Citation

Chew, Lit Jie (2022) Ontology-based hybrid recommender system. Masters thesis, Multimedia University.

Full text not available from this repository.

Abstract

In today’s information-overloaded era, the exponential growth of information we are required to absorb is huge, and we are hard to get the information that we needed. Hence, Recommender System (RS) has been introduced to tackle this problem. RS captures the user preferences and behaviour and then provides a relevant recommendation to the user. This helps users to reach the information that they might not be able to search for by themselves. With this, RS has been widely developed in many companies such as Netflix and Amazon to boost their sales and revenue. Current RS that are using only content-based or collaborative filtering are hard to adapt to the continuous changes of user preferences and have some issues such as cold-start and data sparsity. On the other hand, ontologies define rules to structure data, including interrelations between entities in the database. As such, it offers greater semantic relations within a particular domain. As compared to relational databases, the ontology approach is more flexible, scalable and faster. With the previous success of ontology in RS, we propose to design an RS, in which the semantic information about the domain is constructed as an ontology to represent both the user profile and the recommendable items. The drawback of the current RS is the result may be limited by sparse data and the information is not sufficient for the model to generate a good recommendation. Essentially, good data will improve the accuracy and personalization of an RS model. Hence, this motivates us to enrich the data prior to passing it to an RS model to achieve higher accuracy. In this research, we proposed an ontology-based hybrid RS. We aimed to improve the accuracy of the matrix factorization model. We enhanced the matrix factorization model accuracy by utilizing ontology to enrich the information of the user-item matrix by item-based (IB) and user-based (UB) collaborative filtering techniques. In our approach, the IB and UB have been modified to adapt to the ontology approach. Instead of using just the rating pattern as the similarity calculation in IB and UB, we proposed to use it together with the semantic similarity of item features and user features. The enriched data will then be forwarded to the matrix factorization model to produce the recommendation. The proposed method is evaluated with real-world data to verify the accuracy of our proposed method compared to the existing method. The evaluation performed on real-world datasets demonstrated that our proposed approach outperformed the baseline model and existing models in every dataset. With the ontology constructed with hierarchy structure, our proposed approach Ontology-based Dataset Enriching Recommendation Method (Onto-DERM+) has archived Root Mean Square Error (RMSE) value of 0.9357 in MovieLens 100K dataset, 0.9312 in MovieLens 1M dataset, 1.6321 in Book-Crossing dataset and 0.9653 in Yahoo! Movie dataset.

Item Type:	Thesis (Masters)
Additional Information:	Call No.: ZA3084 .C44 2022
Uncontrolled Keywords:	Recommender systems (Information filtering)
Subjects:	Z Bibliography. Library Science. Information Resources > ZA3038-5190 Information resources (General)
Divisions:	Faculty of Computing and Informatics (FCI)
Depositing User:	Ms Nurul Iqtiani Ahmad
Date Deposited:	11 Jan 2024 01:43
Last Modified:	11 Jan 2024 01:43
URII:	http://shdl.mmu.edu.my/id/eprint/12040

Downloads

Edit (login required)