Context-Aware Recommender System (CARS)

Shaina Raza
Sep 7, 2018 · 1 min read

A typical RS tries to estimate the rating function R as R: User X Item →Rating (Bobadilla,Ortega,Hernando & Gutiérrez,2013). This recommendation function is called two-dimensional (2D) as it only considers User and Item entities in the process. In the case of CARS, we incorporate not only user and item entities but also consider context either as another dimension (e.g. matrix factorization method (Paatero & Tapper, 1994) ) or as additional dimension in the rating function as R: User x Item x Context → Rating (Adomavicius & Tuzhilin, 2008) (e.g. tensor methods (Kolda & Bader, 2009). As an example, if we have a movie recommender system, then we have Movies and Users in the rating function; with contextual information such as time, location, companion, community. In order to understand multi-dimensional context-aware data, we need to be familiar with the terms: context dimension, context condition and context situation (Baltrunas, Ludwig, & Ricci, 2011). For instance, if we have a user-item dataset with time, location and companion as context, then we can say that these three elements are context dimensions. Each of these context dimensions may encompass additional information as context conditions such as weekend/weekday for time, home/hospital/park for location; and spouse/sibling for companion. Each context condition in turn can be further decomposed into nuclear information as the context situation; continuing with the previous example, we may select weekend from time, home from location and spouse from companion as more specific information.

Shaina Raza

Written by

Data Scientist, PhD scholar