User profiling for diverse user contexts

Abstract

The amount of content available for consumption online is increasing tremendously. This make the job of recommender systems more important, and at the same time, more demanding. Context-aware recommender systems might be a solution to this problem.

This work set out to discover user contexts dynamically by collecting contextual information from user actions and perform cluster analysis on the data collected. User interests are collected from user actions as well, and are sorted into groups based on the contexts discovered. These sorted interests are considered the users’ user profile. The user profiles are in turn used to recommend news articles based on the interests of the users, where the users can select what context to receive recommendations from.

The results and evaluation of the system show that the approach used in this work is not very successful and adjustments are recommended to improve the results. The system designed and implemented in this work is only able to identify two very broad contexts based on user data.