Evaluation Methodologies of Recommendation System: An Experimental Approach

Information overload is a common issue faced by internet users due to the huge amount of freely available content. A single search could return millions of results and it is not feasible for a user to go through each of it to find the information relevant to his/her needs. It a tedious and time-consuming task. To address this, recommendation systems have been introduced. It is a tool that provides suggestions to users based on the user’s preference of a particular item or predicted using the ratings gathered via feedback from previous users with similar taste. The emergence of various types of recommendation system has raised the question of how these systems can be evaluated in a standardized and effective manner. It is important for researchers to quantitatively compare the performance of new recommender against the existing ones to establish that the proposed solution is indeed an improvement to the current one. There is a need for researchers as well as academics to be able to compare between two different systems and establish which the better one is based on a clearly defined numerical value. Based on this numerical value, researchers will be able to decide whether the new system requires further fine tuning. Repeated evaluation can be made after each fine tuning to determine the margin of improvement that has been made. This article will propose a methodology to perform such an evaluation of recommendation systems regardless of the domain in which the recommendation system has been used.