Efficient Models and Techniques for Recommendation Systems

This presentation will be dedicated to presenting the findings outlined in my doctoral dissertation, which concentrate on the exploration and improvement of Recommender Systems. The effectiveness of any recommender system is typically evaluated based on three key aspects:

• the accuracy of recommendations in terms of relevance,
• the system’s capability to address the cold-start problem, and
• (the time taken to generate recommendations.

In the presentation, a literature review pertaining to state-of-the-art recommendation systems will be discussed. Then, various methods will be proposed to enhance each of these aspects. Regarding the accuracy of recommendations, a novel recommender system that utilizes the contextual information to find more relevant recommendations will be presented. Next, we introduce Clustering-based FPRS - a novel recommender system, which provides several strategies to address the cold-start problem using frequent pattern mining. When it comes to the third aspect (minimizing latency in RS), we introduce an innovative recommender system, known as Factorization Machines and Association Rules (FMAR). This approach significantly reduces the volume of items processed by the recommendation model by integrating association rules into the recommendation generation process.