Ostvarenje sustava preporučivanja na temelju analize mreže sastavnica i podataka o popularnosti proizvoda

master's thesis

• Title (english): Recommendation System Implementation Based on Product Component Network Analysis and Product Popularity
• Author: Marin Jovanović
• Mentor: Klemo Vladimir (mentor)
• Committee member: Klemo Vladimir (predsjednik povjerenstva)
• Committee member: Marin Šilić (član povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-04, Croatia

Abstract

Algoritmi preporučivanja sadržaja imaju široku primjenu. Postoje razne vrste algoritama koje prihvaćaju različite ulazne varijable. Algoritmi najčešće primaju značajke i ocjene na temelju kojih uče preporučivati sadržaj. U ovom radu pažnja je stavljena na mreže (matrice) koje opisuju međuodnos sastojaka (sastavnica) te se takve matrice koriste kao značajke koje su algoritmu poznate, a normalizirane prosječne ocjene kao oznake što algoritam mora zaključiti. Opisano je kako se grade matrice. Za usporedbu s generičkim rješenjem koristi se algoritam suradničkog filtriranja.
Definirana je programska potpora za dohvaćanje skupa podataka s interneta, parsiranje i perzistentnu pohranu u bazi podataka. Opisano je kako se matrice mogu spremiti u relacijsku bazu podataka te kakve optimizacijske tehnike se mogu primjeniti nad tim matricama.
Koristeći dohvaćeni skup podatka napravljene su ulazne matrice te su predane na ulaz algoritmu regresor za povećanje gradijenta. Napravljene su ocjene mjera uspješnosti korištenjem MSE, RMSE i MAE te je dana usporedba algoritama međusobno i s algoritmom suradničkog filtriranja. Dano je tumačenje rezultata.
Opisana je arhitektura programske potpore za kreiranje matrica.
Otkrivene su zajednice sastojaka te su predloženi načini daljnjeg razvoja.

Abstract (english)

Content recommendation algorithms have a wide range of applications. There are various types of algorithms that accept different input variables. Algorithms usually receive features and ratings based on which they learn to recommend content. In this paper, attention is paid to networks (matrices) that describe the interrelationship of components, and such matrices are used as features that are known to the algorithm, and normalized average scores as markers of what the algorithm must conclude. It is described how matrices are built. A collaborative filtering algorithm is used for comparison with the generic solution.
Software for retrieving a set of data from the Internet, parsing and persistent storage in a database is defined. It is described how matrices can be saved in a relational database and what optimization techniques can be applied to these matrices.
Using the retrieved dataset, input matrices were created and submitted to the gradient boosting regression algorithm. Assessments of performance measures were made using MSE, RMSE and MAE, and a comparison of the algorithms with each other and with the collaborative filtering algorithm was given. An interpretation of the results is given.
The architecture of the software for creating matrices is described.
Communities of the ingredients were discovered, and ways of further development were suggested.

• Keywords: sustav preporučivanja; analiza mreža; analiza mreža sastavnica; popularnost proizvoda; optimizacija matrica; detektiranje zajednica
• Keywords (english): recommendation system; network analysis; component network analysis; product popularity; matrix optimization; community detection
• Language: croatian
• URN:NBN: urn:nbn:hr:168:713859
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-10-11 09:10:09