tensorkelechi/arxiv-cstext · Datasets at Hugging Face

text stringlengths 0 89.3k
surfing path its similarity with the users interests the re
sponse time and the surfing depth
We observe that both approaches are capable of finding rel
evant tweets that can potentially satisfy the users interests
However the main difference between the two approaches
resides in the score and the response time In fact we can
see that EEHOLSIF can achieve a higher score in almost all
cases Furthermore its response time is considerably faster
We believe that this gain in performance is the result of the
improvements we brought to the algorithm to make it able to
undertake large scale information foraging and in particular
the territory concept and the migration mechanism
Figure 13 displays the comparison results between EHO for
information foraging and enhanced EHO for large scale in
formation foraging in terms of relevance score surfing depth
convergence and response time This comparison was made
by testing 70 different users interests generated randomly
with both approaches Figure 13a shows that EEHOLSIF
is able to achieve better relevance scores with an average
of077against 065for EHOIF Moreover EEHOLSIF can
reach very high scores exceeding 09 while EHOIF is lim
ited to 074 The opposite can be said regarding the surfing
depth which is generally higher with EHOIF as shown in
Figure 13b This is due to the fact that in EHOIF the surf
ing process is initialized in a complete random way without
taking into consideration the content of the tweets On the
other hand the territories concept in EEHOLSIF allows to
target tweets similar to the users interest which helps bet
ter guiding the surfing and thus shortening the surfing paths
As for convergence we remark that EEHOLSIF has a faster
convergence rate while reaching a higher optimum compared
to EHOIF We can see from Figure 13c that EEHOLSIF con
verges after 25generations achieving a relevance score of
about 077 EHOIF on the flip side converges after 40gen
erations with a score of 065 This results confirm that the
migration mechanism introduced in EEHOLSIF helps pre
venting stagnation and hence allows the algorithm to reach
better solutions rapidly Figure 13d exhibits the run time re
sults which show that EEHOLSIF is remarkably faster in all
cases with an average response time of 09seconds against
265seconds for EHOIF
D Comparative study
A crucial step to validate our work is to compare it to other
metaheuristicbased information foraging approaches from
the literature In this section we pay particular attention to
two approaches
The first approach we consider is based on Ant Colony Sys
tem ACS and was already used to address Web information
foraging in 16 First we implemented ACS and adapted
it to tackle information foraging on social media Then we
conducted a series of tests to set ACS empirical parameters
with the aim of maximizing the systems performance Table
7 indicates the parameters values we fixed following theseEnhanced Elephant Herding Optimization for Large Scale Information Access on Social Media 42
a Similarity variation regarding αandβ
b Time variation regarding αandβ
c Similarity variation regarding the number of clans and ele
phants
d Time variation regarding the number of clans and elephants
e Similarity variation regarding the number of generations
f Time variation regarding the number of generations
Figure 12 Setting the empirical parameters for enhanced EHO for large scale information foraging43 Y Drias et al
EHOIF EHOKMIF
Users interests Most relevant surfing path ScoreTime
sSurfing
depthMost relevant surfing path ScoreTime
sSurfing
depth
Machine
Learning
IA PythonPython for Machine Learning and
Data Mining DeepLearning
datamining learning via
httpstcoqcC4wrx6m6
httpstcokLvs68HzEQ071 26 s 1Introduction To Machine
Learning with Python
MachineLearning deeplearning
learning via
httpstcolWfQGVjKXK
httpstcoZobXDvNWRO071 1 s 1
American
Express
free speech
democracyAmerican Express
httpstcoo9suYDdsV3064 27s 1American Express
httpstcobGaqvvZp69064 11s 1
Public
Relations
communicationA public relations strategy is
critical now more than ever
PublicRelations
httpstcoKZmGOD2Xjg054 29 s 1RT The Relevance
Of Public Relations
Communication In Fashion
httpstcoYrOhuv0mwX
fashion Fashionista
bloggerstr076 08s 1
COVID19
immunity
transmissionCoocoLa Vrej WHO is
still not sure if those

