Ontologie v rekomendačních systémech
Ontologies in Recommender Systems
Typ dokumentu
disertační prácedoctoral thesis
Autor
Stanislav Kuznetsov
Vedoucí práce
Kordík Pavel
Oponent práce
Peška Ladislav
Studijní obor
InformatikaStudijní program
InformatikaInstituce přidělující hodnost
katedra aplikované matematikyPráva
A university thesis is a work protected by the Copyright Act. Extracts, copies and transcripts of the thesis are allowed for personal use only and at one?s own expense. The use of thesis should be in compliance with the Copyright Act http://www.mkcr.cz/assets/autorske-pravo/01-3982006.pdf and the citation ethics http://knihovny.cvut.cz/vychova/vskp.htmlVysokoškolská závěrečná práce je dílo chráněné autorským zákonem. Je možné pořizovat z něj na své náklady a pro svoji osobní potřebu výpisy, opisy a rozmnoženiny. Jeho využití musí být v souladu s autorským zákonem http://www.mkcr.cz/assets/autorske-pravo/01-3982006.pdf a citační etikou http://knihovny.cvut.cz/vychova/vskp.html
Metadata
Zobrazit celý záznamAbstrakt
The cold-start problem is typical in recommender systems (RS). There are several ways to reduce this problem. There are naive approaches, or there are algorithms that try to reduce the problem by using attributes or generating additional information. We find a way to automatically or semi-automatically generate more in-depth knowledge of the domain and use that knowledge to minimise the cold-start problem. This knowledge is, in our case, represented by a semantic level that can add essential information about items and also be understandable to users. Our approach is to use ontology in the form of a graph for semantic purposes. We use the benefits of graph architecture to automatically generate an ontological profile for the items, and if needed, we can extend profiles with the most similar semantic nodes, and use these profiles for recommendation purposes. In this thesis, we describe the principle of automatically generating ontologies and improving them with explicit word embeddings. We present the general methodology for evaluating algorithms for cold-user and cold-item problems. Next, we show our ontology-based algorithms for reducing the cold-start problem (OBACS). The algorithm is a combination of ontology, and collaborative filtering (CF) approaches. It is a universal algorithm that could reduce the cold-start problem in an earlier phase where more of the items are cold, and also in the transition phase, where the number of cold-start items decreases. Also, these algorithms could be used for diversification and to help CF approaches with the “filter bubble” problem. We prove our approach with an offline experiment in the area of movie recommendation and discuss the results. The cold-start problem is typical in recommender systems (RS). There are several ways to reduce this problem. There are naive approaches, or there are algorithms that try to reduce the problem by using attributes or generating additional information. We find a way to automatically or semi-automatically generate more in-depth knowledge of the domain and use that knowledge to minimise the cold-start problem. This knowledge is, in our case, represented by a semantic level that can add essential information about items and also be understandable to users. Our approach is to use ontology in the form of a graph for semantic purposes. We use the benefits of graph architecture to automatically generate an ontological profile for the items, and if needed, we can extend profiles with the most similar semantic nodes, and use these profiles for recommendation purposes. In this thesis, we describe the principle of automatically generating ontologies and improving them with explicit word embeddings. We present the general methodology for evaluating algorithms for cold-user and cold-item problems. Next, we show our ontology-based algorithms for reducing the cold-start problem (OBACS). The algorithm is a combination of ontology, and collaborative filtering (CF) approaches. It is a universal algorithm that could reduce the cold-start problem in an earlier phase where more of the items are cold, and also in the transition phase, where the number of cold-start items decreases. Also, these algorithms could be used for diversification and to help CF approaches with the “filter bubble” problem. We prove our approach with an offline experiment in the area of movie recommendation and discuss the results.