tensorkelechi/arxiv-cstext · Datasets at Hugging Face

text stringlengths 0 89.3k
gunshotsFacebook groups are falling
apart over Black Lives
Matter posts
httpstcog0eff0hcLC
Socialmedia
httpstcoR3dov8nCAq059 28s 1Black Lives STILL Matter
just in case you forgot
And ALL Lives wont
matter until Black
Lives do
httpstcobULE7LLNk7068 1s 1
Table 6 Information Foraging Results EHOIF vs EEHOLSIFEnhanced Elephant Herding Optimization for Large Scale Information Access on Social Media 44
a Similarity score
b surfing depth
c Convergence
d Response time
Figure 13 EHOIF vs EEHOLSIF
tests
Parameter Value
α 02
β 04
ρ 08
q0 08
Number of ants 50
Number of generations 50
Table 7 ACSIF empirical parameters setting
The second approach we implemented is based on Particle
Swarm Optimization PSO To our best knowledge this is the
first time PSO is used to address information foraging Just
like with ACS we carried out several tests to find the optimal
values of the empirical parameters which are presented in
Table 8 We denote this approach as PSOIF
Parameter Value
c1 15
c2 04
Number of particles 600
Number of generations 90
Table 8 PSOIF empirical parameters setting
Once we fixed the empirical parameters for both approaches
we conducted extensive experiments to make a comparative
study between EHOIF EEHOLSIF ACSIF and PSOIF Fig
ure 14a exhibits the difference in relevance sore between the
four approaches We can see that EEHOLSIF achieves the
highest score followed by EHOIF then ACSIF and finally
PSOIF We also notice that the difference in score is quitesignificant between EEHOLSIF and the other approaches
When it comes to response time Figure 14a shows that EE
HOLSIF is the fastest with an average lower than 1 second
The three other approaches have an average response time
between 25 and 35 seconds This means that EEHOLSIF is
more than 25 times faster than the other approaches
We assume that the main problem faced by EHOIF ACSIF
and PSOIF resides in the social graphs size For instance
ACS is well adapted to work on graphs since it was first
developed to solve the traveling salesman problem It was
also applied to information foraging and gave good results
both in terms of score and response time However it was
only tested on limited size web graphs which contain less
than 2000 web pages 16 Once we increase the number
of the web pages or in our case social posts the complexity
of the problem causes a noticeable slow down in terms of
response time and a decrease in the relevance score
VIII Conclusion and perspectives
A novel bioinspired approach to large scale information
foraging using enhanced elephant herding optimization and
clustering was proposed in this paper First we adapted the
Elephant Herding Optimization algorithm to information for
aging on social media EHO was originally proposed to solve
continuous optimization problems so to make it able to work
on combinatorial problems and more precisely information
foraging we undertook modifications on some important as
pects including the elephants positions implementation the
solution construction and the solution evaluation To our best
knowledge this is the first attempt to use EHO to address in45 Y Drias et al
a Similarity
b Response time
Figure 14 Comparison with ant colony optimization
formation foraging The first results show that our approach
based on the adaptation of EHO has the ability to find rel
evant information on social graphs However the response
time can be very slow especially when the social graph is
big
To overcome this issue and better handle large scale informa
tion foraging we proposed a new enhanced version of EHO
We introduced several new concepts to the algorithm includ
ing two natural phenomena related to elephants behavior
namely territories delimitation and clan migration Cluster
ing and more precisely kmeans algorithm was used for the
implementation of the territories delimitation Thanks to this
phase we were able to define a better representation of the
elephants positions taking into account the real distance be
tween the potential solutions on the search space Further
more dividing the search space into multiple small territo
ries and bounding the foraging to just a few of them using a
newly introduced pseudo random proportional rule helped to
substantially reduce the problem complexity In addition to
that the clan migration phenomenon prevents the algorithm
from stagnation and premature convergence
In order to evaluate the proposed approach we built a dataset
containing more that 14 million tweets covering different
topics We conducted extensive experiments to test both