surfing path its similarity with the users interests the re

sponse time and the surfing depth

We observe that both approaches are capable of finding rel

evant tweets that can potentially satisfy the users interests

However the main difference between the two approaches

resides in the score and the response time In fact we can

see that EEHOLSIF can achieve a higher score in almost all

cases Furthermore its response time is considerably faster

We believe that this gain in performance is the result of the

improvements we brought to the algorithm to make it able to

undertake large scale information foraging and in particular

the territory concept and the migration mechanism

Figure 13 displays the comparison results between EHO for

information foraging and enhanced EHO for large scale in

formation foraging in terms of relevance score surfing depth

convergence and response time This comparison was made

by testing 70 different users interests generated randomly

with both approaches Figure 13a shows that EEHOLSIF

is able to achieve better relevance scores with an average

of077against 065for EHOIF Moreover EEHOLSIF can

reach very high scores exceeding 09 while EHOIF is lim

ited to 074 The opposite can be said regarding the surfing

depth which is generally higher with EHOIF as shown in

Figure 13b This is due to the fact that in EHOIF the surf

ing process is initialized in a complete random way without

taking into consideration the content of the tweets On the

other hand the territories concept in EEHOLSIF allows to

target tweets similar to the users interest which helps bet

ter guiding the surfing and thus shortening the surfing paths

As for convergence we remark that EEHOLSIF has a faster

convergence rate while reaching a higher optimum compared

to EHOIF We can see from Figure 13c that EEHOLSIF con

verges after 25generations achieving a relevance score of

about 077 EHOIF on the flip side converges after 40gen

erations with a score of 065 This results confirm that the

migration mechanism introduced in EEHOLSIF helps pre

venting stagnation and hence allows the algorithm to reach

better solutions rapidly Figure 13d exhibits the run time re

sults which show that EEHOLSIF is remarkably faster in all

cases with an average response time of 09seconds against

265seconds for EHOIF

D Comparative study

A crucial step to validate our work is to compare it to other

metaheuristicbased information foraging approaches from

the literature In this section we pay particular attention to

two approaches

The first approach we consider is based on Ant Colony Sys

tem ACS and was already used to address Web information

foraging in 16 First we implemented ACS and adapted

it to tackle information foraging on social media Then we

conducted a series of tests to set ACS empirical parameters

with the aim of maximizing the systems performance Table

7 indicates the parameters values we fixed following theseEnhanced Elephant Herding Optimization for Large Scale Information Access on Social Media 42

a Similarity variation regarding αandβ

b Time variation regarding αandβ

c Similarity variation regarding the number of clans and ele

phants

d Time variation regarding the number of clans and elephants

e Similarity variation regarding the number of generations

f Time variation regarding the number of generations

Figure 12 Setting the empirical parameters for enhanced EHO for large scale information foraging43 Y Drias et al

EHOIF EHOKMIF

Users interests Most relevant surfing path ScoreTime

sSurfing

depthMost relevant surfing path ScoreTime

sSurfing

depth

Machine

Learning

IA PythonPython for Machine Learning and

Data Mining DeepLearning

datamining learning via

httpstcoqcC4wrx6m6

httpstcokLvs68HzEQ071 26 s 1Introduction To Machine

Learning with Python

MachineLearning deeplearning

learning via

httpstcolWfQGVjKXK

httpstcoZobXDvNWRO071 1 s 1

American

Express

free speech

democracyAmerican Express

httpstcoo9suYDdsV3064 27s 1American Express

httpstcobGaqvvZp69064 11s 1

Public

Relations

communicationA public relations strategy is

critical now more than ever

PublicRelations

httpstcoKZmGOD2Xjg054 29 s 1RT The Relevance

Of Public Relations

Communication In Fashion

httpstcoYrOhuv0mwX

fashion Fashionista

bloggerstr076 08s 1

COVID19

immunity

transmissionCoocoLa Vrej WHO is

still not sure if those