The Effect of Third Party Implementations on Reproducibility

Reproducibility of recommender systems research has come under scrutiny\nduring recent years. Along with works focusing on repeating experiments with\ncertain algorithms, the research community has also started discussing various\naspects of evaluation and how these affect reproducibility. We add a novel\nangle to this discussion by examining how unofficial third-party\nimplementations could benefit or hinder reproducibility. Besides giving a\ngeneral overview, we thoroughly examine six third-party implementations of a\npopular recommender algorithm and compare them to the official version on five\npublic datasets. In the light of our alarming findings we aim to draw the\nattention of the research community to this neglected aspect of\nreproducibility.\n