preview

Strengths And Weaknesses Of Recommender System

Decent Essays

Recommender Systems are based on well-structured and incremental algorithms that have different strengths and weaknesses \cite{ricci2011introduction}. In general, the main existing RSs are concerned with improving accuracy with the premise of being useful to users. In this context, there is an increasing number of custom techniques that analyze the profile of the target user in order to better satisfy them \cite{bobadilla2013recommender}. In this sense, we conducted a broad study of the several metrics used in the literature to evaluate RSs. By organizational purposes, we divided these metrics into three groups: {\it Effectiveness-based}, {\it Complementary Dimensions of Quality} and {\it Domain Profiling}.\looseness=-1 \subsection{Effectiveness-based metrics} The main evaluation metrics are related to the accuracy, precision, and recall concepts \cite{herlocker2004evaluating}. Precision and recall are concepts that aim to quantify information …show more content…

We calculate serendipity by the complement of the cosine similarity between the items in the user's history and the items in a recommendation list \cite{zhang2012auralist}. Lower values indicate that recommendations deviate from a user's traditional behavior, and therefore bring greater surprise.\looseness=-1 \begin{equation} \label{seren} ser = 1 - \sum_{u \in S} \frac{1}{|S||H_u|} \sum_{h \in H_u} \sum_{i \in R_{u,20}} \frac{CosSim(i,h)}{20} \end{equation} \item \textbf{Catalog Coverage:} Catalog Coverage represents the fraction of relevant items that are recommended at least once, taking into consideration all the users. Larger catalog coverage indicate that the recommender balances the popularity bias by covering a large part of the set of items \cite{puthiya2016coverage}.\looseness=-1 \begin{equation} \label{catalogo} CC = \frac{| \bigcup_{u \in U}^{} R^{+}|}{|U|}

Get Access