gunshotsFacebook groups are falling

apart over Black Lives

Matter posts

httpstcog0eff0hcLC

Socialmedia

httpstcoR3dov8nCAq059 28s 1Black Lives STILL Matter

just in case you forgot

And ALL Lives wont

matter until Black

Lives do

httpstcobULE7LLNk7068 1s 1

Table 6 Information Foraging Results EHOIF vs EEHOLSIFEnhanced Elephant Herding Optimization for Large Scale Information Access on Social Media 44

a Similarity score

b surfing depth

c Convergence

d Response time

Figure 13 EHOIF vs EEHOLSIF

tests

Parameter Value

α 02

β 04

ρ 08

q0 08

Number of ants 50

Number of generations 50

Table 7 ACSIF empirical parameters setting

The second approach we implemented is based on Particle

Swarm Optimization PSO To our best knowledge this is the

first time PSO is used to address information foraging Just

like with ACS we carried out several tests to find the optimal

values of the empirical parameters which are presented in

Table 8 We denote this approach as PSOIF

Parameter Value

c1 15

c2 04

Number of particles 600

Number of generations 90

Table 8 PSOIF empirical parameters setting

Once we fixed the empirical parameters for both approaches

we conducted extensive experiments to make a comparative

study between EHOIF EEHOLSIF ACSIF and PSOIF Fig

ure 14a exhibits the difference in relevance sore between the

four approaches We can see that EEHOLSIF achieves the

highest score followed by EHOIF then ACSIF and finally

PSOIF We also notice that the difference in score is quitesignificant between EEHOLSIF and the other approaches

When it comes to response time Figure 14a shows that EE

HOLSIF is the fastest with an average lower than 1 second

The three other approaches have an average response time

between 25 and 35 seconds This means that EEHOLSIF is

more than 25 times faster than the other approaches

We assume that the main problem faced by EHOIF ACSIF

and PSOIF resides in the social graphs size For instance

ACS is well adapted to work on graphs since it was first

developed to solve the traveling salesman problem It was

also applied to information foraging and gave good results

both in terms of score and response time However it was

only tested on limited size web graphs which contain less

than 2000 web pages 16 Once we increase the number

of the web pages or in our case social posts the complexity

of the problem causes a noticeable slow down in terms of

response time and a decrease in the relevance score

VIII Conclusion and perspectives

A novel bioinspired approach to large scale information

foraging using enhanced elephant herding optimization and

clustering was proposed in this paper First we adapted the

Elephant Herding Optimization algorithm to information for

aging on social media EHO was originally proposed to solve

continuous optimization problems so to make it able to work

on combinatorial problems and more precisely information

foraging we undertook modifications on some important as

pects including the elephants positions implementation the

solution construction and the solution evaluation To our best

knowledge this is the first attempt to use EHO to address in45 Y Drias et al

a Similarity

b Response time

Figure 14 Comparison with ant colony optimization

formation foraging The first results show that our approach

based on the adaptation of EHO has the ability to find rel

evant information on social graphs However the response

time can be very slow especially when the social graph is

big

To overcome this issue and better handle large scale informa

tion foraging we proposed a new enhanced version of EHO

We introduced several new concepts to the algorithm includ

ing two natural phenomena related to elephants behavior

namely territories delimitation and clan migration Cluster

ing and more precisely kmeans algorithm was used for the

implementation of the territories delimitation Thanks to this

phase we were able to define a better representation of the

elephants positions taking into account the real distance be

tween the potential solutions on the search space Further

more dividing the search space into multiple small territo

ries and bounding the foraging to just a few of them using a

newly introduced pseudo random proportional rule helped to

substantially reduce the problem complexity In addition to

that the clan migration phenomenon prevents the algorithm

from stagnation and premature convergence

In order to evaluate the proposed approach we built a dataset

containing more that 14 million tweets covering different

topics We conducted extensive experiments to test both