore is looking forward to connecting with Y, who is a software programmer in Brazil there
is no way to find out a path of reference. In this paper we propose an algorithm, which with the
help of Friendship Score (fScore), a score obtained to identify the probability of friendship
between to unconnected individuals, will provide the most relevant Friendship Link, on which if
chain of reference is invoked, can provide the best possibility of attaining a connect.
If a node needs to connect to a distant unconnected node, the best way to do it is to get
introduced to the node of interest through the already existing linked node path from source node
to destination node. If a source node is connected to destination node by multiple unknown paths
then it is always a best way to get connected to the destination node by invoking introduction
reference along the most relevant path from source node to destination node. To find such a
node path the most important steps would be:
a) Finding multiple paths existing between source node and destination node
b) Most relevant path of all the existing paths between source node and destination node
The Importance of the problem, as stated in the paper “The small world problem” [6], brings under
discussion the network of acquaintances and friends in the social structure. The two schools of
thought about the small world problem are:
i. Two individuals how much ever remotely situated they are, they are linked to each other
by a finite number of links between acquaintances and friends and the number of
intermediate links between them are relatively small than expected.
ii. There are unbridgeable gaps between various groups and, therefore, the two distant
individuals will never link up. A message from one circle will never be able to jump to the
other circle.
In this paper, we do not take sides of any of the schools of thought stated above. Here, we
propose to find out the most relevant path between two individuals which with recommendation
and connects followed, will lead to a friendship between the two individuals or else report that no
relevant or strong path exists between the two individuals.
2. RELATED WORK
There is a variety of local similarity measures [5] (i.e. FOAF algorithm, Adamic/Adar index,
Jaccard Coefficient, etc.) for analyzing the “proximity” of nodes in a network and suggesting links
and friendship. FOAF [2] is adopted by many popular OSNs, such as facebook.com and hi5.com
for the friend recommendation task. FOAF is based on the common sense that two nodes x, y are
more likely to form a link in the future if they have many common neighbors. In addition to FOAF
algorithm, there are also other local-based measures such as Jaccard Coefficient [5] and
Adamic/Adar index [1]. Adamic and Adar have proposed a distance measure to decide when two
personal home pages are strongly “related”. There is a variety of global approaches in paper[5].
As comparison partners of global approaches, we consider Katz status index [4], Random Walk
with Restart algorithm [9] and RWR algorithm [7]. Katz defines a measure that directly sums over
all paths between any pair of nodes in graph G, exponentially damped by length to count short
paths more heavily. RWR considers a random walker that starts from node x, and chooses

randomly among the available edges every time, except that, before he makes a choice, with
probability α, he goes back to node x (restart).
3. PROBLEM STATEMENT
A person ‘A’ in social network wishes to connect unknown person ‘B’ because of some
professional interest, what will be the best path to connect person Y?
In most of the OSNs today the request won’t be sent by the OSN functionality as “A” doesn’t
directly know “B”. If “A” sends a friend request to “B” then even if the request lands with “B”, “B”
won’t accept the friendship request as “B” doesn’t know “A”. The only way by which “B” will
consider “A”’s request is when “A” is recommended or introduced to “B” by some of “B”’s
acquaintance or friend.
In this paper, we propose a novel method to identify the most relevant path between “A” and “B”
when the recommendations/introductions can be forwarded by the nodes on the path, with least
cost and better accuracy, to form the most reliable friendship path between “A” and “B”. If A and
B exist in disjoint graphs or have weak linkage path, then the method proposed will specify that
“No reliable and strong reference path exists between A and B at this moments snapshot of the
network”.
4. METHODOLOGY
4.1 Friendship Score (fScore)
Non-Topological Attribute values of two nodes help determine Friendship Score (fScore) [12]. If
two individuals have Same Gender then a specific weighted score is added to the fScore. In the
paper titled “Use of Non-Topological Node Attribute values for Probabilistic Determination of Link
Formation”, we have detailed the mathematical models to calculate fScore between two
individuals.
If a Link is present between two nodes, a and b, then:
Where FNT(a,b) is a function non-topological attribute of a and b. If a link is not formed between
two nodes, a and b, then friendship score needs to be calculated using non-topological attribute
data. The fScore is calculated as:
FNT(a,b) is a weighted sum of non-topological attributes. The weights are calculated by collecting
sample data from the Facebook. The sample data for calculating weights in function consists of
7637 user data and 101904 existing friend links. The mathematical formula, to calculate the
function value
Where:
i: Attributes/Features e.g. Gender (if there is no value associated with the attributes in any/both of
the nodes a or/and b then that attribute will not be taken up for the calculation of fScore)
j: Different Attribute Values possible e.g. Same Gender, Different Gender (No gender available is
also a valid possible value, but it will already be excluded from the equation because of the
elimination of attributes while collating the final set of i’s )

