The Effect of Random Seeds for Data Splitting on Recommendation Accuracy
Mohamed Amezian El Idrissi · 2024
To measur the performance of the recommender system algorithms, the dataset is split randomly into training and test sets using ran?dom seeds, which may cause a big difference in the results. The experiments done in this work are those outlined in the paper “ The Effect of Random Seeds for Data Splitting on Recommendation Accuracy” and we also extended the analysis with two additionaldatasets and algorithms used in the study. Our replication also established the fact that the random seed selection influences the performance of the recommendation algorithms and the resultsshowed a similar deviation in accuracy to the original work using this new algorithms. Therefore, our work indicates that it is essen?tial for the recommender systems researchers to pay attention to the random nature of their assessment processes to ensure that the results obtained can be reproduced and to provide reliable measure?ments of the performance. Here we sum up the general conclusions and open the codes to the public so that others can use the data and further the study in this area.