GRAPH CLUSTERING METHODS IN SOCIAL NETWORKS FOR BUILDING RECOMMENDATION SYSTEMS
DOI:
https://doi.org/10.26906/SUNZ.2019.2.129Keywords:
graph clustering, recommendation systems, social networks, modularity, graph labeling, random walk algorithmsAbstract
The subject matter of the article is the process of graph clustering in social networks. The goal is to investigate the graph clustering in methods social networks that can be used to build recommender systems for social media. The tasks to be solved are: to study the existing graph clustering methods in social network and investigate the possibility and feasibility of using them in recommender systems. The following results were obtained: The study of existing graph clustering methods in social networks of two types, to obtain clusters that do not intersect, and to obtain clusters that can intersect was conducted. The possibility of using the considered methods for building recommendatory systems of social media is investigated. The possibilities of the graph DBMS Neo4j in the implementation of graph clustering algorithms are investigated. Conclusions. The study was conducted on various graph clustering methods in social networks. Methods based on optimization of the graph modularity, on the graph labeling and on the methods of random walks and a separate group of methods that find on a graph into intersecting clusters are considered. The possibility and feasibility of using graph clustering methods for constructing recommender systems has been investigated. The possibilities of the graph database management system Neo4j for the implementation of graph clustering methods are investigated. It has been established that Neo4j provides wide possibilities for the implementation of the considered methods. In order to define clusters in a graph, the DBMS Neo4j offers several algorithms implemented in its Graph Algorithms library, namely the Louvain, Label Propagation and Triangle Counting algorithms. The functions that implement the Louvain, Label Propagation and Triangle Counting algorithms in Neo4j were tested. Other graph clustering algorithms need, if necessary, to be implemented on their own, but the DBMS Neo4j provides many convenient tools for working with data that can be used to implement various graph clustering algorithms with less effort than without using Neo4j.Downloads
References
Выделение сообществ в графе взаимодействующих объектов / М.И. Коломейченко, И.В. Поляков, А.А. Чеповский, А.М. Чеповский // Фундаментальная и прикладная математика. – Т. 21. № 3. – 2016. – С. 131–139.
Никишин Е.С. Методы выделения сообществ в социальных графах [Електронний ресурс] / Е.С. Никишин. – 2016. – Режим доступу: http://www.machinelearning.ru/ wiki/images/8/8a/Nikishin_coursework_community_detection.pdf.
Форман Д. Много цифр. Анализ больших данных при помощи Excel / Джон Форман. – М.: Альпина Паблишер, 2016. – 464 с.
Пархоменко П.А., Григорьев А.А., Астраханцев Н.А. Обзор и экспериментальное сравнение методов кластеризации текстов // Труды ИСП РАН. 2017. №2. [Електронний ресурс]. Режим доступу: https://cyberleninka.ru/article/n/obzor-ieksperimentalnoe-sravnenie-metodov-klasterizatsii-tekstov
neo4j [Електронний ресурс]. Режим доступу: https://neo4j.com/.
Мелешко Є.В. Дослідження методів побудови рекомендаційних систем в мережі Інтернет / Є.В. Мелешко, Г.С. Семенов, В.Д. Хох. // Збірник наукових праць "Системи управління, навігації та зв’язку". Випуск 1(47). – Полтава: ПНТУ ім. Ю. Кондратюка. – 2018. – С. 131–136.
Recommender Systems Handbook / Editors Francesco Ricci, Lior Rokach, Bracha Shapira, Paul B. Kantor. – 1st edition. – New York, NY, USA: Springer-Verlag New York, Inc., 2010. – 842 с.
Мелешко Є.В. Розробка рекомендаційної системи на базі субд neo4j. / Є.В. Мелешко, В.В. Босько, В.А. Резніченко // V Міжнародна науково-практична конференція "Інформаційні технології та взаємодії", 20-21 листопада 2018 року, м. Київ. – 2018. – С. 351–352.
