Publication Date

Fall 2020

Degree Type


Degree Name

Master of Science (MS)


Computer Engineering


Mahima Agumbe Suresh


Distance Metric Learning, Matrix Factorization, Neural Networks, Recommender Systems, Social Networks, Social Recommendation

Subject Areas

Computer engineering; Computer science; Artificial intelligence


Recommender systems are powerful tools that filter and recommend content relevant to a user. One of the most popular techniques used in recommender systems is collaborative filtering. Collaborative filtering has been successfully incorporated in many applications. However, these recommendation systems require a minimum number of users, items, and ratings in order to provide effective recommendations. This results in the infamous cold start problem where the system is not able to produce effective recommendations for new users. In recent times, with escalation in the popularity and usage of social networks, people tend to share their experiences in the form of reviews and ratings on social media. The components of social media like influence of friends, users' interests, and friends' interests create many opportunities to develop solutions for sparsity and cold start problems in recommender systems. This research observes these patterns and analyzes the role of social trust in baseline social recommender algorithms SocialMF - a matrix factorization-based model, SocialFD - a model that uses distance metric learning, and GraphRec - an attention-based deep learning model. Through extensive experimentation, this research compares the performance and results of these algorithms on datasets that these algorithms were tested on and one new dataset using the evaluations metrics such as root mean squared error (RMSE) and mean absolute error (MAE). By modifying the social trust component of these datasets, this project focuses on investigating the impact of trust on performance of these models. Experimental results of this research suggest that there is no conclusive evidence on how trust propagation plays a major part in these models. Moreover, these models show slightly improved performance when supplied with modified trust data.