Beyene, Melkamu (PhD)Terefe, Badimaw2020-12-222023-11-182020-12-222023-11-182020-06-06http://etd.aau.edu.et/handle/12345678/24262This thesis is about designing a community detection model on Facebook using relational data. It is mainly aimed at helping companies by grouping Facebook users based on their interests which has many applications. The researchers selected four algorithms based on the concept of modularity optimization/maximization. Modularity based algorithms were selected since modularity is the best quality measure of a community detection algorithm. The algorithms selected and then compared were the ‘girvan_newman’ algorithm, the ‘greedy_modularity_communities’ algorithm, the ‘fast_greedy’ algorithm, and the ‘Louvain’ algorithm. Each of these four algorithms was studied separately before, but no detailed study was done to show which algorithm best detects communities especially when it comes to relational learning. Five datasets required to undergo an experiment (five experiments since it is difficult to ensure an algorithm as the best algorithm by doing a single experiment) were extracted from Facebook and prepared into matrix and graph data structures. An attempt was made to select the best algorithm among the aforementioned algorithms and the results gained were cheering. The results gained in the experiment showed that the Louvain algorithm was the best in many aspects. First and for most, it is the one that detected communities with the best modularity value (average modularity of 0.491). Second, it is the algorithm that fulfilled all of the conditions to say an algorithm as the best algorithm such as ease of use, ability to detect overlapping communities, state of the art, support to the visualization of networks, and the possibility to run it on a single or distributed computing resource. Finally, it is also the one that best balances the speed quality tradeoff in community detection. That means, it is the algorithm that detected quality communities within an acceptable time. There were many challenges the researchers faced during community detection. But two of them were the most difficult ones. The first challenge was scrapping data from Facebook and the second one was understanding the data type algorithms need to perform community detection. The first challenge was solved by understanding the HTML structure of Facebook pages and the second one was solved by studying graph data structures and file extensions in detail.enFacebookCommunity DetectionModularity-Based Community DetectionThesis