introduction.tex

\section{Introduction}
Today's use of \textit{recommender systems} finds an increased and yet unconscious access to our everyday life.
More and more areas of life are therefore subject to constant optimisation.
Companies such as \textit{Netflix}, \textit{Amazon} and \textit{YouTube} adapt their product proposals to the individual wishes of their customers.
To make this possible, the various \textit{collaborative-filtering} and \textit{content-based} \textit{recommender systems} are used.

Since \citet{JuKa90} first presented \textit{recommender systems} as a kind of intelligent bookcase, much effort has been put into the development and research of such systems.
The most diverse subject areas were not only illuminated by the industry.
 A whole new branch of research also opened up for science.
 
 In their work ``\textit{On the Diffculty of Evaluating Baselines A Study on Recommender Systems}`` \citet{Rendle19} show that current research on the \textit{MovieLens10M} dataset leads in a wrong direction.
 In addition to general problems, they particulary list wrong working methods and missunderstood \textit{baselines} by breaking them by a number of simple methods such as \textit{matrix-factorization}.
 
 They were able to beat the existing baselines by not taking them for granted.
 On the contrary, they questioned them and transferred well evaluated and understood properties of the baselines from the \textit{Netflix-Challenge} to them.

As a result, they were not only able to beat the \textit{baselines} reported for the \textit{MovieLens10M}, but also the newer methods from the last 5 years of research. Therefore, it can be assumed that the current and former results obtained on the \textit{MovieLens10M} dataset were not sufficient to be considered as a true baseline. Thus they show the community a critical error on which can be found not only in the evaluation of \textit{recommender systems} but also in other scientific areas.

As a first problem, the authors point out that scientific papers whose focus is on better understanding and improving existing \textit{baselines} do not receive recognition because they do not seem innovative enough. In contrast to industry, which tenders horrendous prizes for researching and improving such \textit{baselines}, there is a lack of such motivation in the scientific field. From the authors point of view, the scientific work on the \textit{MovieLens10M} dataset is misdirected, because one-off evaluations leading to one-hit-wonders, which are then used as a starting point for further work. Thus \citet{Rendle19} points out as a second point of criticism that the need for further basic research for the \textit{MovieLens10M} dataset is not yet exhausted.

This submission takes a critical look at the topic presented by \citet{Rendle19}. In addition, basic terms and the results obtained are presented in a way that is comprehensible to the non-experienced reader.
For this purpose, the submission is divided into three subject areas. First of all, the non-experienced reader is introduced to the topic of recommender systems in the section ``\textit{A Study on Recommender Systems}``. Subsequently, building on the first section, the work in the section ``\textit{On the Diffculty of Evaluating Baselines}`` is presented in detail. The results are then evaluated in a critical discourse.