Profiliranje uporabnikov in dinamično priporočanje produktov z vektorskimi bazami

Derenda Cizel, Denis (author), Žitnik, Slavko (mentor)

Količina podatkov se iz dneva v dan povečuje. Z namenom filtriranja velikega toka podatkov so bili razviti različni priporočilni sistemi, ki izvajajo preslikavo med uporabniki in predmeti priporočanja z namenom čim hitrejše interakcije med njimi. V magistrskem delu se posvetimo priporočilnim sistemom na podlagi sodelovanja in delovanje preverimo na podatkih o telekomunikacijskih storitvah uporabnikov. Priporočanje ovrednotimo z različnimi merami uspešnosti. Sodelovalno priporočanje z namenom izboljšanja priporočanja nadgradimo v različne hibridne pristope. Hibridni pristop z dodatkom demografskih podatkov pravilno predlaga 85 odstotkov uporabniških priporočil. Z upoštevanjem zaporedja interakcij je mogoče pravilno napovedati naslednjo uporabniško storitev v 74 odstotkih. Implementirano je bilo tudi shranjevanje vektorskih predstavitev v vektorsko bazo, ki naredi priporočilni dostop bolj dostopen za uporabo.

• Language: Slovenian
• Keywords: priporočilni sistemi, profiliranje, vsebinsko osnovana metoda, metoda izbiranja s sodelovanjem, vektorska baza
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2023
• PID: 20.500.12556/RUL-149118
• COBISS.SI-ID: 163977219
• Publication date in RUL: 04.09.2023

Secondary language

• Language: English
• Title: User profiling and dynamic product recommendation with vector databases

Abstract

The amount of data is increasing day by day. In order to filter a large flow of data, various recommendation systems have been developed, which perform the mapping between users and items for recommendation to facilitate faster interaction between them. This master's thesis focuses on collaborative filtering-based recommendation systems and verifies their performance using telecommunications service user data. The recommendations are evaluated using various performance measures. The collaborative filtering method is enhanced with different hybrid approaches to improve recommendations. By incorporating demographic data into the hybrid approach, the developed recommendation system correctly suggests 85 percent of user recommendations. Considering the sequence of interactions, the next user service can be accurately predicted in 74 percent of cases. Additionally, storage of vector representations in a vector database has been implemented, making the recommendation access more user-friendly.

• Keywords: recommender systems, profiling, content-based method, collaborative filtering method, vector database