text
stringlengths
0
89.3k
0otherwise
else
Pci Sj dI m j
Pk
l1dI m l12
WherePci Sjis the probability to place clan cion cluster
Sj
dI m jis the euclidean distance between the users
interests Iand the centroid of cluster Sj
C Solution construction
Another improvement is related to the construction of the so
lution which is in our case a surfing path Unlike in the
original algorithm where the elephants consider readymade
solutions we give the task of constructing a solution to each
elephant starting from its initial position For this purpose
we incorporate information foraging concepts including the
information scent Formula 1 and the surfing decision rule
Formula 2 The solution construction process is given by
Algorithm 1
Note that during the solution construction an elephant can
leave its territory if the surfing path leads it towards a post
located on a neighbor territory We consider two territories
as neighbors if they share adjacent edges Two edges are ad
jacent if they are both incident with a common vertex
D Clan migration
In nature an elephant clan might separate from the larger
herd in response to limited food supplies encountered during
a dry season If food sources are scarce it is more efficient
for elephants to travel as individual clans rather than large
herds
We incorporate this natural phenomenon in the enhanced
EHO for large scale information foraging as a stagnation pre
vention mechanism We believe there is a strong analogy
between the lack of food sources in nature on one side and
the inability of the elephants to improve their solution after
several generations on the other We introduce a migration
parameter t0 which serves as a threshold that controls the
maximum number of generations a clan can spend without
improving its best solution If a clan exceeds this threshold
it migrates towards a new territory with the hope of finding
better solutions The migration is performed by choosing a
new cluster randomly and defining the migrating clan posi
tions within that cluster
The enhanced elephant herding optimization for large scale
information foraging is presented in Algorithm 6
VII Experiments
This section is organized in four subsections first we de
scribe the dataset we use in the evaluation We then present
the results obtained with the adapted EHO for information
foraging Next we follow up with the results of the enhanced
EHO for large scale information foraging Finally we finish
by doing a comparative study with other approaches from the
literature
All the experiments were implemented using Java program
ming language and were conducted on a laptop running Win
dows 10 with an Intel Core i54300M CPU at 260GHz and
6GB of RAM37 Y Drias et al
Algorithm 6 Enhanced EHO for large scale information foraging
Input I the users interests G the social graph
Output a list of surfing paths ranked by relevance
1Divide the search space into kdifferent territories using Algorithm 5
2Set the generations counter t1 the solutions list sols and the stagnation counter for each clan gci0
3Initialize empirical parameters αβq0t0 maximum generations MaxGen number of clans nClans population size
and number of elephants in each clan nci
4Initialize the positions of the elephants according to the users interests Iusing Formula 12 and with respect to distClan
anddistElephant
5while tMaxGen do
6 fori1to nClans do for all clans in elephant population
7 forj1toncido for all elephants in clan ci
8 Build the elephants jsurfing path using Algorithm 1
9 Calculate the elephants fitness using Formula 3
10 end for
11 ifclanciimproved its best solution compared to the previous generation then
12 Update bestSol ci
13 else
14 gcigci 1
15 end if
16 ifgcit0then
17 Migrate clan citowards a new territory chosen randomly
18 gci0
19 else
20 Update the positions of the elephants xcijusing Formula 4
21 Update the matriarchs position using Formula 6
22 Locate the worst elephant to leave clan ciaccording to the fitness function
23 Generate a new elephant in the clan ciusing Formula 7
24 end if
25 end for
26 Append the best surfing paths found in generation ttosols
27 Update the generation counter tt 1
28end while
29Return the best surfing paths ranked relevanceEnhanced Elephant Herding Optimization for Large Scale Information Access on Social Media 38
A Dataset description
We tested our algorithm on Twitter which is one of the most
popular social networks and microblogging platforms We
constructed a dataset composed of 1 410 246 tweets that we
grouped in one big social graph The data crawling was per
formed using NodeXL 35 and took place during the end of
2020 Figure 6 and Table 3 showcase the main topics covered
by the dataset
Figure 6 Topics covered by the dataset