User to user topic recommendation model for student social interaction

• Main Author: Ahmad Subhi, Zolkafly
• Format: Thesis
• Language: English; English; English
• Published: 2025
• Subjects: HM Sociology.
• Online Access: https://etd.uum.edu.my/11652/1/permission%20to%20deposit-s901852.pdf; https://etd.uum.edu.my/11652/2/s901852_01.pdf; https://etd.uum.edu.my/11652/3/s901852_02.pdf; https://etd.uum.edu.my/11652/

Social connectedness is important especially in academic setting because it can establish the sense of belonging, increase the emotional and academic support that result to higher motivation and better academic outcomes. Recommender systems can play a significant role in promoting social connectedness by fostering interactions, shared experiences, and a sense of belonging within communities or networks. Current social connectedness efforts heavily focus on digital interactions, online communities, and global reach that often powered by data-driven personalization. The current recommender systems also need to consider more homophily and homogeneity based on refined psychographic and personality traits. Hence, it is important to have a recommender system that suggested to solve these issues by recommending mutual topics for supporting social interactions and focusing on user-user relationship. This study focuses on developing a User to User Topic Recommendation Model for Student Social Interaction (TReSIT). This model is grounded on the theory of homophily, which describes individuals' tendency to interact with or be linked with others based on comparable characteristics or specific shared ideals. A predictive TReSIT model is developed based on psychological (the "big 5") and psychographics (values and lifestyles) factors. This predictive analytics model compared single and multi-classifier accuracy in predicting the importance of a topic for an individual. Then, a newly developed algorithm called the Social Recommendation (SAR) algorithm ranked and recommended the best topic for social interaction. The results show that the Parameterized Basic Ranking Evaluation Function (PBREF) has higher accuracy (69.90%) than Parameterized Multiple Classifier Ranking Evaluation Function (PMCREF) with 64.58%. An experiment also has been conducted to evaluate satisfaction towards the topic recommended. A total of 16 groups were formed based on the topic recommended by SAR and interacted with their team members. Later, they rated their interactions' satisfaction. The results show that the mean satisfaction level is 3.873. This model can be used in a working setting to trigger social interactions based on mutual interest. It can be implemented on a social robot that can trigger topics for conversations among individuals in a shared space like a resting area. The conversation later can foster social connectedness, reduce isolation, and possibly increase mental health.