Search

Odkar so poslušalske platforme, kot je Spotify, postale prevladujoče sredstvo glasbene potrošnje, se le te zanašajo na priporočilne sisteme, da uporabnike pomagajo usmerjati po svojih vse obsežnejših glasbenih zbirkah. V preteklosti so bile pri tej nalogi najuspešnejše metode sodelovalnega filtriranja (angl. collaborative filtering, CF), ki izkoriščajo pretekle interakcije med uporabniki in artikli. Vseeno pa te metode niso brez pomanjkljivosti. Pogosto so npr. znatno manj uspešne med slabše poznanimi artikli z le malo podatki o interakcijah. Nasprotno modeli, ki temeljijo na vsebini, ta problem popolnoma zaobidejo, saj tvorijo priporočila izključno na osnovi vsebine artiklov (npr. zvočnega signala skladb), vendar pa take metode še niso primerljivo zmogljive. Strojno učenje z grafi --- rastoče področje, ki prekaša pretekle pristope v mnogih domenah --- morda lahko naslovi tudi tukajšno nalogo. Zlasti najnovejše metode, nevronske mreže na grafih (angl. graph neural networks, GNN), obljubljajo, da lahko učinkovito izkoristijo tako uporabniške informacije v obliki grafa kot tudi vsebino in tako generirajo hibridna priporočila. V pričujočem delu sodoben algoritem GNN, PinSage, v vlogi hibridnega priporočilnika apliciramo na javne podatke platforme Spotify. Algoritem naučimo na novem grafu vključenosti skladb v seznamih predvajanja in ga evalviramo v primerjavi z metodami na osnovi vsebine, matričnimi metodami CF in metodami CF na osnovi grafov, pri čemer obravnavamo tudi kriterije onkraj natančnosti (angl. beyond accuracy objectives). Naši eksperimenti kažejo sledeče. Implementirani algoritem je med najuspešnejšimi in izstopa kot najbolj robusten. Metode CF na osnovi grafov so bistveno boljše od matričnih metod CF, kar nakazuje, da je uporabniške informacije bolje predstaviti kot graf. V okviru evalvacijske naloge metode CF ne kažejo padca učinkovitosti v dolgem repu, kjer hibridni pristop posledično ne ponuja prednosti. Na podlagi rezultatov zaključimo, da je strojno učenje z grafi obetavna nova paradigma priporočanja glasbe, ki pa zahteva nekatere nadaljnje raziskave. Since streaming has emerged as the predominant means of music consumption, platforms such as Spotify rely on recommender systems to help users navigate within their increasingly large song catalogues. Collaborative filtering (CF) methods, which rely on past user-item interactions, have historically been most successful. They do, however, have various limitations, like performing poorly among lesser-known items with little or no interaction data. Conversely, content-based models attempt to circumvent the data-sparsity issue by generating recommendations based on item content (song audio) alone, but have seen limited success. A growing paradigm, which is seeing significant success in other fields and may help address the task at hand, is graph-based ML. Graph neural networks (GNN) in particular, promise to learn from both the complex relationships within a user-interaction graph, as well as from content to generate hybrid recommendations. In our work, we introduce graph-based ML to the field of music recommendation by applying a state-of-the-art GNN, PinSage, to public Spotify data. We train the implemented algorithm on a newly collected playlist-song membership graph and evaluate it against traditional CF, graph-based CF and content-based methods on a related-song prediction task, venturing beyond-accuracy in our evaluation. Our experiments show the following. Our approach is among top performers and stands out as the most well rounded. Graph-based CF methods significantly outperform matrix-based CF approaches, suggesting that user interaction data may be better represented as a graph. In the scope of our evaluation task, CF methods do not exhibit a performance drop in the long tail, where the hybrid approach does not offer an advantage. Based on our results we conclude that, although requiring further research, graph-based ML is a promising future direction for music recommendation.

V diplomski nalogi razširimo spletno platformo Trubadur z novim sistemom za upravljanje učenja. Izhajamo iz problema, da spletna platforma učitelju trenutno ne omogoča vpogleda v napredek njegovih učencev in možnosti zastavljanja interaktivnih domačih nalog. V ta namen v nalogi najprej raziščemo nekaj sistemov za upravljanje učenja in preverimo, kako jih pravilno zastaviti. Po razširitvi platforme lahko učitelj ustvari učilnico, sistemu poda pravila za domačo nalogo in ta nato sestavi naloge, obvesti učence in ovrednoti odgovore. Ker so pretekle raziskave o uporabniški izkušnji na platformi trajale relativno kratek čas, izvedemo trimesečno raziskavo o motivaciji uporabnikov. Ugotovimo, da so uporabniki tudi po daljšem časovnem obdobju še vedno aktivno prisotni na platformi. Opazimo rahel upad motivacije zaradi prelahkega nivoja zastavljenih nalog, kar je potrebno v naslednjih verzijah platforme še izboljšati. In this thesis, we extend the Trubadur web platform with a new learning management system. We start from the problem that the online platform does not currently allow teachers to see their students' progress and set interactive homeworks. To this end, in this assignment we first explore some learning management systems and check how to set them up correctly. After extending the platform, the teacher can create classrooms and interactive homeworks. The platform will then generate the homework quizes, inform the students and evaluate the answers. As past research on user experience on the platform has been relatively short, we are conducting a three-month study on user motivation. We find that even after a long period of time, users are still actively present on the platform. There is a slight decrease in motivation due to the level of exercises being too easy, which needs to be improved in future versions of the platform.

