Priporočilni modeli za izbor všečnih kolesarskih tur

MILAVEC, JAKA (author), Zupan, Blaž (mentor)

Cilj diplomske naloge je bil razviti zametek priporočilnega sistema za kolesarske poti. V ta namen smo najprej ustvarili podatkovno bazo kolesarskih poti v Sloveniji, ki smo jih pridobili iz opisov kolesarskih poti na različnih spletnih straneh. Zgradili smo besedilne opise poti ter poti opisali z numeričnimi atributi. Naš osnovni cilj je bil ovrednotiti metode, ki bi za neko izbrano referenčno pot oziroma njen opis v bazi poiskale referenci najbolj podobne poti. V ta namen smo zgradili tudi bazo referenčnih kolesarskih poti, in v namene vrednotenja tem ročno poiskali ustrezne najbolj podobne poti iz prej zgrajene podatkovne baze. Med sabo smo primerjali algoritme, ki podobnost ocenjujejo na atributnem zapisu in algoritme, ki za osnovni opis poti uporabljajo besedila ter ta vlagajo v vektorske prostore. Izsledki naše raziskave so pokazali, da je pristop z atributnimi zapisi učinkovitejši in bolj uporaben za iskanje podobnih kolesarskih poti, kar lahko prispeva k razvoju boljših priporočilnih sistemov v prihodnosti.

• Language: Slovenian
• Keywords: kolesarske trase, metode primerjave, vektorske vložitve opisov
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2025
• PID: 20.500.12556/RUL-167904
• COBISS.SI-ID: 231424515
• Publication date in RUL: 19.03.2025

Secondary language

• Language: English
• Title: Prototypes of recommendation models for a selection of attractive cycling tours

Abstract

The aim of the thesis was to develop the outline of a recommendation system for cycle routes. To this end, we first created a database of cycling routes in Slovenia, which we obtained from descriptions of cycling routes on various websites. We built text descriptions of the routes and described the routes with numerical attributes. Our main goal was to evaluate methods that would find the most similar routes in the database for a given reference route or its description. To this end, we also built a database of reference cycling routes, and for the purpose of evaluating these, we manually retrieved the corresponding most similar routes from the previously built database. We compared between algorithms that evaluate similarity on attribute notation and algorithms that use text as the basic route description and embed this in vector spaces. The results of our study showed that the attribute-based approach is more efficient and more useful for finding similar cycling routes, which may contribute to the development of better recommender systems in the future.

• Keywords: cycling routes, comparisson methods, description embeddings