RF-Rec Empfehlungen

Recommendation systems have been an important area of research since the publication of the first scientific articles on collaborative methods in the mid-1990s. Research has been focused for years on developing more complex algorithmic approaches to improve results in terms of precision metrics such as the absolute mean error rate. This paper analyzes a simple recommendation method called RF-Rec, which is based on rating frequencies, and presents additional variants. Evaluations on six well-known datasets show that RF-Rec predictors can achieve comparable results to popular methods like SVD and Slope One, especially when data is scarce, which poses a challenge for recommendation systems. Statistical tests demonstrate that the observed differences are not significant. Furthermore, it is shown that simple optimizations can lead to further improvements. This may contribute to the ongoing debate questioning the validity of precision metrics.