recommendation system, cold start problem, collaborative filtering, content-based filtering, embeddings, nearest neighbor
Recommendations systems are software solutions for finding high-quality and relevant content for a given user type ranging from online shoppers, to music listeners, to video game players. Traditional recommendation systems use user review data to make recommendations, but we still want recommendations to perform well for new users with no review data. Currently, one of the problems that exists in recommendations is poor recommendation accuracy when only a small amount of data exists for a user, called the cold start problem. In this research we investigate solutions for the cold start problem in video game recommendations and we propose a solution that uses a hybrid neural network and keyword ranking approach. We evaluate this system with precision and recall metrics and compare the results to a traditional recommendation system. We present that this hybrid system offers performance gains when recommending to users who have low-medium previous reviews.
Faculty of Applied Science & Technology (FAST)
© Nicholas Crawford, 2019
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.
Original Publication Citation
Crawford, N. (2019). Improving video game recommendations using a hybrid, neural network and keyword ranking approach (Unpublished thesis). Sheridan College, Ontario, Canada.
Crawford, Nicholas, "Improving Video Game Recommendations Using a Hybrid, Neural Network and Keyword Ranking Approach" (2019). Faculty of Applied Science and Technology - Exceptional Student Work, Applied Computing Theses. 4.