Attribute Value Weights calculated using Conditional Probability Bayes rule on the existing data
set from Facebook:
W (Same School) 0.0235561 W (Same Gender) 0.0047786
W (Same Location) 0.0106998 W (Different Gender) 0.0038902
W (Different Favorite Athlete) 0.0064703 W (Different Location) 0.0037144
W (Same Favorite Athlete) 0.0055419 W (Different School) 0.0032468
So, formula for finding fScore for two nodes, having values for gender, location, School and
favorite athlete attributes, is as follows:
Topological attributes can also be used to effectively to calculate friendship score [13]. We have
used Non Topological attribute Value between two nodes to calculate friendship score [12].
Method to use Topological Attribute to calculate friendship measure [13] combined with the
method suggested above to calculate Friendship score using Non Topological attribute values
can result in a more inclusive friendship score calculation. Calculation of Friendship Score
(fScore) is not within the purview of this paper. We will be going ahead with the Non Topological
Function FNT equation stated above.
5. PROPOSED MODEL
We propose a novel approach, which uses fScore, as the linkage probability predictor, will
provide an existing path in the network, from source to destination, which can be termed as the
most relevant path for invoking chain of reference.
5.1 Detailing Steps In The Proposed Approach
Identify most relevant path from Source to destination.
Source Destination
A B
Step 1:
Are A and B Friends (fScore [A, B] >= 1)?
i. Yes: A -> B has an existing friendship. No Need to proceed further.
ii. No: Process to Step 2.
Step 2: Initialization
Configure Algorithm Parameters: Max Hops, Min fScore Threshold, Qualifying percentage
Initialize Source Set with Source Node and Destination Set with Destination node.
Level = 0
Source Set Destination Set
A B
Step 3:
a. Identify Neighbors for nodes in Destination Set at this level
NB={NB1, NB2, ---, NBi}
i connected neighbors of B
b. Get fScore of all neighbors with all nodes in Source Set
c. Sort the resultant list with the decreasing order of fScore

d. Trim the sorted list by taking only top Qualifying percentage of the list with maximum
values of fScore selected.
e. Remove all the items in the trimmed list that have fScore less than Min fScore Threshold.
f. Check for duplicates in the list and remove the duplicates, retaining maximum fScore in
the trimmed list
g. Add the trimmed list to the destination set.
TNB = Trimmed NB
A B + TNB
Step 4:
a. Check in the destination set if there is any node with fScore more than or equal to 1
a. Yes: Path is identified from Source to Destination which has highest relevance w.r.t.
fScore
i. Process the Source Set and Destination Set to print relevant path
ii. Exit
b. No: Continue with next steps
Step 5:
a. Identify Neighbors for nodes in Source Set at this level
NA={NA1, NA2, ---, NAj}
j connected neighbors of A
b. Get fScore of all neighbors with all nodes in Destination Set
c. Sort the resultant list with the decreasing order of fScore
d. Trim the sorted list by taking the only top Qualifying percentage of the list with maximum
values of fScore selected.
e. Remove all the items in the trimmed list that have fScore less than Min fScore Threshold.
f. Check for duplicates in the list and remove the duplicates, retaining maximum fScore in
the trimmed list
g. Add the trimmed list to source set.
TNA = Trimmed NA
A +TNA B + TNB
Step 6:
a. Check in the source set if there is any node with fScore more than or equal to 1
a. Yes: Path is identified from Source to Destination which has highest relevance w.r.t.
fScore
i. Process the Source Set and Destination Set to print relevant path
ii. Exit
b. No: Continue with next steps
Step 7:
If no path is identified, till this step:
a. Level ++;
b. Process Step 3,4,5,6 for the next hop with set of neighbors for this level.
Step 8:
Is the path identified?”
a. Yes: We have the most relevant path
No: We can process again by changing the configuration algorithm parameter (Max Hops, Min
fScore Threshold, Qualifying percentage) values

6. ALGORITHMIC REALIZATION
The proposed approach is realized in the form of an algorithm. The algorithm, Relevant Path
algorithm, will propose the best path, which when traversed from source to destination, will
provide the most relevant path. The Path given as the output of the algorithm is most relevant as
fScore is the parameter used to weigh the relevance of the node considered for finding the
linkage path.
We have put below the components, which will give the Relevant Path algorithm outline with all
the other components needed to execute Relevant Path algorithm.
 fScore function calculates the Friendship Score between two nodes as stated above in this
paper
 getRelevantNeighbors function provides the list of relevant neighbors as detailed in step 2
and 4 above
 getRelevantConnectPath function using the getRelevantneighbor and fScore function finds
out the most relevant path, inside the network, from source to destination
 printRelevantPath just helps to print the path identified by getRelevantConnectPath
Relevant Path algorithm when processed on a dataset obtained from Facebook® using graph
query, the same data set which was used to obtain the weights stated above. After processing
the above algorithm on the network to find out relevant path for few set of nodes, the results are
as follows.
6.1 Configuration Parameter Values
max_hops = 6
minfScore =0.001
slab_percentatge = 10
Query 1:
getReleventConnectPath('651556926', '540415531')
Output:
Source Set
Level 0: fScore=>0.004552525, Node ID =>651556926
Destination Set
Level 1: fScore =>1.0039132666667 Node ID =>619017668
Level 0: fScore => 0.004552525 Node ID =>540415531
Query 2:
getReleventConnectPath('100002435033939', '100001284823018')
Output:
Source Set
Level 0: fScore] => 0 Node ID =>100002435033939
Destination Set
Level 0: fScore] => 0 Node ID =>100001284823018
Query 3:
getReleventConnectPath('1641471744', '895005331');

