Publication: Comparative evaluation of a real network with network-level adjusted network generation model equivalents
Abstract
Sosyal ağ analizi, etkileşim halindeki nesneler içeren sosyal yapılar hakkında bilgi edinmek için ağ teorisini kullanan bir veri bilimi alanıdır. Düğümler ve bunları birbirine bağlayan bağlantıları içeren yapılar üzerine çalışır. Ağ analizinin araştırma alanlarından biri rastgele ağ modeli oluşturma konusudur. Rastgele ağ modelleri gerçek ağların tipik özellikleri hakkındaki soruları yanıtlamaya çalışır. Karmaşık ağlar oluşturmak veya ağların evrimini anlamak için çeşitli rastgele ağ modelleri geliştirilmiştir. Erdös-Renyi, Watts-Strogatz, Newman-Watts-Strogatz ve Barabasi-Albert ağ modelleri, farklı bağlanma mekanizmalarına sahip önemli rastgele ağ modelleridir. Bu tezde, oluşturduğumuz gerçek bir ağın ağ düzeyi ölçümleri ile ayarlanmış rastgele ağ model karşılıkları üretilmiş ve yapısal özellikleri gerçek ağımızla karşılaştırılmıştır. Çok yönlü gerçek ağları üretmede bu modellerin başarısı gerçek bir ağ üzerinden incelenmiştir. Rastgele ağ üretim modellerinde öne sürülen tercihli bağlanma mekanizmasının gerçek ağlarımız için çalışıp çalışmadığı araştırılmıştır. Oluşturulan gerçek ağların topolojik özellikleri ortaya konmuş ve bunlar ağ modeli üretiminde kullanılan ölçeksizlik ve küçük dünya özellikleri bakımından incelenmiştir. Barabasi-Albert modelinin gerçek ağ üretmede başarılı olduğu, diğer modellerden Erdös-Renyi modelinin birçok ölçekte diğer ağ üretim modellerinden üstün olduğu tespit edilmiştir.Bu tezde ayrıca ağ üretim modellerinin gerçek ağların derece dağılımlarını ve ağ görsellerini oluşturmadaki başarıları karşılaştırılmış olup Barabasi-Albert modelinin diğer modellerden daha başarılı olduğu tespit edilmiştir. Gerçek ağlarda sıklıkla görülen güç yasası derece dağılımının üslerinin genel olarak 2 ile 3 arasında yer aldığı bilinmektedir. Oluşturduğumuz gerçek ağların güç yasası derece dağılımlarının üsleri araştırılmış olup üslerin genel olarak 2’nin üzerinde ve ortalama 2.1 seviyesinde gerçekleştiği bulunmuş, literatürde genel kabul gören değerler doğrulanmıştır.
Social network analysis is a field of data science that uses network theory to learn about social structures that contain interacting objects. It explores structures that include nodes and links that connect them. One of the research areas of network analysis is the random network generation models. Random network models attempt to answer questions about typical features of real networks. Various random network models have been developed to create complex networks or to understand the evolution of networks. Erdös-Renyi, Watts-Strogatz, Newman-Watts-Strogatz and Barabasi-Albert network models are important random network models with different attachment mechanisms. In this thesis, counterparts of these random network models adjusted with network-level measures of our real networks are produced and their structural properties are compared with our real networks. The success of these models in generating our multi-faceted real networks have been examined. It is investigated whether the preferential attachment mechanism suggested in random network generation models works for our real networks. The topological properties of created real networks have been revealed and they have been examined in terms of scale-free and small-world properties. It has been determined that the Barabasi-Albert model is successful in producing real networks, while Erdös-Renyi model is superior to other network generation models in many measures.In this thesis, the success of network generation models in creating degree distributions and network visuals of real networks are compared, and it is determined that the Barabasi-Albert model is more successful than other models. It is known that the exponents of the power-law degree distribution, which is common in real networks, are generally between 2 and 3. The exponents of the power-law degree distributions of the real networks we have created are investigated and it is found that the exponents are generally above 2 and the average level is 2.1, so the generally accepted exponents in the literature have been confirmed.
Social network analysis is a field of data science that uses network theory to learn about social structures that contain interacting objects. It explores structures that include nodes and links that connect them. One of the research areas of network analysis is the random network generation models. Random network models attempt to answer questions about typical features of real networks. Various random network models have been developed to create complex networks or to understand the evolution of networks. Erdös-Renyi, Watts-Strogatz, Newman-Watts-Strogatz and Barabasi-Albert network models are important random network models with different attachment mechanisms. In this thesis, counterparts of these random network models adjusted with network-level measures of our real networks are produced and their structural properties are compared with our real networks. The success of these models in generating our multi-faceted real networks have been examined. It is investigated whether the preferential attachment mechanism suggested in random network generation models works for our real networks. The topological properties of created real networks have been revealed and they have been examined in terms of scale-free and small-world properties. It has been determined that the Barabasi-Albert model is successful in producing real networks, while Erdös-Renyi model is superior to other network generation models in many measures.In this thesis, the success of network generation models in creating degree distributions and network visuals of real networks are compared, and it is determined that the Barabasi-Albert model is more successful than other models. It is known that the exponents of the power-law degree distribution, which is common in real networks, are generally between 2 and 3. The exponents of the power-law degree distributions of the real networks we have created are investigated and it is found that the exponents are generally above 2 and the average level is 2.1, so the generally accepted exponents in the literature have been confirmed.