V diplomski nalogi rešujemo problem označevanja zvočnih posnetkov vibracijske komunikacije. Pri pregledu sorodnih aplikacij nismo zasledili, da obstaja aplikacija za tak tip specifičnega označevanja. Gre za označevanje datotek vibracijske komunikacije žuželk, kar pomeni, da mora biti spektrogram prilagojen za nižje frekvence potrebno pa je imeti tudi nekakšen sistem, da bodo označbe pregledne. Zato v diplomski nalogi implementiramo spletno orodje za prikaz in označevanje zvočnih posnetkov vibracijske komunikacije, ki bo poenostavila proces označevanja zbranih zvočnih posnetkov. Orodje je izdelano z namenom za čim lažjo in hitro uporabo in je tudi podlaga za strojno učenje in avtomatizacijo označevanja v prihodnjih različicah aplikacije. Podlaga za strojno učenje je, saj bomo s tem, ko uporabljamo naše orodje ali zgolj funkcionalnost za uvoz predhodno narejenih označb za potrebe digitalizacije ali pa datoteke tudi označimo s pomočjo naše aplikacije, dobili obširno zbirko označenih podatkov. Rezultata diplomske naloge sta orodje, namenjeno označevanju zvočnih posnetkov vibracijske komunikacije in uvoz ročno izdelane zbirke označenih zvočnih posnetkov v našo aplikacijo, ki nam bo omogočila avtomatizacijo označevanja. In this bachelor's thesis, we tackle the problem of annotating audio recordings of vibrational communication. Our overview of audio-related applications and tools showned a lack of programs specifically suited for this kind of annotation. Given that the goal is an annotation of vibrational communication of insects, the spectrogram must be adjusted for lower frequencies. A system to ensure clarity of the annotations must be put in place. In this thesis, we thus implement a web tool for visualisation visualization and annotation of vibrational communication audio files, designed to simplify this process. We developed this tool with speed and ease-of-use in mind. Additionally, it allows for the implementation of machine learning and automatic annotation in future versions. The web tool makes machine learning possible by accruing annotated files into an extensive library either by importing previously annotated ones or by creating new annotations during use. The results of this thesis are a web tool designed for annotation of vibrational communication audio files and a hand-curated library of annotated files to be imported and used for automatisation automation of the workflow.

V sodobni organski kemiji znanstveniki razpoznavajo organske spojine iz različnih naborov spektrov, kjer vsak posamezni spekter vsebuje podatke o teh spojinah, kot so vsebovani element, struktura, itd. Kemiki so veliko spojin že posneli in razpoznali, računalničarji pa izdelali programe, ki s pomočjo iskanja po podatkovnih bazah najdejo preiskovano spojino. Kljub temu kemiki vedno znova odkrivajo oz. srečujejo nove, do sedaj še neznane spojine, ki jih je potrebno ročno prepoznati oz. razvozlati, za kar lahko potrebujejo dlje časa, lahko tudi več ur ali v primeru zelo kompleksnih molekul, več dni. V okviru diplomske naloge je bil razvit algoritem, ki organske spojine analitično razpoznava, brez vnaprej znane podatkovne baze. Z uporabo IR, NMR in MS spektrov je bilo mogoče dobiti zadovoljive rezultate pri reševanju organskih molekul. Algoritem, ki je nastal v okviru diplomskega dela, je javno dostopen v obliki spletnega aplikacijskega vmesnika (ang. application programming interface - API), za katerega je bil izdelan tudi uporabniški vmesnik za enostavnejšo uporabo. In modern chemistry scientists elucidate organic compounds from different sets of spectrums in which every spectrum contains it's own information about the compound i.e.: known fragments, structure, etc. Since chemists have solved and documented a lot of these compounds computer scientists quickly created softwares which, with the help of an external databases, find the researched compound. Regardless chemists still discover new yet unrecorded compounds that they need to elucidate by hand which can take them multiple hours or, in case of highly complex molecules, even days. With this purpose an algorithm, described in this document, was developed which analitically solves organic compounds without the use of any external database. By using IR, NMR and MS spectrums the algorithm was able to get accurate results while elucidating organic compounds. The developed algorithm is available to the public in the form of a web application interface together with a user interface for simple usage.