Please use this identifier to cite or link to this item:
https://ptsldigital.ukm.my/jspui/handle/123456789/513410Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor.advisor | Shahrul Azman Mohd Noah, Prof. Dr. | - |
| dc.contributor.author | Sumaia Mohamed Abdullah Al-Ghuribi (P89124) | - |
| dc.date.accessioned | 2023-10-16T04:36:22Z | - |
| dc.date.available | 2023-10-16T04:36:22Z | - |
| dc.date.issued | 2021-04-09 | - |
| dc.identifier.other | ukmvital:130134 | - |
| dc.identifier.uri | https://ptsldigital.ukm.my/jspui/handle/123456789/513410 | - |
| dc.description | Recommender system (RS) has been proven to be significantly crucial in many fields and is widely used by various domains. Most of the RSs approaches rely on a singlecriterion rating as a primary source for the recommendation process. However, a single-criterion rating like the overall rating is not enough to gain high accuracy of recommendations. The overall rating cannot express fine-grained analysis behind the user's behaviour. The multi-criteria RS (MCRS) has been developed to improve the accuracy of the RS performance to solve this problem. Additionally, a new source of information represented by the user-generated reviews is incorporated in the recommendation process because of the rich and numerous information included (i.e. review elements) related to the whole item or a specific feature. These review elements are used in this study to propose different MCRS models based on different combinations of these elements. To develop the MCRS models, three major steps are achieved: extracting the review elements, developing the users’ profile based on the extracted elements, and exploring different multi-criteria recommendation models based on different combinations of review elements represented in the users’ profiles profile. Three elements are extracted for the review elements extraction step: total review polarity score, aspect, and comparative words. The existing extraction approaches are applied on small-sized labelled datasets; thus, using these approaches on large-scale datasets may produce inaccurate results. Furthermore, current approaches have significant drawbacks, such as extracting many irrelevant aspects, neglecting infrequent aspects, and making some unrealistic assumptions in some circumstances. As a result, this study proposes an extraction approach applicable for real large-scale unlabeled datasets. The proposed approach is a combination of hybridizing a frequency-based approach and a syntactic-relation based approach. It was enhanced further with a semantic similarity-based approach to extract aspects relevant to the domain; even terms (related to the aspects) are not frequently mentioned in the reviews. According to the proposed approach, the extracted aspects are used to generate a total review polarity score after estimating the weight and the rating of each extracted aspect mentioned in the review. The proposed approach is evaluated using two real-world datasets: Amazon and Yelp. The results (in terms of Fmeasure and accuracy) show that the proposed approach outperformed all the baselines. The previously extracted elements with the helpfulness element are used to develop the users' profile. The profile includes several records, and each record comprises different cells that reflect the user's detailed information about the consumed items. Finally, different combinations of the review elements available in the users’ profile are proposed for developing different MCRS models for rating prediction. Before implementing the MCRS models, the effectiveness of the clustering process for users and aspects are evaluated by implementing several cases separately. Results show that clustering of aspects improves recommendation performance, despite that clustering of users does not. After clustering of aspects, several MCRS models are proposed in which the criteria of each model are a combination of review elements. An extensive series of experiments on the Amazon dataset are carried out to evaluate the performance of these models. The experimental results showed that the proposed MCRS models outperformed existing aspect-based and single-criterion rating RS models in the rating prediction process, demonstrating the benefit of the user reviews elements in improving the MCRS performance.,Ph.D | - |
| dc.language.iso | eng | - |
| dc.publisher | UKM, Bangi | - |
| dc.relation | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | - |
| dc.rights | UKM | - |
| dc.subject | Universiti Kebangsaan Malaysia -- Dissertations | - |
| dc.subject | Dissertations, Academic -- Malaysia | - |
| dc.subject | Recommender system (RS) | - |
| dc.subject | Multi-criteria RS (MCRS) | - |
| dc.subject | Decision support systems | - |
| dc.title | Multi-criteria collaborative filtering recommender systems based on a combination of user reviews elements | - |
| dc.type | Theses | - |
| dc.format.pages | 274 | - |
| dc.identifier.barcode | 005867(2021)(PL2) | - |
| Appears in Collections: | Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat | |
Files in This Item:
| File | Description | Size | Format | |
|---|---|---|---|---|
| ukmvital_130134+Source01+Source010.PDF Restricted Access | 3.78 MB | Adobe PDF |  View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.