Recommender System Techniques

By: Nunung Nurul Qomariyah, Ph.D

Not long after the Recommender System concept was introduced, many researchers developed techniques to implement it in real online systems. Businesses were interested in implementing the concept to increase sales by recommending suitable products for their customers. Researchers in this field worked more to find the best method to learn user preferences and collect them as historical data. The more they knew about a user’s preferences, the more accurate predictions of recommended items they could produce. Machine learning or datamining techniques were then used to explore users’ historical data.

How techniques in RS are categorised differs from one source to another, and sometimes they use different names for the same concepts. The most common techniques are Collaborative Filtering (CF), Content-Based Filtering (CBF) and Hybrid. How the techniques work is described below:

• Collaborative Filtering (CF): This technique recommends items by looking at other users who have similar preferences.
• Content-Based Filtering (CBF): This technique recommends items by looking at how the user rates items.
• Hybrid: This technique recommends items by combining two or more techniques.

Other techniques mentioned in the literature include community-based, demographic, knowledge-based, context-aware, rule filtering, stereotypes, item-centric/co-occurrence based, graph-based and global relevance. Each e-commerce site will need different techniques or approaches, depending on their specific characteristics and goals. Each of these techniques has its own advantages and disadvantages.

Sources:

Qomariyah, N. N. (2018). Pairwise Preferences Learning for Recommender Systems (Doctoral dissertation, University of York).