Output:
Source Set
Destination Set
7. CONCLUSION AND FUTURE SCOPE
Relevant Path Algorithm proposed here, using fScore between nodes as a linkage probability
predictor, identifies a network path, from the source node to the destination node, as the most
relevant path to invoke the chain of reference. This relevant path identified using Relevant Path
algorithm will yield the best result on invocation of chain of reference on the path than any other
network path between source node and destination node.
The other algorithms to predict friendship using a network path used in graph networks is very
restrictive. The example of the same is FOAF (Friend of a Friend) algorithm. It restricts the
traversal of the network only to the depth of 2 from the source network. Katz Algorithm defines a
measure that directly sums over all paths between any pair of nodes in the graph and is
exponentially damped by length to count short paths more heavily. Other algorithms for shortest
path identification using edge weights, in our case fScore, like Bellman-Ford Algorithm need to
calculate the path from each node to every other node. With continuous changes in the weights
due to change in any Node attribute value makes the maintenance of the resultant matrix in this
algorithm continuously changing due to which it has huge computational needs.
The proposed Relevant Path Algorithm is a real-time path calculation algorithm and is capable of
accommodating the continuous changes in the network topology and Node attribute values. It is
very much possible that a different path can be proposed by relevant path algorithm between the
same set of source node and destination node at two different time instances t1 and t2.
In this proposed method, Relevant Path Algorithm, specifies the strength of a friendship link using
Non Topological attributes. The proposed method, bases the relevant path algorithm on strength
of the link. We have proposed an algorithm to find the most relevant path algorithm which uses
fScore between nodes as a linkage probability predictor. The algorithm proposed can be
strengthened more by enriching the formula for fScore calculation.
In the future scope, the strength of the link can also incorporate topological attributes along with
the non-topological attribute. The integrated model, thus designed, will strengthen the relevance
of the path, by taking topological along with Non Topological attributes.
8. REFERENCES
[1] L. Adamic and E. Adar. How to search a social network. Social Networks, 2005, pp. 187–
203.
[2] J. Chen, W. Geyer, C. Dugan, M. Muller, and I. Guy. Make new friends, but keep the old:
recommending people on social networking sites. Proceedings of the 27th international
conference on Human factors in computing systems, 2009, pp. 201–210.
[3] K. C. Foster, S. Q. Muth, J. J. Potterat, and R. B. Rothenberg. A faster katz status score
algorithm. Compute Math Organ Theory, Dec 2001, pp. 275–285,

[4] L. Katz. A new status index derived from sociometric analysis. Psychometrika, 1953, pp. 39-
43.
[5] D. Liben-Nowell and J. Kleinberg. The link prediction problem for social networks.
Proceedings of the 12th International Conference on Information and Knowledge
Management (CIKM), 2003
[6] S. Milgram. The small world problem. PsychologyToday, 1967, pp. 61–67.
[7] J. Pan, H. Yang, C. Faloutsos, and P. Duygulu. Automatic multimedia cross-modal correlation
discovery. Proceedings of the 10th ACM SIGKDD international conference on Knowledge
discovery and data mining, 2004, pp. 653– 658.
[8] F. Rubin. Enumerating all simple paths in a graph. IEEE Transactions on Circuits and
Systems, 1978, pp. 641–642.
[9] H. Tong, C. Faloutsos, and J. Pan. Fast random walk with restart and its applications. In
ICDM ’06: Proceedings of the 6th International Conference on Data Mining, 2006, pp. 613–
622.
[10] S.Wasserman and K. Faust. Social network analysis: Methods and applications. 1994.
[11] A. Papadimitriou, P.Symeonidis and Y. Manolopoulos. Friendlink: Link Prediction in Social
Networks via Bounded Local Path Traversal. International Conference on Computational
Aspects of Social Networks (CASoN), 2011.
[12] A. Gandhe, P. Deshpande. Use of Non-Topological Node Attribute values for Probabilistic
Determination of Link Formation. International Journal of Advanced Computer Science and
Applications(IJACSA), Volume 6 Issue 2, 2015, PP. 186-191.
[13] M. Fire, L. Tenenboim, O. Lesser, R. Puzis, L. Rokach, Y. Elovici. Link Prediction in Social
Networks using Computationally Efficient Topological Features. IEEE International
Conference on Privacy, Security, Risk, and Trust, and IEEE International Conference on
Social Computing, 2011.

Identifying Most Relevant Node Path To Increase Connection Probability In Graph Network

More Related Content

What's hot (20)

Similar to Identifying Most Relevant Node Path To Increase Connection Probability In Graph Network (20)

Recently uploaded (20)

Identifying Most Relevant Node Path To Increase Connection Probability In Graph Network