Abstract
The internet and the digital libraries are major sources of information forresearchers, and there is an enormous growth of information on these sources. A large
number of research papers are available which leads to the information overload
problem and hence finding research papers that are related to users’ interests become
difficult and time consuming. The field of recommender systems aims to solve the
information overload problem by filtering information and providing users with
relevant results. Although the current recommender systems provide recommendation
services to users, different limitations and challenges have not been adequately
addressed in the research paper domain. The work presented in this thesis contributes
to the development of models and algorithms to the recommender systems in the
research paper domain. The main aim of this thesis is to develop a dynamic multiconcept
system that is able to recommend research papers of interest at appropriate
times. The first contribution of this thesis is modelling dynamic user profiles that are
able to adapt to the changes in multiple user interests and to be compatible with the
requirements of advanced ontologies. The second contribution is analysing users’
reading behaviour with research papers to develop novel short-term and long-term
models that are able to adapt dynamically according to a user’s changing behaviour
during his/her short and long term goals. These models can effectively learn different
users’ reading behaviours implicitly without the need for any intervention from the
user. The third contribution is predicting user’s future interests using a novel
collaborative filtering approach without the need for the user ratings. All our proposed
models are evaluated using offline evaluations with the BibSonomy dataset that
contains actual users’ records. Our results show that our models outperform the
baselines used for comparisons. Finally, we integrated our models to one unified
dynamic hybrid system in order to provide recommendations which most closely
represent the users’ research interests at particular times. The evaluation results
indicate that the dynamic hybrid system that models and integrates multiple user
interests and concepts can bring substantial benefits to a recommender system in the
research paper domain.
| Date of Award | 2018 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Gulden Uchyigit (Supervisor) |