magistrsko delo
Nina Mrzelj (Author), Lovro Šubelj (Mentor)

Abstract

Veliko kompleksnih sistemov iz realnega sveta lahko predstavimo z uporabo heterogenih omrežij. Algoritmi za napovedovanje povezav uporabljajo informacijo o strukturi omrežja za identifikacijo manjkajočih podatkov ali napovedovanje povezav, ki se bodo z veliko verjetnostjo pojavile v prihodnosti. V magistrski nalogi implementiramo in med seboj primerjamo različne modele za napovedovanje povezav v heterogenih omrežjih, ki temeljijo na predstavitvah omrežja v vektorskem prostoru. Uporabimo preprost model z ročnim načrtovanjem značilk, pristop, ki temelji na podlagi naključnih sprehodov po meta poteh v omrežju, in globoke metode za učenje na homogenih in heterogenih omrežjih. Metode eksperimentalno vrednotimo na štirih realnih podatkovnih množicah, v katerih se pojavijo različne vrste povezav. Kot metriko za merjenje uspešnosti metod uporabimo površino pod krivuljo ROC, rezultate pa med seboj primerjamo z uporabo neparametričnih testov, kot sta Friedmanov test in post-hoc Nemenyi test. Rezultati kažejo, da so za reševanje problema najprimernejše konvolucijske mreže grafov, prilagojene za heterogena omrežja. Slabost pristopov globokega učenja pa je, da ne moremo utemeljiti njihovega sprejemanja odločitev, zato so včasih primernejši v kombinaciji z drugimi metodami.

Keywords

analiza omrežij;heterogena omrežja;napovedovanje povezav;konvolucijske mreže grafov;

Data

Language: Slovenian
Year of publishing:
Typology: 2.09 - Master's Thesis
Organization: UL FRI - Faculty of Computer and Information Science
Publisher: [N. Mrzelj]
UDC: 004.42
COBISS: 18656345 Link will open in a new window
Views: 1234
Downloads: 316
Average score: 0 (0 votes)
Metadata: JSON JSON-RDF JSON-LD TURTLE N-TRIPLES XML RDFA MICRODATA DC-XML DC-RDF RDF

Other data

Secondary language: English
Secondary title: Link prediction in heterogenous networks by embedding nodes in the vector space
Secondary abstract: Many complex real-world systems can be modeled as heterogeneous networks. Link prediction in such networks can be used to detect missing information or predict future relationships based on currently observed connections. In the thesis, we compare various methods for the task of link prediction on heterogeneous networks. We implement four different models, all based on embeddings of nodes in vector space. We compare a simple model with manually selected link features, a method based on random walks on meta paths in the graph and an autoencoder model with graph convolutional networks for homogenous networks and its adaptation for heterogeneous networks. Area under ROC curve is used to evaluate algorithms' performance. We conduct experiments on four real-world datasets, resulting in various edge types to test on. To measure if the results between classifiers are statistically significant, non-parametric statistical tests such as the Friedman test and post-hoc Nemenyi test are used. Results show that graph autoencoder model modified for heterogeneous networks outperforms other methods. The main drawback of deep learning models is, that they are not interpretable and their process of decision making cannot be explained. In some cases, it is better to use them in combination with other methods.
Secondary keywords: network analysis;heterogeneous networks;link prediction;graph convolutional networks;
Type (COBISS): Master's thesis/paper
Study programme: 0
Embargo end date (OpenAIRE): 1970-01-01
Thesis comment: Univ. v Ljubljani, Fak. za matematiko in fiziko, Oddelek za matematiko, Računalništvo in matematika - 2. stopnja
Pages: XI, 57 str.
ID: 11159807